TECHNOLOGY DIGEST - Draper Laboratory
TECHNOLOGY DIGEST - Draper Laboratory
TECHNOLOGY DIGEST - Draper Laboratory
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
THE DRAPER<br />
<strong>TECHNOLOGY</strong> <strong>DIGEST</strong><br />
2006 Volume 10 CSDL-R-3005 #*
The <strong>Draper</strong> Technology Digest (CSDL-R-3005) is published annually by the Education Office of<br />
The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc. Editorial questions and requests for reprints can be<br />
directed to:<br />
Media Services<br />
Phone: (617) 258-1491<br />
Fax: (617) 258-4535<br />
E-mail: techdigest@draper.com<br />
Mailing address changes can be forwarded to:<br />
communications@draper.com<br />
Editor-in-chief<br />
Dr. George Schmidt, Director, Education Office<br />
Creative Director<br />
Charya Peou<br />
Designer<br />
Pamela Toomey<br />
Editor<br />
Beverly Tuzzalino<br />
Photography Coordinator<br />
Drew Crete<br />
Photographers<br />
Jay Couturier<br />
Michael Brooks<br />
Copyright © 2006 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc. All rights reserved.
Introduction by Vice President, Engineering, Dr. Eli Gai<br />
Papers<br />
The Silicon Oscillating Accelerometer: A High-Performance MEMS<br />
Accelerometer for Precision Navigation and Strategic Guidance Applications<br />
RALPH HOPKINS, JOSEPH MIOLA, ROY SETTERLUND, BRUCE DOW, WILLIAM SAWYER<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
DAVID CARTER, SEAN GEORGE, PHILIP HATTIS, LEENA SINGH, STEVEN TAVAN<br />
An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock<br />
MARK MESCHER, ROBERT LUTWAK, MATHEW VARGHESE<br />
Detection of Biological and Chemical Agents Using Differential Mobility<br />
Spectrometry (DMS) Technology<br />
MELISSA KREBS, ANGELA ZAPATA, ERKINJON NAZAROV, RAANAN MILLER,<br />
ISAAC COSTA, ABRAHAM SONENSHEIN, CRISTINA DAVIS<br />
Autonomous Mission Management for Spacecraft Rendezvous Using an<br />
Agent Hierarchy<br />
MARK JACKSON, CHRISTOPHER D'SOUZA, HOBSON LANE<br />
Fluid Effects in Vibrating Micromachined Structures<br />
PETER KWOK, MARC WEINBERG, KENNETH BREUER<br />
2005 Published Papers<br />
Patents<br />
Patents Introduction<br />
Optically Rebalanced Accelerometer: Patent No. 6,867,411 B2<br />
Issued: March 15, 2005<br />
WILLIAM KELLEHER, STEPHEN SMITH, RICHARD STONER<br />
2005 Patents Issued<br />
The 2006 Charles Stark <strong>Draper</strong> Prize<br />
The <strong>Draper</strong> Distinguished Performance Awards<br />
The Howard Musoff Student Mentoring Award<br />
2005 Graduate Research Theses<br />
2005 Technology Exposition<br />
2<br />
4<br />
14<br />
26<br />
32<br />
42<br />
56<br />
70<br />
77<br />
78<br />
81<br />
82<br />
84<br />
85<br />
86<br />
88<br />
1
This year is the 10 th anniversary of <strong>Draper</strong>’s Technology<br />
Digest. For the past 10 years, the Digest has included<br />
over 60 of the best papers <strong>Draper</strong> staff members have published,<br />
demonstrating the breadth of <strong>Draper</strong> <strong>Laboratory</strong><br />
technology. It has also included lists of all patents issued<br />
to <strong>Draper</strong> staff members. I would like to take this<br />
opportunity to thank the Media Services Group, who are<br />
responsible for the publication of the Digest, for continuously<br />
improving the visual look of every issue.<br />
This year’s issue includes six papers that were published<br />
and patents that were issued during the calendar year<br />
2005. Committees established by the Vice President of<br />
Engineering evaluated all papers and patents during this<br />
period and selected those to be included in this issue, as<br />
well as the winner of the annual Vice President’s Awards.<br />
The papers in this issue can be divided into two groups<br />
covering two growth areas of <strong>Draper</strong> technologies. Four<br />
papers describe applications of Microelectromechanical<br />
Systems (MEMS) for inertial instruments, the atomic<br />
clock, and biomedical instruments. The other two papers<br />
cover autonomous systems – one to guide a parachute, the<br />
other for spacecraft rendezvous.<br />
The first paper, “The Silicon Oscillating Accelerometer: A<br />
High-Performance MEMS Accelerometer for Precision<br />
Navigation and Strategic Guidance Applications” by Ralph<br />
2<br />
10th An<br />
D R A P E R T E C H<br />
Introduction by Vice President, Engineering, Dr. Eli Gai<br />
Hopkins, Joseph Miola, Roy Setterlund, William Sawyer,<br />
and Bruce Dow was selected for the Vice President’s Award<br />
for Best Paper published in 2005. This paper describes the<br />
theory of operations, performance goals, and the fabrication<br />
process for two silicon oscillating accelerometer<br />
(SOA) designs, one for strategic missile guidance and one<br />
for submarine navigation. Test data show that both<br />
versions of the SOA met the performance required for<br />
their respective applications.<br />
The second paper, “Autonomous Guidance, Navigation, and<br />
Control of Large Parafoils” by David Carter, Sean George,<br />
Philip Hattis, Leena Singh, and Steven Tavan deals with<br />
the design and flight testing of a guidance, navigation, and<br />
control (GN&C) software package that enables precision<br />
payload airdrop delivery using large parafoils. The<br />
modular software design is structured to accommodate<br />
payloads ranging from 2000 to 30,000 lb. The requirements<br />
for low components cost resulted in a limited<br />
onboard processor and a laptop personal computer. In<br />
spite of these limitations, flight test data results showed a<br />
miss distance of 200 m.<br />
The third paper, “An Ultra-Low-Power Physics Package for<br />
a Chip-Scale Atomic Clock” by Mark Mescher, Robert<br />
Lutwak, and Matthew Varghese reports on the design of a<br />
low-power, small-package, thermally-isolated physics<br />
package for an atomic clock. This physics package will
niversary<br />
N O L O G Y<br />
D I G E S T<br />
enable communication and navigation systems that<br />
require an atomic frequency standard with the above tight<br />
requirements. The design demonstrated a package requiring<br />
only single milliwatts of power, which represented a<br />
reduction of two orders of magnitude compared with the<br />
lowest of existing commercial technology.<br />
The fourth paper, “Detection of Biological and Chemical<br />
Agents Using Differential Mobility Spectrometer<br />
Technology” by Melissa Krebs et al. describes a new<br />
MEMS-based sensor, the Differential Mobility Spectrometer<br />
(DMS) that was developed to detect chemical or<br />
biological agents down to the parts per trillion within<br />
seconds. The sensor is small, robust, can detect multiple<br />
agents simultaneously, and is nondestructive during its<br />
operation. Several experiments were conducted to demonstrate<br />
the ability of the sensor to detect Bacillus subtilis<br />
spores.<br />
The fifth paper, “Autonomous Mission Management for<br />
Spacecraft Rendezvous Using an Agent Hierarchy” by<br />
Mark Jackson, Christopher D’Souza, and Hobson Lane is<br />
the second paper in this issue that deals with vehicle<br />
autonomy. In this case, it is a chaser satellite, and the<br />
mission management task is to rendezvous with another<br />
satellite. The software manager is based on a hierarchy of<br />
“activity agents,” each of which monitor, diagnose, plan,<br />
and execute the activity. The algorithms in this agent use<br />
low-thrust guidance to arrive in the neighborhood of<br />
the target, followed by high-thrust burns to arrive at a<br />
specified point at a specified time.<br />
The last paper by Peter Kwok, Marc Weinberg, and<br />
Kenneth Breuer entitled “Fluid Effects in Vibrating<br />
Micromachined Structures,” has resulted in a closed-form<br />
solution for fluid damping and lift that can be used to<br />
design accelerometers, tuning-fork gyroscopes (TFG),<br />
and other micromechanical devices. The closed-form and<br />
the numerical solutions are compared against experimental<br />
data over a range of pressures using TFG samples, and<br />
the results show agreement among the three.<br />
Three patents were nominated for the Vice President’s<br />
Award for Best Patent issued during calendar year 2005.<br />
The winning patent, “Optically Rebalanced Accelerometer,”<br />
was authored by Bill Kelleher, Steve Smith, and<br />
Richard Stoner. Acceleration is observed by measuring<br />
the position changes of a proof mass. The sensitivity of<br />
the accelerometer is predicted to be 1 µg with less than 1ppm<br />
scale factor and maximum acceleration of 100 g,<br />
thus making it a candidate for strategic guidance<br />
applications.<br />
3
The Silicon Oscillating Accelerometer: A High-Performance<br />
MEMS Accelerometer for Precision Navigation and Strategic<br />
Guidance Applications<br />
Ralph Hopkins, Joseph Miola, Roy Setterlund, Bruce Dow, William Sawyer<br />
Copyright © 2005 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc.<br />
Presented at the Institute of Navigation 61 st Annual Meeting, Cambridge, MA, June 27-29, 2005<br />
INTRODUCTION<br />
SOA Applications and Performance Goals<br />
The ICBM/submarine-launched ballistic missile (SLBM) has<br />
the most demanding requirements of any inertial guidance<br />
application. The high degree of accuracy required of the<br />
weapon system, combined with the high acceleration levels<br />
and large velocity at reentry body deployment place an especially<br />
stringent performance requirement on the guidance<br />
system accelerometers. The SSBN SINS system requires similarly<br />
precise inertial performance from the navigation<br />
system accelerometers, although compared with the missile<br />
application, the SINS accelerometers enjoy a more benign<br />
4<br />
The intercontinental ballistic missile (ICBM) and submarine-launched strategic missiles developed<br />
over the past 50 years have employed successive generations of increasingly accurate inertial guidance<br />
systems. The comparatively short time of guided flight and high acceleration levels characteristic of the<br />
ballistic missile application place a premium on accelerometer performance to achieve desired weapon<br />
system accuracy. Currently, the U.S. strategic missile arsenal relies on variants of the pendulous<br />
integrating gyro accelerometer (PIGA) to meet the high-performance, radiation-hard requirements of<br />
the weapon system.<br />
Likewise, precision navigation systems such as the currently deployed SSBN ship inertial navigation<br />
systems (SINS) employ highly specialized and complex electromechanical instruments that, like the<br />
PIGA, present a system life-cycle cost and maintenance challenge.<br />
The PIGA and the electromagnetic accelerometer (EMA) demonstrate unsurpassed performance, however,<br />
their life-cycle cost has motivated a search for a high-performance, solid-state, strategic<br />
accelerometer.<br />
<strong>Draper</strong> <strong>Laboratory</strong> is currently developing the silicon oscillating accelerometer (SOA), a microelectromechanical<br />
system (MEMS)-based sensor that has demonstrated in laboratory testing the part-per-million<br />
(ppm)/µg scale-factor and bias performance stability required of strategic and precision navigation<br />
applications.<br />
The ICBM and SSBN applications have significantly different environmental, acceleration dynamic<br />
range, and resolution requirements that are best satisfied by optimizing the SOA geometry for each<br />
application. The design flexibility and wafer-scale fabrication methods of the silicon MEMS process<br />
enable manufacturing both instrument designs with essentially zero incremental cost associated with<br />
the additional instrument assembly line. That is, the SOAs developed for the ICBM guidance and SSBN<br />
navigation applications share a common sensor package, electronics architecture, main housing, and<br />
instrument assembly process. This paper presents an overview of <strong>Draper</strong>’s SOA and compares and<br />
contrasts performance data taken to date on both versions of the SOA.<br />
operational environment and a smaller dynamic range<br />
requirement. [5]<br />
Although there are many system-derived performance parameters<br />
specified for inertial-grade accelerometers (see Table 1), in<br />
broad terms, accelerometer performance can be characterized<br />
with two parameters: bias and scale-factor (SF) stability.<br />
Accelerometer bias is the DC offset indicated from the instrument<br />
output under zero applied acceleration. Scale factor is the<br />
instrument gain or sensitivity that relates the applied acceleration<br />
to the instrument output signal (e.g., V/g, Hz/g, etc.).
Bias<br />
Table 1. Typical SOA Performance Goals.<br />
Parameter Units Missile High-<br />
Guidance Performance<br />
Navigation<br />
Long-Term Stability* µg 1 - 100 1<br />
Short-Term Stability**<br />
Scale Factor<br />
µg 1 - 5 0.5<br />
Long-Term Stability* ppm 1 - 100 10<br />
Short-Term Stability** ppm 1 - 5 5<br />
Asymmetry ppm TBD TBD<br />
g2 (Compensated) µg/g2 0.1 - 0.2 1<br />
g3 (Compensated) µg/g3 TBD TBD<br />
VRW ft/s√h 0.014 - 0.030 0.0014<br />
White Noise<br />
Misalignment<br />
µg/√(Hz) 10 - 22 1<br />
Long-Term Stability* arcsec 0.1 - 5 TBD<br />
Short-Term Stability** arcsec 0.1 - 2 0.4<br />
Vibration Rect. µg/(grms) 2
Figure 1. SOA schematic.<br />
The SOA input axis lies in plane as indicated in Figure 1;<br />
under acceleration, the proof mass axially loads the two<br />
resonator pairs. The vibration frequency of each resonator<br />
changes under the applied load. This frequency change is<br />
measured and serves as the indicated acceleration output<br />
of the SOA. Note that the resonators are arranged so they<br />
are loaded differentially by the proof mass. That is, one<br />
resonator is placed in tension, the other in compression.<br />
This differential design doubles the sensitivity or scale<br />
factor of the accelerometer and furnishes a cancellation of<br />
error sources common to both resonators.<br />
The resonators are excited by an electrostatic comb<br />
drive, [1],[2] similar to that used in <strong>Draper</strong>’s micromechanical<br />
tuning-fork gyro (TFG). The comb drive has both inner<br />
and outer motor stator combs that are fixed to the glass<br />
substrate. The outer motor combs apply the drive force, the<br />
inner motor combs sense the drive amplitude and frequency.<br />
A detail of the comb geometry is shown in Figure 2.<br />
Figure 2. SOA oscillator detail.<br />
The goal of the SOA design is to achieve a high scale-factor,<br />
high-Q resonator to achieve high performance. Large scale<br />
factor is desirable because it decreases the degree of frequency<br />
stability required to resolve a given acceleration<br />
level. For example, 0.1-mHz frequency stability is required<br />
of a 100-Hz/g SF unit to resolve 1 µg. A 10-Hz/g unit has a<br />
10 times more restrictive frequency stability requirement<br />
(10 µHz) to resolve the same 1-µg input.<br />
6<br />
Anchored<br />
Component<br />
Suspended<br />
Component<br />
Electrostatic<br />
Component<br />
The Silicon Oscillating Accelerometer<br />
It can be shown [3] that the lateral stiffness of an axially<br />
loaded beam with fixed ends is:<br />
where:<br />
K = stiffness<br />
E = Young’s modulus<br />
I = moment of inertia<br />
L = beam length<br />
P = axial load<br />
If a lumped mass is supported between two beams, the<br />
natural frequency of the mass-beam system as a function<br />
of axial load is given by:<br />
where:<br />
f = resonant frequency<br />
m = mass of lumped oscillator<br />
Rearranging Eq. (2) gives:<br />
where:<br />
(1)<br />
(2)<br />
(3)<br />
f = resonant frequency<br />
f0 = nominal unloaded (bias) resonant frequency<br />
=<br />
m= resonator mass<br />
L = beam length<br />
E = Young’s modulus<br />
I = beam inertia<br />
P = applied axial load<br />
Note that the frequency versus applied acceleration load<br />
relationship in the SOA is nonlinear, as indicated by Eq.<br />
(3) and shown in Figure 3.<br />
Frequency (Hz)<br />
Buckling Load<br />
Bias Frequency fo<br />
Linear SF (Hz/g)<br />
f = fo<br />
Input acceleration (g)<br />
1 + M g<br />
π2EI L2 Figure 3. SOA frequency vs. acceleration curve.
A series expansion of Eq. (3) can be used to determine the<br />
linearized SOA scale factor (the slope about zero acceleration<br />
in Figure 3) and higher-order g-coefficients:<br />
where S = L2/π2EI. Equation (4) can be rewritten as:<br />
where:<br />
Kn =K1bn(K1/f0) n-1 (Hz/gn) bn =bn-1 (3-2n)/n<br />
K1 = S/2<br />
b1 =1<br />
g = acceleration<br />
Note that the values of the g2 and g3 coefficients (K2, K3)<br />
are controlled by the linear SF (K1) and bias frequency (f0).<br />
The linearized SF is dependent on resonator dimensions,<br />
Young’s modulus, and the mass of the resonator (m) and<br />
proof mass (M):<br />
(6)<br />
The SF stability of the SOA will be largely controlled by<br />
the Young’s modulus sensitivity to temperature (∆E/E/∆T =<br />
-50 ppm/°C), as it is an order of magnitude larger than<br />
silicon’s thermal coefficient of expansion (TCE) (2.5<br />
ppm/°C), the parameter that would control the resonator<br />
dimensional stability. Given the square root relationship,<br />
the linear SF temperature coefficient is approximately 25<br />
ppm/°C, indicating that 0.01°C temperature control will<br />
maintain better than 1-ppm SF performance.<br />
From Eqs. (4) and (5), the values of the g2 and g3 coefficients<br />
(K2, K3) can be expressed by the linear SF (K1) and<br />
bias frequency (f0):<br />
For a 100-Hz/g (per side) SF, 20-kHz nominal bias frequency<br />
unit, Eqs. (7) and (8) project g2 and g3 coefficients<br />
of 0.25 Hz/g2 and 0.0125 Hz/g3. Normalizing these coefficients<br />
by dividing by the linear SF gives 2500 µg/g2 and<br />
12.5 µg/g3, respectively.<br />
Compensating these coefficients to the sub-µg level is feasible<br />
because their stability will be of the order of the linear<br />
SF and bias. Differentiation of Eqs. (7) and (8) gives:<br />
The Silicon Oscillating Accelerometer<br />
(4)<br />
(5)<br />
(7)<br />
(8)<br />
(9)<br />
(10)<br />
Note that 1-µg performance implies a resonator frequency<br />
stability (∆f/f) on the order of 5 parts per billion (ppb)<br />
(given a100-Hz/g per side SF and 20-kHz nominal frequency<br />
unit). Combined with 1-ppm SF, this implies that<br />
the above SF nonlinearity coefficients should be stable to<br />
approximately 2 to 3 ppm. This high degree of stability<br />
should enable compensating the raw nonlinear coefficients<br />
to sub-µg levels (although calibrating the higher-order gcoefficients<br />
will require precision centrifuge testing).<br />
Additionally, the net SOA g 2 coefficient and other evenorder<br />
terms will be reduced as the g 2 contribution from<br />
each resonator will be common mode differenced in the<br />
net SOA output.<br />
SOA Electronics<br />
Figure 4 shows the SOA electronics block diagram. As<br />
mentioned above, the SOA employs an electrostatic comb<br />
drive to excite the resonators, and to pick off the resonator<br />
displacement. The electrostatic drive force is given by:<br />
where:<br />
F = drive force<br />
dC/dx = comb position sensitivity<br />
V = applied voltage<br />
Proof<br />
mass<br />
Gain<br />
X1<br />
-X1<br />
Gain<br />
X1<br />
-X1<br />
90<br />
deg<br />
Drive<br />
gen<br />
90<br />
deg<br />
Drive<br />
gen<br />
Freq<br />
det<br />
Ampl<br />
det<br />
C(s)<br />
Freq<br />
det<br />
Ampl<br />
det<br />
C(s)<br />
(11)<br />
Figure 4. SOA electronics block diagram.<br />
The drive amplitude stability furnished by the electronics is<br />
critical to maintaining nominal resonator frequency stability.<br />
The resonator beams stiffen with lateral deflection, causing<br />
a dependence of the resonant frequency with drive amplitude.<br />
This nonlinear stiffening effect introduces a bias<br />
uncertainty from the resonator drive amplitude instability.<br />
It can be shown [4] that if the resonator is modeled as a<br />
linear plus cubic stiffness element, the resonator frequency<br />
dependence on amplitude is given by:<br />
Fa<br />
– +<br />
Fb<br />
– +<br />
REF<br />
REF<br />
7
(12)<br />
where:<br />
f = frequency at amplitude x<br />
f0 = nominal resonant frequency<br />
K = linear stiffness of resonator<br />
K3 = cubic stiffness coefficient<br />
x = drive amplitude<br />
The stability requirement on the drive amplitude can be<br />
determined by differentiating Eq. (12) to get:<br />
(13)<br />
From Eqs. (12) and (13), it can be seen that a small drive<br />
amplitude will minimize the resonator frequency variance<br />
from drive amplitude instability and noise. An alternate<br />
means of maximizing frequency stability is to minimize<br />
the amount of nonlinear stiffening, i.e., design a resonator<br />
with a low K3 coefficient.<br />
The resolution or noise floor of the SOA can be estimated<br />
by calculating the amplitude and phase noise associated<br />
with the sense comb frequency readout. From Ref. [1], the<br />
capacitance across a set of engaged comb drive fingers is<br />
given by:<br />
(14)<br />
where:<br />
C = capacitance<br />
εo = permittivity of air<br />
N = Number of teeth per side<br />
α = fringing factor<br />
t = comb finger thickness<br />
g = air gap between fingers<br />
L = engaged length of fingers<br />
The capacitance sensitivity to position (i.e., engaged<br />
length) is given by:<br />
(15)<br />
where dC/dx = sensitivity to position.<br />
Equation (12) gives the relationship between resonator<br />
amplitude and frequency, which gives the resonator frequency<br />
power spectral density (PSD) as:<br />
where:<br />
8<br />
φf = frequency PSD in Hz/√Hz<br />
x = nominal drive amplitude<br />
φA = amplitude noise PSD<br />
f0 = nominal resonant frequency<br />
K3/K = stiffness coefficient ratio<br />
The Silicon Oscillating Accelerometer<br />
(16)<br />
The contribution of phase noise in the drive frequency<br />
electronics can also be estimated. The PSD of the oscillator<br />
phase noise is approximately equal to the PSD of the<br />
amplitude noise divided by the peak amplitude.<br />
At resonance, the phase noise is related to frequency noise<br />
by:<br />
where:<br />
φf = frequency noise PSD<br />
φp = phase noise PSD<br />
(17)<br />
ωn = nominal resonant frequency<br />
Q = Q of resonator<br />
The high Q’s achieved in the SOA oscillators (~100,000)<br />
significantly reduce the frequency noise in the output from<br />
phase jitter. The net frequency noise in the SOA readout is<br />
dominated by oscillator amplitude noise. Consequently,<br />
frequency readout resolution is improved with increasing<br />
bias voltage and decreasing drive amplitude.<br />
SOA MEMS Sensor Design, Fabrication, and Screening<br />
The ICBM application and the SSBN SINS system have significantly<br />
different environmental, acceleration dynamic<br />
range, and resolution requirements, as mentioned above<br />
(Table 1). The SOA instrument can be easily adapted to<br />
either application by adjusting the SOA MEMS sensor element<br />
geometry for the specific operating acceleration<br />
range required. Figure 5 is similar to Figure 3 and shows<br />
the load vs. acceleration curve for two different SOA<br />
designs: a higher g-capable design suitable for<br />
ICBM/SLBM environments and a lower g design tailored<br />
for the more benign SINS environment. Note that the<br />
higher sensitivity (higher SF) SINS design has a correspondingly<br />
lower buckling load. This is a consequence of<br />
beam mechanics and is a fundamental design trade-off in<br />
selecting SOA acceleration sensitivity.<br />
f = fo<br />
1 + M g<br />
π2EI L2 Linear SF (Hz/g)<br />
Buckling Load<br />
Frequency (Hz)<br />
Bias Frequency fo<br />
Low-g<br />
High-g<br />
Input acceleration (g)<br />
Figure 5. SOA dynamic range.<br />
The geometries of the two different SOAs are optimized for<br />
each application by adjusting oscillator and proof mass<br />
geometry. The advantages of the silicon MEMS design<br />
approach is apparent as two different applications can be<br />
serviced with only the incremental expense of the MEMS
photo-maskset needed to fabricate the application-specific<br />
SOA sensor element. The design flexibility and wafer-scale<br />
fabrication methods of the silicon MEMS process enable<br />
manufacturing both instrument designs with essentially<br />
zero incremental cost associated with the additional sensor<br />
design. That is, both versions of the SOA share a common<br />
sensor package, electronics architecture, main housing,<br />
and instrument assembly process. The only difference<br />
between the two instrument assembly processes is the use<br />
of a different MEMS photo-maskset in the fabrication of<br />
the SOA sensor element.<br />
<strong>Draper</strong> uses the silicon-on-glass dissolved-wafer process to<br />
fabricate the SOA, [6] a mature MEMS fabrication process<br />
that <strong>Draper</strong> employs elsewhere in MEMS-based inertial<br />
technology. [2]<br />
The main process steps of the dissolved-wafer process<br />
are illustrated in Figure 6. The starting wafers used in<br />
processing are a boron-doped epitaxial silicon layer<br />
grown on an undoped silicon handle layer. The thickness<br />
of the epitaxial layer determines the final<br />
thickness of the silicon device layer. The first process<br />
step is to etch mesas in the epitaxial silicon layer; this<br />
step defines the eventual operating air gap between the<br />
suspended silicon sensor element structures and the<br />
glass substrate.<br />
Silicon<br />
p ++ Epitaxial Growth<br />
Form Mesas<br />
Deep Trench Etch<br />
Glass Wafer<br />
Metalization<br />
EDP Etch Anodic Bond<br />
Figure 6. SOA fabrication process.<br />
The device structural layer is then photolithographically<br />
patterned on the silicon, and the wafers are etched using<br />
high-aspect-ratio micromachining in a reactive ion etch<br />
(RIE) machine. This step forms the 2-D geometry of the<br />
SOA silicon sensor element. Recent advances in RIE etching<br />
technology have enabled a very high degree of feature<br />
size control (e.g., beam width uniformity, side wall perpendicularity,<br />
etc.) in MEMS processing, a critical factor<br />
affecting as-built SOA sensor performance.<br />
The Silicon Oscillating Accelerometer<br />
The patterned wafers are then anodically bonded to glass<br />
substrates that have been metalized with the SOA electrode<br />
pattern. Finally, the silicon wafer is dissolved in an<br />
anisotropic wet etchant such as ethylenediamine pyrocatechol<br />
(EDP), which removes the silicon handle layer. The<br />
boron doping in the epitaxial layer stops the chemical<br />
etching action of the EDP, leaving the finished device layer<br />
array on the glass wafer substrate. Individual SOA sensors<br />
are then diced and screened prior to packaging.<br />
The first screening step performed on the SOA MEMS sensor<br />
die is a particle and defect inspection, followed by<br />
electrical probing for proper resistance and continuity.<br />
Units passing this step are “wiggle tested” by energizing<br />
the oscillator elements to verify oscillator freedom at the<br />
expected drive frequency.<br />
After probe testing, SOAs with satisfactory performance<br />
are packaged and vacuum sealed in an aluminum oxide<br />
leadless ceramic chip carrier (LCCC). The residual pressure<br />
achieved in the LCCC after vacuum seal is less than 1<br />
mTorr, which ensures high oscillator Q factors. A plot of Q<br />
vs. pressure for a typical SOA is shown in Figure 7, indicating<br />
that oscillator drive Q’s on the order of 100,000 are<br />
achieved with this process. The packaged SOA is mated to<br />
a front-end preamplifier electronics module and subsequently<br />
mounted in an instrument main housing<br />
assembly. The final SOA instrument assembly is shown in<br />
Figure 8.<br />
100000<br />
10000<br />
Log Q 1000000<br />
1000<br />
0.01 0.1 1 10 100 1000<br />
Log Pressure (mTorr)<br />
Figure 7. Q vs. pressure relationship.<br />
Figure 8. SOA instrument.<br />
Missile Guidance and SINS SOA Performance Test Data<br />
Both versions of the SOA demonstrate the performance<br />
required for their respective applications. As mentioned<br />
9
above, the SINS SOA is tailored with a higher sensitivity to<br />
exploit the reduced dynamic range environment of the SSBN<br />
application. The SINS SOA sensor design also incorporates<br />
design features to maximize longer time scale drift stability.<br />
The contrast in the performance of both versions of the<br />
SOA is best captured by the acceleration resolution capability<br />
demonstrated via the noise PSD plots shown in<br />
Figure 9. Note that both units have a white noise floor on<br />
the order of 10-11 g2/Hz, however, the SINS design maintains<br />
the resolution capability over longer time scales (i.e.,<br />
lower frequency bands). The missile guidance SOA shows<br />
a noise PSD that starts to break upward with a 1/f signature<br />
in the 10-1 Hz to 10-2 Hz decade. The SINS SOA<br />
shows the white noise floor extending another decade or<br />
two before exhibiting a 1/f signature.<br />
PSD Amplitude [(g) 2 /Hz]<br />
Figure 9. SOA acceleration noise PSD.<br />
This 1/f (or “flicker”) noise is indicative of an acceleration<br />
drift instability that limits the net resolution capability of<br />
the accelerometer. This is also characterized by the Allan<br />
variance charts calculated from the PSD measurements<br />
shown in Figure 10. The missile guidance SOA resolution<br />
10<br />
10 -8<br />
10 -9<br />
10 -10<br />
10 -11<br />
Acceleration Uncertainty (g)<br />
10 -5<br />
10 -6<br />
Missile Guidance SOA<br />
White Noise -1/2 slope<br />
φ = 4.5 µg/√Hz<br />
VRW = 0.006 ft/s/√h<br />
The Silicon Oscillating Accelerometer<br />
Guidance SOA<br />
SOA-SINS<br />
10<br />
Frequency (Hz) [Fs = 1 sample/s]<br />
-12<br />
10-6 10-5 10-4 10-3 10-2 10-1 100 1 µg<br />
0.5 µg<br />
is limited to 0.5 µg, which is achieved over 100-s averaging<br />
times. Longer averaging times do not further improve<br />
resolution in the missile guidance SOA because of the 1/f<br />
drift instability, but the low-frequency extension of the<br />
white noise floor of the SINS SOA enables 80 nano-g<br />
resolution, which is achieved after 1000 s of averaging.<br />
The Allan variance plots the standard deviation of indicated<br />
acceleration against data averaging time. The white noise<br />
floor of the corresponding PSD can be determined from the<br />
portion of the Allan variance having a minus-one-half slope<br />
on the log scale. The equivalent white noise PSD can be calculated<br />
from a point on the minus-one-half slope line from:<br />
where:<br />
σ = acceleration standard deviation<br />
φ = white noise PSD<br />
T = averaging time<br />
(18)<br />
The data from both versions of the SOA (Figure 10) in the<br />
minus-one-half slope region indicate (from Eq. (18)) a<br />
white noise PSD of 4.5 µg/√Hz, which is equivalent to a<br />
velocity random walk coefficient of 0.006 ft/s/√h.<br />
Figure 11 shows SOA bias and SF uncertainty measured<br />
on a missile guidance SOA in a 2-position test, and Figure<br />
12 shows SF and bias uncertainty measured on an SOA-<br />
SINS unit in a 4-position test. The SF and bias data shown<br />
in these figures are uncompensated SOA output and are<br />
the average values over a 30-min dwell as determined from<br />
1-s averaged data. The data shown extend over a roughly<br />
3-day period and show 1σ standard deviations in SF and<br />
bias of 0.56 ppm and 0.92 µg, respectively, for the missile<br />
guidance SOA. The quieter SOA-SINS unit demonstrates<br />
1σ uncertainty in SF and bias of 0.14 ppm and 0.19 µg,<br />
respectively.<br />
SOA-SINS<br />
10<br />
Time (s)<br />
-7<br />
10<br />
Time (s)<br />
-8<br />
10<br />
Figure 10. SOA Allan variance.<br />
0 101 102 103 104 105 100 101 102 103 104 105 10 -4<br />
10 -5<br />
10 -6<br />
10 -7<br />
White Noise (-1/2 slope)<br />
φ = 4.5 µg/√Hz<br />
VRW = 0.006 ft/s/√h<br />
1 µg<br />
0.08 µg
SF (ppm) & Bias (µg)<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
-1.0<br />
-1.5<br />
SF (1σ) = 0.56 ppm<br />
Bias (1σ) = 0.92 µg<br />
SF (ppm)<br />
Bias (µg)<br />
-2.0<br />
0.00 20.00 40.00<br />
Time (h)<br />
60.00 80.00<br />
Figure 11. Missile guidance SOA SF and bias stability.<br />
SF (ppm) & Bias (µg)<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
-0.1<br />
-0.2<br />
-0.3<br />
-0.4<br />
4 Position Calibration<br />
SF (1σ) = 0.14 ppm<br />
Bias (1σ) = 0.19 µg<br />
-0.5<br />
0 10 20 30 40 50 60 70<br />
Time (h)<br />
Figure 12. SOA-SINS SF and bias stability.<br />
The SOA also demonstrates sub-ppm and sub-µg performance<br />
across multiposition tumble tests. The following error<br />
model was used to fit the tumble data:<br />
(19)<br />
where:<br />
aind = indicated acceleration output<br />
f = net (differenced) SOA frequency output<br />
SF = SOA scale factor (Hz/g)<br />
bias = SOA bias frequency<br />
ai = input axis applied acceleration<br />
am, ao = cross axis applied accelerations<br />
K2, K3 = second- and third-order SF nonlinearity<br />
Kim, Kio = cross axis coupling coefficients<br />
Kmm, Koo = cross axis nonlinearity coefficients<br />
The Silicon Oscillating Accelerometer<br />
Table 2 shows the nominal values of the above parameters<br />
for the SOA as determined in a 16-position tumble<br />
test. Note that a value for the Kmm and Koo coefficients<br />
can not be extracted from a single-orientation, multiposition<br />
tumble. The contribution of these terms is<br />
absorbed in the nominal values of bias and the K2 coefficient.<br />
Table 2 also shows the 1σ variation in the<br />
coefficients during the tumble, a test that takes 9 h to<br />
complete.<br />
Residual (µg)<br />
Table 2. Multiposition Tumble Results.<br />
Tumble about OA Tumble about MA<br />
Parameter Nominal Std. Dev. Nominal Std. Dev.<br />
Value (1σ) Value (1σ)<br />
Bias 797.965 0.354 797.724 0.295<br />
Hz µg Hz µg<br />
Scale Factor 126.697 0.989 126.722 0.822<br />
Hz/g ppm Hz/g ppm<br />
K2 121.33 0.652 117.49 0.542<br />
µg/g 2 µg/g 2 µg/g 2 µg/g 2<br />
K3 6.167 1.324 0.24 1.10<br />
µg/g 3 µg/g 3 µg/g 3 µg/g 3<br />
KimKio -7.293 0.629 -6.65 0.523<br />
µg/g 2 µg/g 2 µg/g 2 µg/g 2<br />
Misalignment 2.34 0.295 28.3 0.245<br />
mrad µrad mrad µrad<br />
Figure 13 is a plot of the residual (i.e., the difference<br />
between the actual SOA output and the fitted error<br />
model) for multiposition tumble tests about the instrument’s<br />
two cardinal axes. Note that the 1σ standard<br />
deviations of the residuals are less than 1 µg (0.95 and<br />
0.79 µg) and the residual plot does not show a systematic<br />
variation with table angle, an indication that there<br />
is not a major unmodeled error term present in the<br />
sensor.<br />
OA Residual (1σ) = 0.95 µg<br />
MA Residual (1σ) = 0.79 µg<br />
Table Angle (deg)<br />
OA Tumble<br />
MA Tumble<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
-1.0<br />
-1.5<br />
-2.0<br />
0 100 200 300 400<br />
Figure 13. Multiposition tumble residuals.<br />
11
Finally, Figures 14 and 15 show long-term (30-day) SF<br />
and bias stability measured on two SOA-SINS units (S/Ns<br />
028 and 063). The better of the two units demonstrates a<br />
1σ SF stability of 0.73 ppm and a 1σ bias stability of<br />
approximately 2 µg over 30 days.<br />
ppm<br />
µg<br />
12<br />
S/N 028 Avg. SF = 247.4 Hz/g<br />
S/N 063 Avg. SF = 258.8 Hz/g<br />
S/N 028 SF (1σ) = 0.73 ppm<br />
S/N 063 SF (1σ) = 1.22 ppm<br />
Days<br />
Figure 14. Long-term SOA-SINS SF stability.<br />
Figure 15. Long-term SOA-SINS bias stability.<br />
The Silicon Oscillating Accelerometer<br />
SF 063<br />
SF 028<br />
Linear (SF 063)<br />
Linear (SF 028)<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
dSF/SF = -0.102 ppm/day<br />
-1.0<br />
-1.5<br />
dSF/SF = -0.011 ppm/day<br />
-2.0<br />
0 5 10 15 20 25 30 35 40<br />
S/N 028 Bias (1σ) = 2.04 µg<br />
S/N 063 Bias (1σ) = 2.68 µg<br />
Days<br />
Bias 063<br />
Bias 028<br />
Linear (Bias 063)<br />
Linear (Bias 028)<br />
5.0<br />
4.0<br />
3.0<br />
2.0<br />
1.0<br />
0.0<br />
-1.0<br />
-2.0<br />
dBias = 0.220 µg/day<br />
-3.0<br />
-4.0<br />
dBias = 0.064 µg/day<br />
-5.0<br />
0 5 10 15 20 25 30 35 40<br />
CONCLUSIONS<br />
<strong>Draper</strong> <strong>Laboratory</strong> is currently developing the SOA, a<br />
MEMS-based inertial-grade sensor. Performance data<br />
acquired to date meet requirements for missile guidance<br />
missions and for high-performance navigation applications<br />
such as the SLBM SINS navigation systems.<br />
The missile and SSBN applications have significantly different<br />
environmental, acceleration dynamic range, and<br />
resolution requirements that are best satisfied by optimizing<br />
the SOA geometry for each application. The design<br />
flexibility and wafer-scale fabrication methods of the silicon<br />
MEMS process enable manufacturing both instrument<br />
designs with essentially zero incremental cost associated<br />
with the additional instrument assembly line. That is, both<br />
versions of the SOA share a common sensor package, electronics<br />
architecture, main housing, and instrument<br />
assembly process.<br />
The MEMS technology is low cost and offers a rapidly<br />
expanding commercial business base to leverage and sustain<br />
accelerometer production and deployment in<br />
next-generation strategic system applications. Integration of<br />
the SOA with mixed-signal, application-specific integrated<br />
circuit (ASIC) electronics offers the promise of a low-cost,<br />
small-size, low-power, and high-reliability strategic-grade<br />
instrument.<br />
REFERENCES<br />
[1] Tang, W., Electrostatic Comb Drive for Resonant Sensor and<br />
Actuator Applications, PhD Dissertation, University of California<br />
at Berkeley, November 21, 1990.<br />
[2] Kourepenis, A., J. Borenstein, J. Connelly, R. Elliott, P. Ward, M.<br />
Weinberg, “Performance of MEMS Inertial Sensors,” 24th Joint<br />
Services Data Exchange for Guidance, Navigation, and Control,<br />
November 16-20, 1998, Anaheim, CA.<br />
[3] Roark and Young, Formulas for Stress and Strain, 5th Edition,<br />
McGraw Hill, 1975.<br />
[4] Nayfeh, Mook, Nonlinear Oscillations, Wiley & Sons, New York,<br />
1979.<br />
[5] Hays, K.M., R.G. Schmidt, W.A. Wilson, J.D. Campbell, D.W.<br />
Heckman, M.P. Gokhale, A Submarine Navigator for the 21st Century, The Boeing Co., IEEE 0-7803-7251-4/02, 2002, pp.<br />
179-188.<br />
[6] Borenstein, J.T., N.D. Gerrish, R. White, M.T. Currie, E.A.<br />
Fitzgerald, “Silicon Germanium Epitaxy: A New Material for<br />
MEMS,” Mat. Res. Soc. Symp., Vol. 657, 2000, pp. 7.4.1.
The Silicon Oscillating Accelerometer<br />
(l-r) William Sawyer, Ralph Hopkins,<br />
Roy Setterlund, and Bruce Dow<br />
Ralph Hopkins is a Principal Member of the Technical Staff and Group Leader in the Guidance Hardware Division at<br />
<strong>Draper</strong> <strong>Laboratory</strong>. Currently, he is Technical Director of the SOA development program, a high-performance silicon<br />
MEMS VBA targeted for strategic-grade applications. He has also led and contributed to the development of navigation<br />
and tactical-grade MEMS gyroscopes and accelerometers, and high-performance electromechanical inertial sensors,<br />
such as floated instruments and dynamically-tuned gyros. He holds several patents, has authored several papers,<br />
and is an invited speaker for tutorials on inertial instruments and inertial technology. He is a current member of the<br />
AIAA Guidance Navigation and Control Technical Committee. Mr. Hopkins holds BS and ME degrees in Mechanical<br />
Engineering from Rensselaer Polytechnic Institute, an ME in Engineering Mechanics from Columbia University, and an<br />
MS in Engineering Management from The Gordon Institute of Tufts University.<br />
Joseph Miola joined the <strong>Laboratory</strong> in 1959 and was in the Systems Integration, Test, and Evaluation Division before<br />
his retirement in 2005. He worked in the development programs for Navy Polaris and Trident and for Air Force<br />
Minuteman and Peacekeeper guidance systems, primarily in accelerometer and gyro instrument design and test.<br />
Experience included over 25 years in management with Section Chief and Associate Director positions and Task Leader<br />
for the A-10 GPS/IDM Integration Test Program. Mr. Miola was responsible for the test and evaluation of the SOA 3<br />
accelerometer design. He received a BS in Engineering and an MS in Electrical Engineering from Northeastern University.<br />
Roy Setterlund is an Associate Director in <strong>Draper</strong>’s Strategic Systems Program Office and is responsible for Air Force<br />
ICBM and Navy/Air Force reentry programs. These programs consist of ICBM sustainment activities related to the<br />
Minuteman III guidance system, new strategic instrument development work as part of the Air Force’s Guidance<br />
Applications Program (GAP), various science and technology efforts sponsored by the Air Force Research <strong>Laboratory</strong><br />
(AFRL), and reentry instrumentation and guidance development activities for the strategic Navy’s reentry branch, SP28.<br />
He also oversees <strong>Draper</strong>’s IR&D related to strategic systems technology. Prior to this, Mr. Setterlund was a Business<br />
Development Manager in <strong>Draper</strong>’s Space and Missile’s Program Office, where he concentrated on business development<br />
for space systems, including spacecraft attitude control systems, small satellites, and precision pointing and stabilization<br />
applications. He has published many papers for various professional journals and has worked at <strong>Draper</strong> <strong>Laboratory</strong> since<br />
graduating from MIT with an MS in 1973.<br />
Bruce Dow is a Program Manager in the Strategic Systems Directorate at <strong>Draper</strong>. He has over 15 years experience in<br />
systems engineering and test of advanced guidance, navigation, and control (GN&C) systems and over 3 years experience<br />
in technical director and program management positions supporting both Navy and Air Force ballistic missile<br />
guidance and reentry system applications. He has been a Systems Engineer for undersea and aerial autonomous vehicles<br />
and has also served as Technical Director and Program Manager for various ballistic missile guidance and reentry<br />
system programs in support of AFRL, Navy SP23, and Navy SP28. Mr. Dow has also served as Program Manager for<br />
the design and installation of DoD and DoJ intelligence network systems world-wide, including the White House<br />
Communications Agency, Navy Space Command, National Drug Intelligence Center, Office of Naval Intelligence, and<br />
Joint Analysis Center. Mr. Dow has spent 8 years on active military duty and is a Lieutenant Colonel in the U.S. Army<br />
Reserves, deployed from 2003 to 2004 for Operation Iraqi Freedom. He holds a BA from the U.S. Military Academy,<br />
and an MS in Systems Engineering and an MBA, both from Boston University.<br />
William Sawyer has been employed at <strong>Draper</strong> <strong>Laboratory</strong> for 7 years, during which time he has worked on the manufacturing<br />
process for several <strong>Draper</strong> inertial MEMS devices, including the Tuning-Fork Gyro (TFG)14 and TFG20.<br />
For the past several years, he has concentrated on the SOA. He developed the Bonded Etchback of Silicon (BESOI)<br />
process in collaboration with Jeff Borenstein. This work led to the patent for which Dr. Sawyer and Dr. Borenstein<br />
received <strong>Draper</strong>’s 2004 Best Patent Award. He received a BS in Physics from Colby College (1978), a Diplome (Masters)<br />
in Physics from Albert Ludwig University, Freiburg, Germany (1985), and a PhD in Physics from the University of<br />
Stuttgart, Stuttgart, Germany (1989).<br />
13
Autonomous Guidance, Navigation, and Control of<br />
Large Parafoils<br />
David Carter, Sean George, Philip Hattis, Leena Singh, Steven Tavan<br />
Copyright © 2005 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc.<br />
Published by the American Institute of Aeronautics and Astronautics, Inc., with permission<br />
INTRODUCTION<br />
Under the Joint Precision Airdrop System (JPADS) program, a <strong>Draper</strong> <strong>Laboratory</strong> autonomous<br />
guidance, navigation, and control (GN&C) software package that enables precision payload airdrop<br />
delivery using large parafoils has been developed in prototype form and successfully flight tested.<br />
The modular software design is structured to accommodate parafoil airdrop systems for payloads<br />
ranging from under 2,000 lb to over 30,000 lb. The initial GN&C software implementation has been<br />
demonstrated on the Para-Flite Dragonfly 10,000-lb class parafoil using an airborne guidance unit<br />
(AGU) provided by Wamore, Inc. and an avionics package provided by RoboTek. Among the<br />
primary avionics selection criteria was low component cost, resulting in the use of a processor with<br />
very limited data throughput capability. To accommodate the processor limits, the guidance<br />
algorithm includes table-driven trajectory data that guide the parafoil through precision finaldescent<br />
maneuvers while imposing very limited processor throughput burden. The GN&C<br />
algorithms and associated mission planning software have also been incorporated into the Precision<br />
Airdrop System (PADS) laptop personal computer. This accommodates easy, in the field, ground<br />
loading of the GN&C software onto the AGU and enables PADS updates of the airdrop system<br />
mission files during flight of the carrier aircraft to the airdrop release point. The details of the GN&C<br />
design and flight test results to date are discussed.<br />
A key element of the JPADS Advanced Concept<br />
Technology Demonstration (ACTD) is the development<br />
of GN&C software to autonomously fly the Dragonfly<br />
10,000-lb-capable parafoil. This software must guide the<br />
parafoil from deployment altitudes up to 25,000 ft above<br />
mean sea level (MSL) to landings within a 100-m circular<br />
error probable (CEP) of the target. Other key goals<br />
include robustness to a variety of failure modes, algorithms<br />
that are sufficiently generic to facilitate adaptation<br />
to both smaller and larger decelerators, efficient enough<br />
14<br />
to perform well on a very modest microprocessor, and<br />
capable of meeting system performance requirements<br />
with a navigation sensor suite limited by recurring system<br />
costs. Also important are government ownership of<br />
the resulting code, the ability to handle user-supplied<br />
waypoints, plus efficient and cost-effective integration<br />
with the previously developed Air Force and Army PADS<br />
mission planner. <strong>Draper</strong> <strong>Laboratory</strong>, one of the developers<br />
of the PADS planner, is implementing the<br />
autonomous GN&C.
The Flight Hardware Configuration<br />
As noted, an important goal of this program is to keep<br />
recurring system costs at a minimum. Hence, the avionics<br />
selected are the ultimate in simplicity. The sole navigation<br />
sensor is the CSI Wireless Vector dual-Global Positioning<br />
System (GPS) receiver, which provides not only system<br />
position and velocity, but also heading and heading rate.<br />
The Vector unit provides this by using two tightly coupled<br />
high-performance GPS receivers and two antennae, separated<br />
by 10 in, providing a good balance between heading<br />
accuracy and GPS acquisition time. This approach avoids<br />
including a magnetic compass for the heading reference,<br />
with its attendant difficulties due to local changes in the<br />
magnetic field and field perturbations from the various<br />
metal masses in the AGU. Other means of determining<br />
heading without additional heading sensors were examined<br />
and determined to be operationally unattractive. The<br />
flight processor selected is the Rabbit RCM3400 microcontroller,<br />
augmented by 8 MB of flash memory to hold<br />
guidance data tables and record GN&C parameters during<br />
flight for later analysis. A Freewave 902- to 928-MHz<br />
spread-spectrum modem is also utilized to receive remote<br />
control commands from the ground and to downlink mission<br />
data for post-flight analysis. This modem can be used<br />
to download mission files prior to airdrop system release,<br />
although 802.11-g wireless network components are<br />
being integrated as a replacement for operational use.<br />
Basis for the Initial Airdrop System Dynamics Models<br />
<strong>Draper</strong> has a long history of research in guided ram-air<br />
parafoils, beginning with work on the Precision Guided<br />
Airdrop System (PGAS). [1],[2] As part of this program, an<br />
engineering simulation was implemented based on<br />
numerous technical references detailing the appropriate<br />
structure for a 6-degree-of-freedom (DOF) parafoil<br />
dynamics model complete with high-fidelity environment<br />
models. The initial Dragonfly dynamics models were<br />
derived from this simulation framework and then updated<br />
substantially using the best available experimental and<br />
theoretical analyses of large-scale parafoil systems. In particular,<br />
a wealth of incredibly detailed flight performance<br />
data have been generated by the NASA X-38 program on<br />
two parafoil systems in the same size category as the<br />
Dragonfly system. [3],[4] In addition, an updated theoretical<br />
treatment of apparent mass effects was undertaken in<br />
Reference [5], the results of which were rolled into the initial<br />
Dragonfly simulation. It was understood from the start of<br />
the program that there would be opportunities for<br />
conducting performance estimation tests on the new<br />
canopy, therefore, most of the initial work focused on<br />
developing tools that could be used to quickly update the<br />
original models as performance data were collected. It<br />
should be understood, however, that due to the lack of<br />
inertial instrumentation onboard the Dragonfly system, as<br />
well as a program focus on cost effectiveness, no attempt<br />
has been made to complete a detailed system identification<br />
of all parameters in the Dragonfly parafoil model. Rather,<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
the focus of the modeling effort has been on matching: (1)<br />
steady-state velocity and turn rate, (2) toggle line actuator<br />
dynamics, and (3) the basic turn rate time constant. Given<br />
the uncertainty in measurement data as well as wind<br />
conditions during flight testing, it is believed that engineering<br />
models suitable for GN&C development have<br />
been generated.<br />
Analysis tools have been developed that attempt to fit<br />
longitudinal lift-to-drag and velocity flight data to relatively<br />
simple mathematical expressions for the lift, drag, and<br />
pitch moment characteristics of the parafoil. The parafoil<br />
aerodynamics model is used to generate tabular force and<br />
moment coefficient data as a function of brake setting and<br />
angle of attack as inputs into the simulation. Values for<br />
various aerodynamic parameters were originally set using<br />
past historical data collected on PGAS as well as general<br />
trends seen in the X-38 program. The process to ascertain<br />
whether the parametric aerodynamics model had sufficient<br />
complexity was to attempt to match X-38 L/D and velocity<br />
performance. The aerodynamics model was able to reasonably<br />
produce the steady-state response for this large<br />
parafoil airdrop system; therefore, there was confidence<br />
that subsequent test data on the Dragonfly could be used<br />
to give a good engineering model of the flight performance.<br />
The main uncertainty in “fitting” aerodynamic<br />
parameters against the flight data was that, unlike X-38<br />
flights, angle-of-attack was not measured explicitly by the<br />
Dragonfly AGU. Instead, trends from the X-38 program,<br />
coupled with ensuring physically reasonable constraints<br />
on aerodynamic parameters, were used to help produce a<br />
longitudinal model for the Dragonfly. Detailed analysis of<br />
flight data is shown in the Flight Test Program section:<br />
“System Identification from Flight Test Data,” with comparisons<br />
from the parafoil model updated to match the<br />
steady-state response of the system.<br />
The lateral model parameters for the Dragonfly were based<br />
on both theoretical calculations as well as appropriately<br />
scaled historical data from the PGAS and X-38 programs.<br />
Originally, a very complicated yaw rate response model<br />
was used that tried to match the nonlinear steady-state<br />
turn characteristics seen with the X-38 at low brake<br />
settings. Subsequent flight tests did not show this type of<br />
behavior, therefore, the turn rate response was augmented<br />
to a more linear relationship with differential toggle (see<br />
the Flight Test Program section: “System Identification<br />
from Flight Test Data”). The turn rate dynamics are generally<br />
dominated by two factors: the inertial (both “real” and<br />
apparent) properties of the system and the toggle actuator<br />
response. Effort was given to an appropriate dynamics<br />
model for the toggle motors. Generally, the Dragonfly<br />
motors are capable of ~2 ft/s of maximum line pull rate<br />
and have relatively high acceleration characteristics. Fullscale<br />
toggle changes from 0 to 100 in general take ~5 s. In<br />
addition, yaw inertial and damping characteristics were<br />
originally chosen to give the parafoil turn rate response a<br />
5-s time constant. The overall sluggish turn behavior of<br />
15
the parafoil was well modeled by the original model<br />
parameters used, and generally corresponded to a fairly<br />
highly damped large inertia system. Subsequent flight<br />
tests were used to slightly modify the parameters, but in<br />
general, the original lateral model was sufficient to give the<br />
gross dynamic behavior of the Dragonfly system.<br />
The GN&C Design<br />
Top-Level Overview<br />
The GN&C flight software resides in the AGU. While the<br />
airdrop system is in flight, the software receives information<br />
from a navigation package about system position, velocity,<br />
and heading. Based on these inputs, it computes a trajectory<br />
toward the programmed target landing point. It then flies<br />
the system toward that point via a series of extensions<br />
and/or retractions of the parafoil’s two control lines.<br />
After a stable canopy opening, the GN&C software optionally<br />
enters a control line trim mode in which it adjusts the<br />
control line lengths for straight flight. After completing (or<br />
skipping) the trim mode, state estimates derived by filtering<br />
the position, velocity, and heading inputs from the navigation<br />
package are utilized by guidance to begin directing<br />
the control algorithms to home the airdrop system toward<br />
the target. When the airdrop system comes within about<br />
200 m horizontal distance from the target, it enters an<br />
energy management mode, flying figure-eight patterns in<br />
the vicinity of the target until ground-relative altitude<br />
drops below 500 m, at which point, it steers toward the<br />
target to establish the final approach. From this point until<br />
landing, the guidance software utilizes precomputed steering<br />
commands from a lookup table with a large family of<br />
trajectories that minimize final position and heading error.<br />
This table-driven approach minimizes the trajectory determination<br />
processing burden on the available, small,<br />
low-throughput flight processor.<br />
The navigation instrument is a CSI wireless, dual-GPS<br />
receiver-based navigation package that generates position,<br />
velocity, and heading data. The processed GPS data from<br />
the navigation package is filtered by an algorithm in the<br />
GN&C software, with the resulting state estimates sent to<br />
the guidance software in the form of airdrop system altitude,<br />
heading, and estimated wind velocity. The wind estimates<br />
are derived by comparing the GPS velocity data with a<br />
model of the expected parafoil air speed.<br />
Using gain scheduling, the control software attempts to<br />
maintain the heading rate commanded by the guidance<br />
software by extending or retracting the left and right<br />
parafoil control lines. The control software also attempts to<br />
maintain symmetric operation of the control lines about a<br />
base retraction length when determining the line positions<br />
that will achieve the desired heading rate. Also, a parafoil<br />
flare capability, achieved by fully retracting both control<br />
lines, achieves a stall condition that is used just before<br />
touchdown to reduce the parafoil’s ground impact velocity.<br />
Development of the GN&C software began in<br />
Government FY 2004. <strong>Draper</strong> developed, integrated, and<br />
16<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
validated the autonomous GN&C flight software for initial<br />
use on the Para-Flite 10,000-lb-capable parafoil equipped<br />
with the Wamore, Inc. AGU and RoboTek-supplied avionics.<br />
All aspects of the GN&C design are modularized to readily<br />
accommodate eventual adaptation to other smaller and<br />
larger guided parafoil airdrop systems. Also, both the<br />
GN&C software and mission planning algorithms associated<br />
with the use of guided airdrop systems that apply the<br />
GN&C algorithms have been hosted onto the PADS<br />
mission planner laptop personal computer (PC). This will<br />
facilitate the use of the PADS PC as a common platform to<br />
download the GN&C software and data files to the AGU on<br />
the ground, as well as to generate airdrop load-specific<br />
mission files shortly before release onboard the carrier aircraft.<br />
GN&C Development, Integration, and Test Methodology<br />
The development, integration, and test methodology<br />
applied by <strong>Draper</strong> staff has followed a rigorous process<br />
tailored to the flight prototype demonstration objectives<br />
applicable to the current program. The key process steps<br />
are summarized as follows:<br />
• Algorithm conceptualization, preliminary design, and<br />
evaluation: This development phase involved preliminary<br />
algorithm design and off-line evaluation of the<br />
algorithms by the individual developers. This process<br />
included internal and external (customer) reviews of the<br />
proposed algorithms before their integration into a<br />
complete GN&C implementation.<br />
• GN&C integration and software simulation assessment:<br />
A 6-DOF model of the parafoil dynamics was formulated<br />
and validated against available parafoil test data. A software<br />
simulation was developed using the 6-DOF<br />
parafoil model that provided data interfaces for GN&C<br />
software equivalent to those on the AGU. The GN&C<br />
algorithms were then integrated together and into the<br />
simulation. Closed-loop GN&C assessments were then<br />
performed using this simulation, the results of which<br />
were subjected to internal and customer review. This<br />
simulation tool has subsequently supported the evaluation<br />
of flight test results.<br />
• Hardware-in-the-loop (HWIL) simulation assessment:<br />
An HWIL simulation was assembled at <strong>Draper</strong>, using<br />
actual AGU processors and memory as well as a set of<br />
control line actuation motors. This HWIL facility was<br />
used to evaluate proper code function with the real<br />
timing, space, and word-length limitations of the target<br />
processor and memory, and was used to demonstrate<br />
proper message handling by the control actuation<br />
motors. Subsequently, the CSI navigation package was<br />
added. The navigation package provided a means to<br />
evaluate proper message interfacing, although the static<br />
nature of the HWIL simulation assembly prevents the<br />
use of the actual navigation package during flight<br />
dynamics simulations.<br />
• Preflight code validation and freeze: Prior to each flight<br />
test cycle, the intended GN&C implementation was
integrated and validation tested, first on the software<br />
simulation, and then on the HWIL simulation. When<br />
the intended performance was realized, the GN&C software<br />
was frozen as a defined code release under a<br />
configuration control process, and it was then delivered<br />
for field test use. A final integration test was performed<br />
at the AGU vendor’s facility prior to the first flight of<br />
each new software release.<br />
Guidance Implementation Details<br />
The primary function of guidance is to compute steering<br />
commands for use by control, given position, heading, and<br />
estimated wind profile from navigation. As a part of the<br />
steering calculation, guidance is responsible for timing the<br />
flare (deep brakes) maneuver prior to landing, signaling<br />
mode control to transition to flare; mode control then<br />
commands the other parts of the system accordingly.<br />
Throughout flight, guidance accepts the system mode<br />
from mode control. The mode can be any of the following:<br />
initialize, preflight, trimflight, autoflight, manual, or terminal.<br />
Steering calculations, including timing of the flare maneuver,<br />
are performed only when system mode is autoflight. When<br />
the mode is trimflight, guidance monitors altitude and distance<br />
to the first waypoint (which could be the target); if<br />
the ratio of distance to altitude becomes sufficiently small,<br />
guidance requests that trim be abandoned, causing transition<br />
to autoflight mode. In the modes initialize, preflight,<br />
manual, and terminal, guidance monitors system position<br />
but is otherwise inactive.<br />
Guidance obtains MSL altitude, the north and east<br />
position coordinates with respect to the target, flight path<br />
azimuth (the direction of the air velocity vector), as well as<br />
a table of wind velocity vs. altitude from navigation. It also<br />
accepts a list of waypoints, given as north and east offsets<br />
from the target, from the mission loads file. During the<br />
autoflight mode, guidance computes the commanded rate<br />
of change of the flight path azimuth, which is referred to<br />
as “turning rate,” and sends this to control. Guidance<br />
interfaces are shown in Figure 1.<br />
Mode<br />
Control<br />
Mode<br />
Navigation Wind Table Guidance<br />
Mission<br />
Loads<br />
h, x, y, χ<br />
Waypoints<br />
Figure 1. Dragonfly guidance interfaces.<br />
Guidance does its calculations in a deweighted, wind-fixed<br />
frame; a portion of the anticipated displacement due to<br />
wind is added to the navigated position with respect to<br />
the next waypoint (or the target) so that steering calculations<br />
can be done as if winds were zero. A nominal, single<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
χ, dχ/dt<br />
Submode<br />
Abandon<br />
Trim<br />
Control<br />
Mode<br />
Control<br />
waypoint trajectory in this wind-fixed frame is shown in<br />
Figure 2. The guidance strategy is best understood by<br />
considering this trajectory, which is marked to show the<br />
different flight phases (modes and submodes).<br />
Due North (>0) target relative position (ft)<br />
7000<br />
6000<br />
5000<br />
4000<br />
3000<br />
2000<br />
1000<br />
0<br />
-1000<br />
Table<br />
Lookup<br />
Trim<br />
Homing<br />
Flare<br />
Energy<br />
Management<br />
-6000 -5000 -4000 -3000 -2000 -1000 0 1000 2000 3000 4000<br />
Due East (
Navigation Implementation Details<br />
Navigation uses data from a dual-antenna GPS navigation<br />
package to compute position and flight path azimuth for<br />
use by guidance, and flight path azimuth and flight path<br />
azimuth rate (turning rate) for use by control. Using GPSmeasured<br />
velocity with respect to the ground together<br />
with a simple analytic model of system air speed and GPSmeasured<br />
heading, navigation estimates wind velocity. The<br />
wind velocity estimate is used to correct an a priori table<br />
of wind vs. altitude; the correction is largest for altitudes<br />
close to the altitude at which navigation’s wind estimate is<br />
valid. Navigation sends the corrected wind table to guidance.<br />
Early in the flight, navigation monitors the sink rate<br />
in order to detect canopy inflation; this information is<br />
important for system moding. Navigation interfaces are<br />
shown in Figure 3.<br />
Figure 3. Dragonfly navigation interfaces.<br />
Navigation receives standard National Marine Electronics<br />
Association (NMEA) data messages from the dual-antenna<br />
GPS navigation system. Coordinated Universal Time<br />
(UTC), latitude, longitude, and altitude MSL are<br />
obtained from the “GGA” data message. The GGA message<br />
also contains a quality indicator, the number of<br />
satellites used in the position calculation, and an estimate<br />
of horizontal dilution of precision (HDOP). Navigation<br />
considers the GPS position data to be valid provided the<br />
time has been updated from its previous value, the quality<br />
indicator takes values of 1 or 2, the number of<br />
satellites used in the position fix is at least 4, and HDOP<br />
does not exceed a configurable threshold whose current<br />
value is 10.<br />
Navigation receives ground velocity (speed and course)<br />
from the “VTG” data message. This message is considered<br />
to be valid whenever its numeric fields are fully populated.<br />
Navigation receives true heading from the “HDT”<br />
data message and heading rate (rate of turn) from the<br />
“ROT” message. These messages are considered valid<br />
when their numeric fields are populated; this happens<br />
only when the receiver has lock on both channels.<br />
When the HDT message is valid, navigation sets the<br />
flight path azimuth equal to the HDT heading. This<br />
amounts to asserting that sideslip is zero, an approximation<br />
that seems to work well for low bandwidth GN&C<br />
18<br />
Mode<br />
Control<br />
GPS<br />
Interface<br />
Mission<br />
Loads<br />
Mode<br />
h, x, y, χ<br />
t, h, lat, lon, v,<br />
Wind<br />
course<br />
Table<br />
ψ, dψ/dt, χ, dχ/dt,<br />
Validity Data<br />
Impact Point<br />
Navigation<br />
Submode<br />
Release Point<br />
a priori<br />
Wind Table<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
Submode<br />
Sink Rate<br />
Stabilized<br />
Guidance<br />
Control<br />
Mode<br />
Control<br />
of large parafoil systems. Likewise, navigation identifies<br />
flight path azimuth rate with heading rate from the ROT<br />
message when this message is valid. Wind estimates are<br />
made only when valid GGA, VTG, and HDT data are<br />
available. Horizontal velocity with respect to the ground<br />
is obtained from the VTG message. Altitude is obtained<br />
from GGA and used together with data provided in the<br />
mission loads file to compute atmospheric density and<br />
nominal air speed (which depends on atmospheric density).<br />
Air velocity is estimated using computed nominal<br />
air speed, nominal flight path angle, and measured heading<br />
from HDT; the calculation assumes that sideslip is<br />
negligible. The air velocity estimate is low-pass filtered<br />
and subtracted from ground velocity to obtain the wind<br />
estimate.<br />
Control Implementation Details<br />
When formulating the control algorithm for Dragonfly,<br />
we stipulated the following control objectives: (1) the<br />
closed-loop system must track wind-relative lateral rate<br />
commands issued from guidance, and (2) the control<br />
algorithm must reject external disturbances, in particular,<br />
the frequency of the payload oscillation relative to the<br />
canopy. In this phase of the program, control only regulates<br />
lateral directional heading rate errors based on<br />
commands produced by guidance; flight speed is treated<br />
as a system parameter governed by the base (nominal<br />
brake setting) deflection during the flight. From initial,<br />
manually-controlled, system-identification parafoil flight<br />
drops, we identified the heading and sideslip rate<br />
dynamics in response to the lateral controls or differential<br />
toggle commands (right toggle – left toggle). We<br />
operate the vehicle at a 25% base deflection of 50 in<br />
since this allows the parafoil turning dynamics to be<br />
nearly linear and ensures that flight operations do not<br />
show the reflexive behavior that was observed in the<br />
small toggle deflection dynamics of the X-38 parafoil<br />
canopy. Around this base deflection, we collectively represented<br />
the heading and slideslip lateral dynamics mode<br />
shapes in terms of flight-path azimuth (χ = Ψ – β);<br />
experimental results showed that heading rate and<br />
slideslip had nearly identical natural modes, so we were<br />
able to collect the terms. We identified a second-order<br />
heading rate model shown in Eq. (1) using parameter<br />
identification techniques:<br />
(1)<br />
where ∆ is the control deflection difference between the<br />
left and right toggles and serves as the lateral control<br />
input. G represents the open-loop plant heading rate<br />
transfer function. Because we operate around a 25% base<br />
deflection, we achieve the desired deflection difference<br />
by symmetrically operating the line deflections about the<br />
base setting. Similarly, we identified the plant model<br />
parameters for symmetric toggle commands produced
about a 25% base deflection. The lateral control toggle<br />
commands, ∆, are issued by symmetrically deflecting the<br />
left and right toggles around the base deflection, b: δL =<br />
b - ∆/2; δR=b + ∆/2.<br />
Figure 4 shows the measured values of Dragonfly heading<br />
rate (dΨ/dt) and its sideslip rate in response to a<br />
100-in differential control toggle step from which the<br />
plotted flight path azimuth rate quantity was also<br />
derived. Based on simulations, it was determined that the<br />
natural open-loop bandwidth was 12.5 s with a damping<br />
coefficient of about 0.6. Figure 5 shows the derived<br />
azimuth rates obtained from the parafoil at two different<br />
toggle deflections as well as the corresponding response<br />
of the simulated second-order plant model that has the<br />
input gain, bandwidth, and damping characteristics<br />
determined from system identification. The second-order<br />
model can not capture very precisely the small overshoot<br />
combined with large settling times seen in the high<br />
deflection regimes. Nevertheless, the rise time, overshoot,<br />
and time constant are matched quite well.<br />
Heading Rate (deg/s)<br />
12<br />
10<br />
Azimuth Rate<br />
8<br />
6<br />
4<br />
Heading Rate<br />
2<br />
0<br />
-2<br />
-4<br />
Sideslip Rate<br />
-6<br />
0 10 20 30 40 50 60<br />
Time (s)<br />
Figure 4. Measured Dragonfly rate response for a 100-in<br />
differential control toggle step.<br />
Heading Rate (deg/s)<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
Time (s)<br />
Figure 5. Comparison of predicted and flight-measured<br />
Dragonfly rate response at two differential<br />
control toggle settings.<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
Azimuth Rate at 100-in deflection<br />
(from Flight Test Data)<br />
Azimuth Rate from Second-Order<br />
Matched Model at 100-in deflection<br />
Azimuth Rate from Second-Order<br />
Matched Model at 40-in deflection<br />
Azimuth Rate at 40-in deflection<br />
(from Flight Test Data)<br />
0 0 10 20 30 40 50 60<br />
Figure 6 shows the feedback control block interconnections.<br />
The controller rate errors are based on commands<br />
from guidance; it forms the left and right toggle commands<br />
to the motor controllers. Plant response is measured relative<br />
to the wind frame. Disturbances result from unexpected<br />
wind gusts, unknown and unmodeled plant dynamics,<br />
and payload oscillation frequencies that appear in the<br />
feedback signal. Based on this plant model, we identified<br />
control gains for a proportional, integral, derivative (PID)<br />
control structure that maximized the closed-loop bandwidth,<br />
provided at least 60 deg of phase margin, and<br />
minimized the gain of the auxiliary output-input loop to<br />
limit the impact of the payload oscillation frequency on the<br />
applied control. Since the payload oscillates with pendular<br />
motion beneath the AGU, these oscillations are picked up<br />
by navigation sensors and appear in the feedback signal<br />
even though the canopy itself does not exhibit significant<br />
oscillations. Therefore, one control objective was to synthesize<br />
the control gains so that the control line deflection<br />
command sent to the motors was desensitized to this<br />
disturbance signal and did not attempt to correct for this<br />
spurious measurement. Thus, we designed the control<br />
gains to manage the gain and phase margins of three significant<br />
transfer functions shown in Eq. (2). The transfer<br />
function P measures the gain between the commanded<br />
line deflection and the heading rate command error. S<br />
is the sensitivity function and T the closed-loop plant<br />
transfer function.<br />
Command<br />
from<br />
Guidance<br />
Plant Lateral<br />
Controller Dynamics<br />
r +<br />
- K G +<br />
ε<br />
Disturbance<br />
δleft<br />
δright<br />
Ψ Ψmeas<br />
Figure 6. Dragonfly feedback control block interconnections.<br />
We posed a multi-objective optimization problem and<br />
solved for the control gains that maximize three gain and<br />
phase quantities at suitable frequencies. This produced a<br />
control structure: , where (k1, k2,<br />
k3) are the outputs of the robust optimization problem.<br />
We used the following control gains in the controller: (k1<br />
= 1.65, k2 = 3.0, k3 = 1.07). The controlled-system transfer<br />
function characteristics are shown in Figures 7 and 8. We<br />
designed the control gains to maximize the bandwidth and<br />
phase margin of the controlled-system transfer function<br />
(T), minimized the gain of the input transfer function (D)<br />
at the payload natural frequency, and maintained low sensitivity<br />
at all frequencies.<br />
(2)<br />
19
Magnitude (dB)<br />
Phase (deg)<br />
-270<br />
10<br />
Frequency (rad/s)<br />
Figure 7. Dragonfly controlled system open-loop Bode plot.<br />
-1 100 Magnitude (dB)<br />
Phase (deg)<br />
0<br />
10<br />
Frequency (rad/s)<br />
Figure 8. Dragonfly controlled system transfer function<br />
(GK) sensitivity features.<br />
-1 100 20<br />
0<br />
-20<br />
-40<br />
-60<br />
-80<br />
0<br />
-90<br />
-180<br />
10<br />
0<br />
-10<br />
-20<br />
90<br />
60<br />
30<br />
Guided/Smart<br />
Airdrop<br />
Systems<br />
Navigator or<br />
Navigation<br />
Systems<br />
System Transfer Function<br />
Sensitivity Diagram<br />
Air Force Weather<br />
Agency Atmospheric<br />
Forecast Model - High:<br />
Resolution Nested Grid<br />
Surrounding Drop Zone (s)<br />
INTERNET/SIPRNET<br />
Mesoscale<br />
4D Field<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
PADS Laptop Computer<br />
3-D Field - Wind, Density,<br />
Pressure from Drop Time<br />
Reference<br />
Ballistic<br />
Trajectory<br />
5-km Grid Domain within<br />
15-km Grid Domain<br />
Assimilation<br />
Processor<br />
Computed Air<br />
Release Point<br />
(CARP)<br />
Supporting PADS Mission Planning and File Download<br />
Capability<br />
The PADS program has developed a laptop PC-based airdrop<br />
mission planning capability to support the determination of<br />
proper airdrop release points and to enable updates to guided<br />
airdrop system mission files while aboard the carrier<br />
aircraft in transit to the drop zone. The PADS implementation<br />
architecture is shown in Figure 9. The PADS PC includes<br />
a means to access current meteorological information and<br />
uses the altitude-dependent wind and density data, combined<br />
with models of the release and flight dynamics of<br />
airdrop systems to derive optimized computed air release<br />
points (CARPs) for unguided airdrop systems and allowable<br />
release envelopes for guided airdrop systems. Also, for the<br />
guided airdrop systems, nominal CARPs are designated<br />
within the derived release envelope. Meteorological data can<br />
be collected by PADS from any combination of the following<br />
sources: forecasts loaded before takeoff; data received<br />
through an encrypted satellite link during transit flight to the<br />
drop zone; and sondes released from or near the carrier aircraft,<br />
with the data retrieved by PADS through the carrier<br />
aircraft UHF antenna and processed into suitable form by<br />
software within PADS. All the available meteorological data<br />
are assimilated within PADS into a best estimate of current<br />
conditions near the drop zone.<br />
PADS has an interface to connect to a remote terminal on<br />
the carrier aircraft 1553 data bus to acquire the current<br />
vehicle navigation state, to obtain the current aircraft ataltitude<br />
wind measurement, and to monitor various<br />
airdrop-related mission parameters. The vehicle navigation<br />
state is used to enable PADS to display the aircraft position<br />
relative to the CARP and/or release envelope on a<br />
Airdrop<br />
Dynamics<br />
Simulation<br />
Figure 9. The PADS planning system architecture.<br />
Wind Data Sources<br />
• Satellite-Derived<br />
• TACMET Radiosonde<br />
• Theater Pilot Reports<br />
Combat Track II<br />
Radio<br />
Receiver<br />
Secure<br />
Interface<br />
Dropsonde<br />
Processor<br />
Radio<br />
Receiver<br />
Aircraft 1553<br />
Data Bus<br />
Communications<br />
Satellite<br />
Aircraft<br />
Top<br />
Antenna<br />
Aircraft<br />
Bottom<br />
Antenna<br />
GPS<br />
Dropsonde
FalconView graphical map interface (GMI). The at-altitude<br />
wind measurement serves as an additional source of meteorological<br />
data that is assimilated with the other available<br />
data. The airdrop-related mission parameters are recorded<br />
during release operations to enable post-release assessment<br />
of the release condition accuracy.<br />
PADS also includes a wireless link to enable loading mission<br />
file updates generated by PADS to guided airdrop systems<br />
while onboard the carrier aircraft. These mission file<br />
updates reflect the PADS-computed meteorological state in<br />
a format uniquely designed to be compatible with each airdrop<br />
system class.<br />
More details about the PADS design, current features, and<br />
flight test experience are provided in References [6]<br />
through [9]. Limited quantities of the PADS units are in<br />
field use supporting mission planning with unguided airdrop<br />
systems. Updates to PADS to support field use for<br />
mission planning and mission file updates for several<br />
classes of guided airdrop systems are now in work. Incountry<br />
operational demonstrations of these systems are<br />
expected presently.<br />
A version of PADS has been updated to include models of<br />
the Para-Flite Dragonfly 10,000 lb-class parafoil. This<br />
enables the PADS computer to provide mission file<br />
updates to the AGU prior to flight test release of the<br />
parafoil in tests of the autonomous GN&C system. This<br />
version of the PADS PC also includes a load of the GN&C<br />
software to enable the use of the same PC in the field to<br />
load that software onto the AGU via the wireless link or a<br />
temporary cable connection. This added PADS capability<br />
demonstrates the intended use of PADS as a common<br />
platform for preflight and in-flight support of all future<br />
airdrop operations from Air Force carrier aircraft (e.g., C-<br />
17s and C-130s).<br />
Flight Test Program<br />
Flight Test Program Overview<br />
Flight testing relevant to the development of the<br />
autonomous GN&C software commenced in March 2004<br />
at Red Lake in Kingman, Arizona. Initial flights in March<br />
and April were remote controlled, executing planned<br />
maneuvers to establish the flight characteristics of the<br />
Dragonfly system. <strong>Draper</strong> used the results of these flights<br />
to conduct system identification and to establish GN&C<br />
parameters as described elsewhere in this paper. First<br />
flight of the autonomous flight software occurred in May<br />
2004. Testing has continued since then at approximately<br />
6-week intervals, with flights starting in October at the<br />
Corral Drop Zone (DZ) at Yuma Proving Ground (YPG),<br />
Yuma, Arizona. During this time, the GN&C software was<br />
matured in parallel with the evolution of the canopy,<br />
rigging, and airborne hardware, including a major<br />
upgrade to the AGU involving new actuation motors,<br />
necessitating revised flight software motor interfacing. The<br />
move to YPG was a milestone as this was the first time the<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
system flew from a C-130 airplane, deploying at 130-kn<br />
indicated air speed (KIAS), considerably faster than the C-<br />
123 used in Kingman. Flights from military aircraft<br />
commenced in February 2005. As the flight test program<br />
proceeded, system weights were gradually increased up to<br />
the Dragonfly maximum of 10,000 lb, as were drop altitudes,<br />
heading toward a goal of flights from 18,000 ft MSL<br />
by spring 2005. Initial autonomous flights were deployed<br />
directly over the targeted impact point, and then gradually<br />
more offset from the target was introduced. GN&C software<br />
was initialized in early tests assuming no winds, then<br />
forecast winds were used, and eventually flight tests will<br />
include updates of the GN&C mission file while en route<br />
to the DZ with current winds estimates based on an assimilation<br />
of forecast and dropsonde wind and density data.<br />
System Identification from Flight Test Data<br />
As mentioned previously, flight tests were used to update<br />
the original models of the Dragonfly. A sequence of manual<br />
input tests were conducted to capture isolated flight<br />
conditions under varying brake and differential toggles.<br />
Longitudinal tests using nominally zero turn maneuvers<br />
under varying brakes were used to generate a map of the<br />
glide slope (lift-to-drag (L/D)) and speed characteristics<br />
of the parafoil over a prescribed flight envelope. Figures<br />
10 and 11 show some results from these tests, including<br />
a comparison with the current parafoil model. Figure 10<br />
demonstrates a clear reduction in glide slope with<br />
increasing brake setting; however, the loss in performance<br />
is very gradual up until more than 50 in of toggle.<br />
Tests have been planned to examine performance at toggles<br />
exceeding 100 in, to capture the stall characteristics<br />
of the parafoil, but at the limits tested thus far, no collapse<br />
of canopy cells is evident from video or flight data.<br />
Error bars on the L/D flight data are particularly large<br />
because of the large noise generated by the GPS navigation<br />
package in the vertical channel. The model<br />
comparisons show that a relatively simple aerodynamics<br />
model can be used to capture most of the steady-state<br />
performance of the system. Figure 11 shows the speed<br />
envelope of the parafoil over the range of tested brake<br />
settings. The original toggle limit of 100 in (it is now 200<br />
in) resulted in nearly 20-ft/s variation in flight speed.<br />
Lift-over-Drag<br />
5.0<br />
4.5<br />
4.0<br />
3.5<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
Flight Data<br />
Model<br />
0.0<br />
0 20 40 60 80 100 120<br />
Brake Command (in)<br />
Figure 10. L/D ratio vs. brake setting.<br />
21
Sea-Level Airspeed (ft/s)<br />
0<br />
0 20 40 60 80 100 120<br />
Brake Command (in)<br />
Figure 11. Sea-level air speed vs. brake setting.<br />
Manual steady-state turn rate tests were also conducted on<br />
the Dragonfly system. Figure 12 shows the turn rate vs.<br />
differential toggle deflection. Several conclusions can be<br />
drawn from this plot: (1) maximum steady-state turn rate<br />
is ~9 deg/s at a toggle of 100 in, (2) there is a considerable<br />
amount of yaw oscillatory noise that permeates the heading<br />
data channel causing significant variance in the data, and<br />
(3) the data collected show a near linear relation between<br />
differential toggle and turn rate. The noise in the yaw<br />
channel appears to be caused by some relative motion<br />
between parafoil and payload (with the AGU that holds<br />
the GPS navigation system located close to the payload),<br />
combined with wind perturbations. Small yaw oscillations<br />
persisted throughout most flight tests and did not appear<br />
to correspond to changes in the commanded turn rate of<br />
the parafoil. These oscillations were not indicative of the<br />
highly damped dynamics of the parafoil’s general turn rate<br />
response. Earlier X-38 studies showed a nonlinearity in<br />
turn rate performance at low brake settings that resulted in<br />
nearly zero response up to almost 35-40% of the stall differential<br />
toggle limits. There is insufficient data at this time<br />
to ascertain the exact low-brake response characteristics of<br />
the Dragonfly system since we have conducted the majority<br />
of our autonomous flights near 50 in of brake to reduce<br />
risk and complexity. Future plans call for additional low<br />
brake setting turns that should help document if a turn<br />
rate deadband exists. Thus far, we have seen no evidence<br />
that the Dragonfly exhibits this nonlinear turn-rate behavior.<br />
Turn Rate (deg/s)<br />
22<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
15.0<br />
10.0<br />
5.0<br />
0.0<br />
-5.0<br />
-10.0<br />
Flight Data<br />
Model<br />
-15.0 -125 -100 -75 -50 -25 0 25 50 75 100 125<br />
Differential Toggle (in)<br />
Figure 12. Turn rate vs. differential toggle setting.<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
The variability in the observed Dragonfly turn rate at<br />
zero toggle deflection (seen in Figure 12) results from a<br />
combination of some of the effects just noted (rotation of<br />
the payload relative to the canopy and wind-driven<br />
canopy rotation) as well as some asymmetric control<br />
response. The asymmetric control effect varied from<br />
flight to flight, possibly indicating some flight-specific<br />
rigging effects and/or individual control motor response<br />
variations.<br />
Flight data analysis also resulted in some changes to the<br />
toggle actuator model that included the effects of<br />
aeroloading. Data collected on commanded line toggle<br />
position vs. actual position indicated some response<br />
degradation at higher brake settings, indicative of a falloff<br />
in motor response. Revised torque limits and a more<br />
elaborate line acceleration sensitivity to load were added<br />
to the general actuator model to represent this behavior.<br />
The response characteristics of the first-generation AGU<br />
motor also resulted in a redesign of the toggle actuator<br />
models to increase torque limits on the lines. Insufficient<br />
data were collected on these motors from flight tests;<br />
however, updates to the simulated actuator models has<br />
already been completed based on manufacturer specifications.<br />
In general, the torque limits of the motors did not<br />
impact system response except during flare maneuvers.<br />
Some small modifications to the lateral model parameters<br />
were undertaken following test flights. In general, the<br />
focus was on ensuring proper steady-state response characteristics,<br />
but in addition, some small changes were<br />
made to ensure that the turn rate characteristics matched<br />
flight data, which showed a time of ~10 to 12 s from the<br />
commanded differential toggle until steady-state turn<br />
rate was reached (with the indicated time including actuator<br />
response).<br />
GN&C Flight Test Performance and Associated Design<br />
Evolution<br />
Table 1 gives a summary of the Dragonfly flight tests to date<br />
that have provided canopy dynamics data in response to<br />
scripted manual remote control (R/C) inputs or that enabled<br />
autonomous GN&C flight with onboard logging of system<br />
performance data. Other flight tests that experienced prototype<br />
avionics or canopy-related hardware failures that<br />
precluded GN&C-related testing or that inhibited essential<br />
data logging have not been included in the table. The five listed<br />
flight tests in March-April 2004 were performed under<br />
R/C to support system identification objectives. The first tests<br />
of the autonomous GN&C occurred in May 2004. Post-flight<br />
analysis revealed that large misses in these initial autonomous<br />
GN&C tests were due, at least in part, to unintended limiting<br />
of toggle commands by the flight software. This software error<br />
was corrected before the next test cycle in August 2004.<br />
Some of the initial flights in August experienced read/write<br />
failure of the flash memory chip used for in-flight data<br />
logging and for storage of the large data table applied by<br />
the lookup terminal guidance algorithm. Consequently, to
Table 1. Initial GN&C-Related Dragonfly Flight Test Results.<br />
Date Control Drop Drop Release GRW Miss Comments<br />
State Aircraft Speed Altitude (lb) Dist (m)<br />
(KIAS) (ft MSL)<br />
3/1/04 R/C C-123K 110 9K 8K n/a Good flight characterization data<br />
3/2/04 R/C C-123K 110 9K 8K n/a Good flight characterization data<br />
3/4/04 R/C C-123K 110 9K 8K n/a Good flight characterization data<br />
3/5/04 R/C C-123K 110 9K 8K n/a Good flight characterization data; hard landing but no damage<br />
4/21/04 R/C C-123K 110 12K 8K n/a Good flight characterization data<br />
5/20/04 Auto C-123K 110 12K 8K ~400 Excellent deployment and autoflight; very good landing<br />
5/21/04 Auto C-123K 110 12K 8K ~1000 Excellent deployment and autoflight; very good landing<br />
8/12/04 Auto C-123K 110 12K 8K 183 Guidance table mode disabled; very soft landing<br />
12/8/04 Auto C-130A 130 10K 8K 142 Extended drogue descent; good canopy opening; good energy<br />
management and final approach/flare<br />
12/10/04 Auto C-130A 130 10K 10K 170 Extended drogue; good canopy opening; auto, high offset,<br />
navigation to target; good final approach/flare<br />
2/2/05 R/C, Auto C-130H 130 14K 8K 256 Extended drogue phase; good canopy opening; auto, high<br />
offset, navigation to target; good final approach/flare<br />
avoid additional flash memory problems during that flight<br />
test cycle, the lookup algorithm was disabled for the flight<br />
on August 12, and the system was allowed to fly to the<br />
ground using the energy management technique (circles<br />
with adjustable radius, which was the baseline at the time,<br />
rather than figure-eights). This was moderately successful.<br />
Despite a series of deployment and avionics malfunctions<br />
in October and December 2004, two drops on December<br />
8 and 10 enabled reasonably successful GN&C tests; the<br />
system deployed well and flew autonomously to within<br />
140 m and 170 m, respectively, of the target.<br />
The February 2005 test series was the first to use Wamore’s<br />
second-generation AGU. Of the two drops performed in this<br />
series, on February 2, autonomous GN&C usage was<br />
enabled on one. While the apparent GN&C performance on<br />
this flight was somewhat poorer than anticipated, that deficiency<br />
was later traced to a communication glitch with the<br />
servo motors in the new AGU, which has since been rectified.<br />
Next Design and Development Steps<br />
Given the generic, modular architecture of the<br />
autonomous GN&C software, its adaptation to fly other<br />
parafoils will be straightforward. This year, flight characterization<br />
of two subscale models of 30,000-lb (30 klb)<br />
capable parafoils are planned, followed by adaptation of<br />
the existing GN&C software and then autonomous flight<br />
tests of these canopies. Following this, it is expected that<br />
the software will be adapted to fly the full-size 30-klb<br />
systems when they are developed.<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
Through the Small Business Innovative Research (SBIR)<br />
program, the Natick Soldier Center (NSC) is developing<br />
two navigation sensors to enhance the landing precision of<br />
guided airdrop systems. The Dragonfly parafoil with the<br />
GN&C flight software described herein will be the initial<br />
flight test vehicle for these sensors. The most mature of<br />
these sensors, in the middle of Phase II at this time, is a<br />
precision ground-relative altitude sensor under development<br />
by Creare, Inc. of Hanover, NH. Utilizing primarily<br />
sound detection and ranging (SODAR) to detect the<br />
ground over varied terrain, including through foliage, this<br />
sensor will provide height precision of ±1 ft during the<br />
terminal phase of the flight, allowing GN&C to time the<br />
final braking maneuver precisely, thereby increasing landing<br />
accuracy. This sensor will be small and lightweight (less<br />
than 5 lb) with a recurring cost of less than $500 per unit,<br />
cheap enough to be installed in just about any airdrop<br />
system from the 2000-lb class and up. Testing of this sensor<br />
will take place this calendar year.<br />
Wind uncertainty remains a significant error source for airdrop<br />
systems despite the improvements provided by the<br />
PADS mission planner discussed above. For guided systems,<br />
which have varying degrees of wind penetration capability,<br />
the wind uncertainty at lower altitudes when there is less<br />
time for correction is a significant problem. Also under<br />
development, nearing completion of Phase I SBIR trade<br />
studies, are several competing light detection and ranging<br />
(LIDAR) wind sensors that will provide real-time wind<br />
23
speed and direction to the GN&C software along the sensor<br />
line of sight ahead of the vehicle. This will allow GN&C to<br />
refine its onboard wind estimate during final maneuvers,<br />
further improving overall system landing performance.<br />
Initial tests of one of these units should take place in about<br />
1 year.<br />
CONCLUSIONS<br />
A GN&C system that enables autonomous precision payload<br />
delivery using the Dragonfly 10,000-lb-payload-class<br />
parafoil has completed prototype development and has<br />
undergone initial flight testing. The guidance algorithm<br />
uses a proportional scheme for initial homing to the target,<br />
S-turns for energy management near the target, and a table<br />
lookup implementation of optimal terminal control for<br />
final approach. A terminal flare maneuver capability is provided<br />
for landing. The control algorithm is a proportional,<br />
integral, derivative design to account for control actuator<br />
deflection constraints. Navigation relies on a coupled pair<br />
of GPS receivers with two antennae to determine position,<br />
velocity, and heading. Wind velocity is also estimated inflight<br />
from the navigation data for use by the guidance<br />
algorithm. A Dragonfly mission planning capability has<br />
been integrated into the laptop-PC-based PADS that is<br />
used onboard the Dragonfly’s carrier aircraft to determine<br />
the desired aerial release point as well as to wirelessly<br />
transmit the mission plan file to the Dragonfly, including<br />
the best current estimate of the expected winds near the<br />
drop zone during descent. Flight testing of the Dragonfly<br />
has included system identification tests, the results of<br />
which have been analyzed and factored into the dynamics<br />
models used in the GN&C algorithm design. Autonomous<br />
GN&C flight tests have already demonstrated a delivery<br />
accuracy capability of about 200 m despite a variety of<br />
developmental problems with the prototype canopy,<br />
avionics, and actuation systems that have been experienced<br />
to date. Assessment of the simulation and flight<br />
test results suggest that significant improvement in the<br />
payload delivery accuracy will be realized once the canopy<br />
and actuator dynamics are more fully characterized, and<br />
the avionics/actuator developmental problems experienced<br />
to date are overcome by design refinements and/or<br />
component upgrades.<br />
ACKNOWLEDGMENTS<br />
The autonomous GN&C software described in this paper<br />
is one part of the Dragonfly program, developed through<br />
a team effort of many individuals. The tireless efforts of the<br />
rest of the contractor team, ParaFlite, Wamore, and<br />
RoboTek, provided the vehicle to be flown. Test support<br />
by C-123 pilot Jim Blumenthal of Kingman, Arizona, and<br />
his ground support team got us through the early months<br />
24<br />
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
of the program. We would also like to thank the large team<br />
of professional system testers at the U.S. Army Yuma<br />
Proving Grounds who helped bring the system so much<br />
closer to military utility.<br />
The authors gratefully acknowledge the funding support<br />
of Joint Forces Command JPADS ACTD, the U.S. Army<br />
30K Science and Technology Objective, and the Air Force<br />
Air Mobility Command.<br />
The material in this paper is based on work supported by<br />
the U.S. Army Natick Soldier Center under contract Nos.<br />
W9124R-04-C-0154, -0144, and -0118. Any opinions,<br />
findings, and conclusions or recommendations expressed<br />
in this material are those of the authors and do not necessarily<br />
reflect the views of the Natick Soldier Center.<br />
REFERENCES<br />
[1] Hattis, P.D. and R. Benney, “Demonstration of Precision Guided<br />
Ram-Air Parafoil Airdrop Using GPS/INS Navigation,” presented<br />
at the 52 nd Institute of Navigation Annual Meeting, Cambridge,<br />
Massachusetts, June 18-20, 1996.<br />
[2] Hattis, P., B. Appleby, T. Fill, and R. Benney, “Precision Guided<br />
Airdrop System Flight Test Results,” AIAA paper 97-1468 presented<br />
at the 14th AIAA Aerodynamic Decelerator Systems<br />
Conference, June 3-5, 1997.<br />
[3] Madsen, C.M. and C.J. Cerimele, “Updated Flight Performance<br />
and Aerodynamics from a Large Scale Parafoil Test Program,”<br />
presented at the AIAA Modeling and Simulation Conference, CP<br />
2000-4311, Denver, CO, August 14-17, 2000.<br />
[4] Madsen, C.M. and C.J. Cerimele, “Flight Performance,<br />
Aerodynamics, and Simulation Development for the X-38<br />
Parafoil Test Program,” presented at the AIAA Aerodynamic<br />
Decelerator Systems Technology Conference, CP 2003-2108,<br />
Monterey, CA, May 19-22, 2003.<br />
[5] Barrows, T., “Apparent Mass of Parafoils with Spanwise Camber,”<br />
presented at the AIAA Aerodynamic Decelerator Systems<br />
Technology Conference, CP 2001-2006, Boston, MA, May 22-<br />
24, 2001.<br />
[6] Hattis, P., T. Fill, D. Rubenstein, R. Wright, and R. Benney, “An<br />
Advanced Onboard Airdrop Planner to Facilitate Precision<br />
Payload Delivery,” presented at the AIAA Guidance, Navigation,<br />
and Control Conference, CP 2000-4307, Denver, Colorado,<br />
August 14-17, 2000.<br />
[7] Hattis, P., T. Fill, D. Rubenstein, R. Wright, R. Benney, and D.<br />
LeMoine, “Status of an Onboard PC-Based Airdrop Planner<br />
Demonstration,” presented at the AIAA Aerodynamic<br />
Decelerator Systems Conference, CP 2001-2066, Boston<br />
Massachusetts, May 22-24, 2001.<br />
[8] Hattis, P., K. Angermueller, T. Fill, R. Wright, R. Benney, and D.<br />
LeMoine, “An In-Flight Precision Airdrop Planning System,” presented<br />
at the 23nd Army Science Conference, Orlando, Florida,<br />
December 2-5, 2002.<br />
[9] Hattis, P., K. Angermueller, T. Fill, R. Wright, R. Benney, D.<br />
LeMoine, and D. King, “In-Flight Precision Airdrop Planner<br />
Follow-On Development Program,” presented at the AIAA<br />
Aerodynamic Decelerator Systems Conference, CP 2003-2141,<br />
Monterey, California, May 19-22, 2003.
Autonomous Guidance, Navigation, and Control of Large Parafoils<br />
(l-r) Leena Singh, Phil Hattis,<br />
David Carter and Sean George<br />
David Carter is a Principal Member of the Technical Staff. His interests are algorithm development and<br />
design and specification in the areas of guidance, control, and estimation theory. Before joining <strong>Draper</strong>, he<br />
was an Associate Professor of Mathematics at the University of Virginia. Dr. Carter holds a BA in Physics<br />
from Haverford College, an MSE in Electrical Engineering from Princeton, and a PhD in Mathematics from<br />
Columbia University.<br />
Sean George is a Senior Member Technical Staff in the Vehicle Systems Group. He has over 8 years experience<br />
at <strong>Draper</strong> <strong>Laboratory</strong> as an aerodynamicist on engineering projects relating to the analysis, simulation,<br />
and design of a wide range of vehicle platforms, including projectiles, rotorcraft, aerodecelerators,<br />
and airplanes. He has experience with a range of computational modeling tools, including Navier-Stokes<br />
and potential flow solvers. He was chief configuration design engineer for the Wide Area Surveillance<br />
Projectile (WASP) gun-launched unmanned aerial vehicles (UAVs) as well as several folded-flyer variants<br />
proposed for alternative missions. He has also provided analysis support for a number of <strong>Draper</strong> internal<br />
and external rotorcraft projects. His principal role on <strong>Draper</strong>’s airdrop programs is in the areas of vehicle<br />
modeling, flight data analysis, and simulation development for both ballistic and guided parachute systems.<br />
Mr. George holds a BS in Applied Physics from Harvey Mudd College (1996) and an SM in<br />
Aeronautics and Astronautics from MIT (1998).<br />
Philip Hattis has been employed at <strong>Draper</strong> <strong>Laboratory</strong> since 1974, and currently holds the position of<br />
<strong>Laboratory</strong> Technical Staff in the Mission Design and Analysis Group of the GN&C Systems Division.<br />
Responsibilities have included technical leadership for projects involving launch and space vehicle GN&C<br />
system development, precision delivery airdrop systems, precision Mars landing systems, ballistic missile<br />
defense systems, ground warrior systems, helicopter fire control systems, autonomous aerial vehicles,<br />
autonomous space systems, and advanced satellite navigation systems. He now serves as Technical Lead<br />
for the Crew Exploration Program (CEV) GN&C system development as well as Technical Director for precision<br />
airdrop programs and for missile defense programs. Dr. Hattis is a Fellow in the American Institute<br />
of Aeronautics and Astronautics (AIAA). He has a BS in Mechanical Engineering from Northwestern<br />
University, an MS in Aeronautics from Caltech, and a PhD in Aeronautics and Astronautics from MIT.<br />
Leena Singh is a Senior Member of the Technical Staff in the Autonomous Control Group. Dr. Singh has<br />
9 years of research experience exploring new autonomous GN&C algorithms for systems, which have<br />
included UAVs, helicopter fire control systems, and aircraft engines. Her recent research focuses on envelope-extending<br />
advanced autonomous guidance and control algorithms. Dr. Singh received a BS in Physics<br />
and Mathematics from Mt. Holyoke College, an MS from Rensselaer Polytechnic Institute (1993), and a<br />
PhD from MIT (1997).<br />
Steven Tavan is the Precision Airdrop GN&C Research Leader at the U.S. Army Natick Soldier Center,<br />
Airdrop/Aerial Delivery Directorate, Natick, MA. For the last 3 years, he has been the government sponsor<br />
of <strong>Draper</strong> <strong>Laboratory</strong>’s activities in airdrop mission planning and autonomous guidance, navigation,<br />
and control of parafoils. Prior to his government service, Mr. Tavan worked on aerospace projects at the<br />
MITRE Corporation and <strong>Draper</strong> <strong>Laboratory</strong>. He has a BS in Earth Sciences from MIT.<br />
25
An Ultra-Low-Power Physics Package for a Chip-Scale<br />
Atomic Clock<br />
Mark Mescher, Robert Lutwak, and Mathew Varghese<br />
Copyright © 2005 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc. Presented at Institute of Electrical and Electronics<br />
Engineers (IEEE) Transducers ’05, 13 th International Conference on Solid-State Sensors, Actuators, and Microsystems,<br />
Seoul, Korea, June 5-9, 2005<br />
INTRODUCTION<br />
26<br />
We report the design and measured thermal and mechanical performance of an ultra-lowpower<br />
physics package for a chip-scale atomic clock (CSAC). This physics package will enable<br />
communications and navigation systems that require a compact, low-power atomic frequency standard.<br />
The physics package includes a unique combination of thermal isolation, mechanical stability and<br />
robustness, and small package volume. We have demonstrated temperature control at a nominal<br />
operating temperature of 75°C in a room-temperature, vacuum ambient requiring only 7 mW of heating<br />
power. This represents a power reduction of over two orders of magnitude compared with the lowestpower<br />
existing commercial technology [1] and more than an order of magnitude improvement over<br />
other CSAC development efforts. [2]<br />
Atomic clocks play an essential role in the timing and synchronization<br />
of modern communications and navigation<br />
systems. To date, however, their relatively large size and<br />
power consumption (125 cm3 and 6 W) have prevented<br />
their application in portable devices. We are developing a<br />
chip-scale atomic clock, with size
opposite side of the cell back to a photodiode that is integrated<br />
on the VCSEL die. The VCSEL is tuned to the D1<br />
optical transition of cesium at λ = 894 nm and directly<br />
modulated at νHF/2 = 4.6 GHz, where νHF is the hyperfine<br />
transition frequency of cesium. The cell and VCSEL are<br />
temperature-stabilized at approximately 75°C to optimize<br />
clock performance. [4]<br />
The cell and VCSEL are supported by a novel suspension<br />
system that relies on the strength and low thermal conductivity<br />
of polyimide to minimize conductive thermal losses<br />
from the heated resonance cell while providing mechanical<br />
stability and isolation from external vibration. The suspension<br />
structure consists of upper and lower halves (Figure 1).<br />
Mirror<br />
6 mm<br />
Heater/Sensor Elements<br />
Vapor Cell<br />
VCSEL/PD<br />
Figure 1. Left: cross-section view of physics package without<br />
leadless chip carrier (LCC); right: exploded view.<br />
Each half consists of a silicon frame that supports multiple<br />
polyimide tethers converging on a central attachment<br />
plate. During assembly, the two central attachment plates<br />
mount to opposite sides of the vapor cell, while the two<br />
frames are mounted on opposite sides of a spacer. The<br />
spacer and the cell thickness are such that the tethers are<br />
stretched out of plane in the assembly process. The introduction<br />
of a small angle in the tethers greatly stiffens the<br />
suspension structure by eliminating low-frequency bending<br />
modes in the suspension beams; thus, z-axis motion of<br />
the cell causes the tethers to stretch rather than bend.<br />
The superior thermal isolation and mechanical stiffness<br />
provided by the suspension structure is made possible<br />
through appropriate material choice and packaging techniques.<br />
Polyimide has extremely low thermal conductivity<br />
(0.2 W/m°C) that minimizes conduction heat loss and permits<br />
compact suspension designs by reducing required<br />
beam lengths. In addition, similar to other polymers, it has<br />
a high yield strain (3%), which enables the angled suspension<br />
system to be built from components that are<br />
fabricated using planar, batch-fabrication processes. In<br />
addition, the electrical leads to the heated components<br />
may be directly patterned onto the polyimide.<br />
Fabrication<br />
Cell Fabrication<br />
The 2-mm 3 cesium resonance cell is fabricated of silicon<br />
with anodically-bonded Pyrex ® windows. Details of the<br />
cell fabrication and cesium loading have been described<br />
previously. [5] A quarter-wave plate necessary for circular<br />
An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock<br />
Frame<br />
Spacer<br />
Vapor<br />
Cell<br />
Thermal Control Suspension<br />
VCSEL<br />
Suspension<br />
VCSEL/PD<br />
polarization is attached to the tablet after cesium filling but<br />
prior to dicing (Figure 2).<br />
Figure 2. Left: cross-section model of cesium vapor cell<br />
including wave plate; right: 4 x 4 tablet (1 cm 2 ) of<br />
silicon cells before cesium loading and tablet dicing.<br />
Suspension Fabrication<br />
The upper (thermal control) and lower (VCSEL/photodiode<br />
(PD)) portions of the suspension structure are<br />
fabricated similarly on separate silicon wafers. The frame<br />
spacer is conventionally machined from aluminum. The<br />
patterned polyimide layers that form the suspension tethers<br />
are identical except for a 1.5-mm hole in the center<br />
plate of the VCSEL suspension that allows optical transmission<br />
of the laser source and collected light. The process<br />
sequences are identical except for the metallization. These are<br />
outlined in Table 1. Figure 3 shows a fabricated assembly.<br />
Table 1. Suspension Structure Fabrication Sequence.<br />
Grow etch stop for backside silicon etch:<br />
thermal SiO2 (1.0 µm)<br />
Spin on suspension structural material:<br />
photodefined polyimide (5 µm)<br />
Sputter thermal control suspension metal:<br />
titanium (Ti)/platinum (Pt) (0.03 µm/0.25 µm)<br />
Sputter VCSEL/PD suspension metal:<br />
Ti/Pt/gold (Au)/Ti (0.03 µm/0.4 µm/0.4 µm/0.1 µm)<br />
Sputter bond pad metal:<br />
Ti/Au (0.03 µm/0.5 µm)<br />
Etch silicon (Si) deep reactive ion etching (DRIE) for suspension<br />
release:<br />
through wafer (500 µm)<br />
Plasma-etch thermal oxide etch-stop layer<br />
Figure 3. Fabricated physics package assembly mounted<br />
in an LCC (thermal control side facing up).<br />
27
Assembly and Packaging<br />
The cell and suspension structures are assembled with epoxy.<br />
EPO-TEK ® 353ND is used for its optical transmission characteristics<br />
and its low outgassing properties. An exploded<br />
view before assembly is depicted in Figure 1. Individual cells<br />
are mounted in the suspension structure. The thermal control<br />
suspension structure is placed face-down in an<br />
alignment fixture. The cell is mounted mirror-side down on<br />
the central attachment plate of the suspension via manual<br />
dispensing of epoxy. After curing, the adhesive interface<br />
thickness is 5 µm. The aluminum frame spacer is then<br />
aligned via fiducial marks and adhered to the silicon frame<br />
similarly. Finally, the VCSEL/PD suspension is aligned using<br />
another fiducial mark on the frame. Epoxy attaches the center<br />
plate to the cell and the frame to the frame spacer.<br />
The VCSEL/PD die is attached to the cell via solder reflow.<br />
First, 0.018-in solder balls are attached to the VCSEL/PD<br />
pads. The VCSEL/PD solder balls are aligned to pad locations<br />
on the center plate of the VCSEL/PD suspension and<br />
attached by reflowing the solder. Simultaneously, solder balls<br />
are attached to pads on the frame of the VCSEL/PD suspension<br />
to enable mounting of the frame assembly to a ceramic<br />
LCC. The LCC is then aligned and a final reflow is done to<br />
connect the frame pads to those of the LCC. The heater and<br />
temperature sensor are connected via wire bond to corner<br />
pads in the LCC. Finally, an alumina lid containing an activated<br />
getter is attached in vacuum. A solder preform is<br />
mounted onto the seal-ring of the LCC and reflowed to seal<br />
the device. Including the vacuum package, the overall size<br />
of the physics package is approximately 0.6 cm3 .<br />
Performance Characteristics<br />
Thermal Control<br />
The cell temperature is maintained through integrated single-element<br />
resistive platinum temperature-sensing and<br />
heating elements, both of which are patterned directly onto<br />
the central polyimide plate of the thermal control side of the<br />
suspension structure. Measured temperature coefficients of<br />
resistance for the sputtered films are 2300 ppm/°C with a<br />
4% variation across a 4-in wafer. The temperature sensor is<br />
distributed uniformly across the cell face to provide an accurate<br />
average temperature measurement of the vapor cell.<br />
The portion of the trace that runs along the suspension tether<br />
from the center plate to outer frame bond pad is designed<br />
to have relatively low resistance (0.9% in current devices) to<br />
prevent partial sensing of the frame (i.e, ambient) temperature.<br />
The sense resistor is sized (10 kΩ) to achieve<br />
approximately 0.1°C temperature sensitivity with lowpower<br />
readout electronics. The heater resistance is<br />
distributed around the periphery of the cell rather than uniformly<br />
across the plate surface. This provides a more<br />
uniform temperature because most of the heat transfer away<br />
from this surface is through the silicon walls of the cell to all<br />
cell faces, which then radiate heat to the LCC package. The<br />
heater resistance is sized (400 Ω) such that power is<br />
delivered efficiently from the control electronics (3-V supply)<br />
while still maintaining sufficient voltage overhead to<br />
28<br />
An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock<br />
provide fast device turn-on and good control response. In<br />
the case of the heater, the tether traces contribute to inefficiency<br />
by heating the tethers and the frame (7.5% in current<br />
devices). Both the heating and sensing resistors are configured<br />
such that current flowing through one segment of the<br />
resistor is balanced by a current in another segment in close<br />
proximity that is flowing in the opposite direction, which<br />
minimizes magnetic fields in the vicinity of the vapor cell.<br />
Heat loss to the frame and package occurs through radiation,<br />
gas conduction and convection, and conduction in<br />
the suspension tethers. Gaseous conduction and convection<br />
are eliminated by vacuum packaging the device.<br />
However, even at atmospheric pressure, convection is suppressed<br />
due to the small size of the gap between the heated<br />
device and the LCC package. Radiation is driven by surface<br />
area and emissivity. Gas conduction is driven by gap<br />
dimensions, gas composition, and pressure. The bulk conduction<br />
is determined by thermal conductivity and the<br />
shape and size of the suspension tethers and their metal<br />
interconnect. Detailed radiation models must account for<br />
the multiple surface emissivities and shape factors of the<br />
system. However, the heated cesium cell will have radiative<br />
heat loss with a temperature dependence given by<br />
P = ε . σ . Acell . (Tcell 4 – Tamb 4 ) (1)<br />
where ε is a parameter that embodies the relative emissivities<br />
of the cell and heat-reflecting package, Acell is the surface<br />
area of the cell, and σ is the Stefan-Boltzmann constant. As<br />
a general rule, heat loss by thermal radiation is reduced by<br />
minimizing the cell’s surface area and average emissivity. The<br />
cell has four saw-diced silicon surfaces, one polyimide-andplatinum-covered<br />
surface, and one surface of gallium<br />
arsenide (the VCSEL/PD). These materials all have emissivities<br />
around 0.5, with variation depending on surface finish.<br />
Some materials, such as polished aluminum, can have emissivities<br />
as low as 0.03. While we have not yet implemented<br />
surface coatings on the cell sidewalls to reduce radiation, we<br />
have conducted experiments with test aluminum cells in<br />
place of the standard cell that indicate that even lower heater<br />
power requirements are possible (Figure 4).<br />
12<br />
Power (mW)<br />
10<br />
8<br />
6<br />
4<br />
2<br />
Silicon/Pyrex Cell<br />
AI Cell<br />
0<br />
0 30 60 90 120<br />
Temperature (˚C)<br />
Figure 4. Measured steady-state heater power as a function<br />
of cell temperature in a vacuum ambient (24°C,<br />
2 mTorr) for a silicon/Pyrex cell and a low-emissivity<br />
polished aluminum test cell.
Heat loss due to gas conduction was characterized to determine<br />
how package vacuum level affects required heater<br />
power. Figure 5 shows heater power for a cell held at 80°C<br />
in 24°C ambient as a function of gas (air) pressure. For low<br />
pressures (less than about 20 mTorr), the power is dominated<br />
by radiation. For intermediate pressures, heat transfer is<br />
molecular and is proportional to pressure. At high pressures,<br />
the heat transfer is essentially independent of pressure<br />
and depends on the thermal conductivity of the gas.<br />
Power (mW)<br />
80<br />
60<br />
40<br />
20<br />
0<br />
0.001 0.1 1.0 1000<br />
Pressure (Torr)<br />
Figure 5. Heater power required to maintain cell temperature<br />
at 80°C in a 24°C ambient of varying<br />
pressure. Experiment was performed in a bell jar<br />
with the device packaged as shown in Figure 3.<br />
The solid material conduction losses in our CSAC physics<br />
package are not measurable currently and are small compared<br />
with radiation losses. The predicted power loss at<br />
75°C by conduction is 0.9 mW, or 13% of the total. The<br />
specifications and predicted thermal performance characteristics<br />
of the suspension system are shown in Table 2.<br />
Table 2. Tether System Design Values.<br />
Total tether count 16<br />
Polyimide tether 5 x 375 x 1000 µm<br />
VCSEL/PD metal widths 10 µm<br />
Heater metal trace widths 80 µm<br />
Sensor metal trace widths 30 µm<br />
Thermal Resistances:<br />
Polyimide tethers (total)* 167°C/mW<br />
Metal traces (total)† 87°C/mW<br />
Total 57°C/mW<br />
* 0.2 W/m°C polyimide conductivity assumed<br />
† bulk conductivity values assumed<br />
The heat capacity of the system was also estimated from a<br />
time constant. The step response used in the estimate is<br />
shown in Figure 6. The measured time constant was 63 s.<br />
From the steady-state measurements shown in Figure 4, the<br />
effective total thermal resistance (radiation-dominated) at<br />
75°C is approximately 6.3°C/mW, which yields a heat capacity<br />
of 0.010 J/C. This is within 5% of the calculated value.<br />
An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock<br />
Temperature (˚C)<br />
80<br />
78<br />
76<br />
74<br />
72<br />
70<br />
0 200 400<br />
Time (s)<br />
600<br />
Figure 6. Measured thermal step response of physics package.<br />
Mechanical Design<br />
As described earlier, the mechanical design takes advantage<br />
of several material properties of polyimide to meet the<br />
conflicting design requirements of low heat loss and<br />
mechanical rigidity while permitting planar MEMS batch<br />
fabrication techniques for the suspension system. We<br />
describe the design in the context of two mechanical specifications:<br />
fundamental mechanical resonance frequency<br />
and maximum sustainable steady acceleration (“g” load)<br />
without plastic deformation of the polyimide tethers.<br />
Figure 7 shows a cross section of one half of the suspension<br />
system. As is suggested by the figure, the tethers are<br />
dominated by axial rather than bending stresses. They are<br />
more appropriately thought of as cables rather than bending<br />
beams. If one considers only z-axis forces, the system<br />
is equivalent to the mass-spring model shown in Figure 7,<br />
where d is the spring’s displacement from equilibrium after<br />
assembly (both springs are nominally in tension). For large<br />
z-displacements, these springs are highly nonlinear.<br />
θ<br />
d<br />
Mcell<br />
Figure 7. Left: depiction of a cross section of half-device<br />
showing exaggerated tether angle; right: linearized<br />
mass-spring model.<br />
In order for the tethers to act as springs, they must be in<br />
tension at all times. Under this condition, the small signal<br />
resonant frequency of the suspension system is independent<br />
of the magnitude of the tensile stress. It is the angle of<br />
the tethers rather than the pre-tension that sets the resonant<br />
frequency. However, if the stress in the tethers turns<br />
compressive at any point, they buckle and do not provide<br />
any significant mechanical support. Such a condition is<br />
encountered under a high static “g” load and is described<br />
later. The restoring force of a single tether for a displacement<br />
z is<br />
+z<br />
∆z<br />
11<br />
10<br />
9<br />
8<br />
7<br />
6<br />
Heater Power (mW)<br />
29
(2)<br />
This equation is nonlinear in z, but for small motions ∆z<br />
around the equilibrium displacement, d, a linear equation<br />
for the total restoring force in the z-direction can be used<br />
FT = -N . k . ∆z (3)<br />
Here N is the total tether count and k is the linearized<br />
spring constant of a single tether<br />
(4)<br />
E is elastic modulus, A is tether cross-sectional area, L is<br />
tether length, and θ0 is the as-built tether angle (Figure 7).<br />
This spring constant greatly exceeds that of the bending<br />
stiffness for even modest angles. The resonant frequency is<br />
now given by the familiar relation<br />
It should be noted that x- and y-axis resonance frequencies<br />
can be calculated simply by replacing cos (θ0) with sin<br />
(θ0) in the equation for the spring constant. It follows from<br />
this that resonant frequencies and maximum g loads are<br />
higher in the x and y directions.<br />
Test units with a range of built-in tether angles were built<br />
and tested to validate the model. A small magnetically permeable<br />
disk was attached to the cell in these test units and<br />
excited by a miniature electromagnet. Displacement was<br />
measured with an interferometer as the excitation frequency<br />
was swept manually. Fundamental mode resonances<br />
were calculated by fitting the data to a standard massspring-damper<br />
model. Measured resonant frequency data<br />
were estimated to be accurate to better than 1%. The<br />
results are compared to the model in Figure 8. The elastic<br />
modulus assumed for the model was 3.5 GPa. The mass of<br />
the cell and attached magnetic disk was estimated at 27<br />
mg, significantly greater than the cell alone (18 mg). Thus,<br />
the resonant frequency of real devices is expected to be<br />
higher by the square root of the mass ratio of the test unit<br />
and real device (27 mg/18 mg) 0.5 for the same tether<br />
angle. This predicts a resonant frequency of 2450 Hz for a<br />
real device with a tether angle of 10 deg.<br />
In order to calculate the maximum g load of the devices,<br />
we estimate the forces present at the yield strain, εmax, of<br />
the tethers. Under steady acceleration, the maximum g<br />
load, Amax, can be expressed as<br />
30<br />
An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock<br />
(5)<br />
(6)<br />
Resonant Frequency (Hz)<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
Model<br />
Measured<br />
0<br />
0 5 10 15<br />
Tether Angle (deg)<br />
Figure 8. Measured and modeled mechanical resonance frequency<br />
as a function of as-built tether angle. The error<br />
bars are attached to the theoretical curve; they are<br />
based on estimates of fabrication tolerances, including<br />
cell mass, tether dimensions, and tether angle.<br />
We assume that the cell has been displaced such that only<br />
the upper or lower halves of the suspension system remain<br />
in tension and thus provide mechanical support. For εmax<br />
= 0.03, Amax = 2200 g (1500 grms).<br />
CONCLUSIONS<br />
We reported on the design and measured thermal and<br />
mechanical performance of an ultra-low-power physics<br />
package for a CSAC. We have demonstrated temperature<br />
control at a nominal operating temperature of 75°C in a<br />
room-temperature, vacuum ambient requiring only 7 mW<br />
of heating power. Lowering the cell’s emissivity has been<br />
demonstrated experimentally to provide further power<br />
reduction. Measured resonance frequency corresponding<br />
to 2.5 kHz for a physics package was presented. In addition,<br />
models of the suspension system indicate that the<br />
physics package can sustain maximum g loads in excess of<br />
2000 g without damage.<br />
ACKNOWLEDGMENTS<br />
The authors wish to thank Mike Garvey of the<br />
Symmetricom Technology Realization Center (TRC) and<br />
John McElroy of <strong>Draper</strong> <strong>Laboratory</strong> for supporting and<br />
directing this effort. We are also thankful for the technical<br />
and theoretical support provided by Gary Tepolt,<br />
John LeBlanc, and Amy Duwel of <strong>Draper</strong> <strong>Laboratory</strong>,<br />
Don Emmons and Peter Vlitas of the TRC, and Darwin<br />
Serkland of Sandia National Laboratories. We thank Yen-<br />
Wah Ho, Greg Romano, Ryan Prince, Maria Cardoso,<br />
Maria Holmboe, Katherine Ashton, and Bessy Silva for<br />
assistance in the fabrication and assembly of physics<br />
package components. We also thank V.M. Montano and<br />
A.T. Ongstad for assistance in the fabrication of the<br />
VCSEL/RCPD chips, and T.W. Hargett for assistance in<br />
the epitaxial semiconductor growth. This work is supported<br />
by the Defense Advanced Research Projects<br />
Agency, Contract No. NBCHC020050.
REFERENCES<br />
[1] X-72 Data Sheet, http://www.symmetricom.com/media/<br />
pdf/documents/ds-x72.pdf.<br />
[2] Kitching, J., “Chip Scale, Microfabricated Atomic Clocks (An<br />
Emerging Technology),” 29th Annual Time and Frequency<br />
Metrology Seminar, National Institute of Standards &<br />
Technology (NIST), Boulder, CO, June 14-17, 2004.<br />
[3] Lutwak, R. et al., “The Chip-Scale Atomic Clock – Coherent<br />
Population Trapping vs. Conventional Interrogation,”<br />
Proceedings of the 34th Annual Precise Time and Time Interval<br />
(PTTI) Systems and Applications Meeting, December 3-5, 2002,<br />
Reston, VA, pp. 539-550.<br />
An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock<br />
[4] Lutwak, R. et al., “The Chip-Scale Atomic Clock – Recent<br />
Development Progress,” Proceedings of the 35th Annual Precise<br />
Time and Time Interval (PTTI) Systems and Applications<br />
Meeting, December 2-4, 2003, San Diego, CA, pp. 467-478.<br />
[5] Lutwak, R., J. Deng, W. Riley, M. Varghese, M. Mescher, D.K.<br />
Serkland, G.M. Peake, “The Chip-Scale Atomic Clock – Recent<br />
Development Progress,” Proceedings of the 36th Annual Precise<br />
Time and Time Interval (PTTI) Systems and Applications<br />
Meeting, December 7-9, 2004, Washington, DC.<br />
[6] Liew, L-A. et al., “Microfabricated Alkali Atom Vapor Cells,”<br />
Applied Physics Letters, Vol. 84, 2004, p. 2694.<br />
(l-r) Mark Mescher, Mathew Varghese<br />
Mark Mescher is a Principal Member of Technical Staff in the Advanced Hardware Development Division.<br />
He won the <strong>Draper</strong> 2005 Distinguished Performance Award with fellow team members for their work on<br />
the chip-scale atomic clock. Past or current project leadership includes transcleral drug delivery development,<br />
MEMS acoustic arrays for underwater imaging, sensor systems for biological sensing applications,<br />
and microfluidic system development. Dr. Mescher received a BS in Computer Engineering from Wright<br />
State University (1993), and MS and PhD degrees in Electrical and Computer Engineering from Carnegie<br />
Mellon University (1995 and 1999, respectively).<br />
Robert Lutwak is a Senior Scientist in the Research group of the Symmetricom TRC. His areas of expertise<br />
include the engineering of atomic instrumentation, particularly the design of laser oscillators and laser<br />
systems, ultra-high vacuum and atomic beam design, and electronic and computer control of complex systems.<br />
He is currently a Principal Investigator on the DARPA-sponsored CSAC program, developing atomic<br />
clocks two orders of magnitude smaller and lower power than any existing technology. He also leads the<br />
Symmetricom Advanced Technology Atomic Frequency Standard (ATAFS) program to develop next-generation,<br />
high-performance atomic clocks for deployment onboard the GPS satellite constellation. Dr.<br />
Lutwak is a member of the American Physical Society and the IEEE Ultrasonics, Ferroelectrics, and<br />
Frequency Control (UFFC) Society. He serves on the Technical Program Committee for the IEEE<br />
Frequency Control Symposium. Dr. Lutwak received a BS in Physics, summa cum laude, from Miami<br />
University (1988) and a PhD in Atomic and Optical Physics from MIT (1997).<br />
Mathew Varghese currently heads the Microsystems Integration Group and is a Principal Member of<br />
Technical Staff at <strong>Draper</strong> <strong>Laboratory</strong>. His research interests focus on the fabrication, design, and analysis<br />
of microsystems. He has led projects to build microphones, drug delivery devices, MEMS RF filters, and<br />
CSAC. He won the 2005 <strong>Draper</strong> Distinguished Performance Award for leading the CSAC development<br />
effort at <strong>Draper</strong>. Dr. Varghese received a BS in Electrical Engineering and Computer Science with a minor<br />
in Physics from the University of California, Berkeley and SM and PhD degrees in Electrical Engineering<br />
and Computer Science from MIT (1997 and 2001, respectively).<br />
31
Detection of Biological and Chemical Agents Using<br />
Differential Mobility Spectrometry (DMS) Technology<br />
Melissa Krebs, Angela Zapata, Erkinjon Nazarov, Raanan Miller, Isaac Costa, Abraham Sonenshein, Cristina Davis<br />
Copyright © 2005 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc. Printed with permission in IEEE Sensors Journal, Vol. 5, No. 4, August<br />
2005, pp. 696-703<br />
INTRODUCTION<br />
With international concern growing over the potential for chemical and biological terrorism,<br />
there is an urgent need for a sensor that can detect chemical and biological agents quickly and accurately.<br />
Such a sensor must be portable, robust, and sensitive, with fast sample analysis time. We will<br />
demonstrate the use of a micromachined differential mobility spectrometer (DMS) with all these characteristics<br />
that is able to detect multiple agents simultaneously on a time scale of seconds. In this study,<br />
we have demonstrated the ability of the DMS to detect Bacillus subtilis spores, a surrogate for Bacillus<br />
anthracis spores, the causative agent of anthrax. Pyrolysis was used as the sample introduction method<br />
to volatilize the spores before introducing material into the DMS. In addition, we examined the effect<br />
of pyrolysis on B. subtilis spores suspended in sterile water using SDS-PAGE. These experiments<br />
showed that the spores must be heated at 650°C or greater for 5 s or at 550°C for at least 10 s to be<br />
fragmented into particles considerably smaller than 10 kilodaltons (kDa), which the DMS is able to<br />
detect. Several major biomarkers can be distinguished easily above the background of the sterile water<br />
in which the spores are suspended, and we hypothesize that additional biomarkers could be liberated<br />
by further optimizing conditions. The DMS also has shown promise as a detector for chemical weapon<br />
agents, and we have also demonstrated the ability of the DMS to detect nerve and blister agent simulants<br />
at clinically relevant levels.<br />
The use of microbes for bioterrorism is well documented.<br />
As long ago as the Middle Ages, warriors catapulted<br />
bodies of plague victims over the walls of villages under<br />
siege in an attempt to hasten their victory. [1] In the 18 th<br />
century, blankets that had been used by smallpox victims<br />
were intentionally distributed to the Native American<br />
population. [1] Within the last century, there are many<br />
more examples of the use of biological weapons. Japan<br />
32<br />
conducted biological warfare experiments in the 1930s<br />
and 1940s, to which outbreaks of plague, cholera, and<br />
typhus were credited. [2] In 1979, there was an epidemic<br />
of inhalation anthrax in Sverdlovsk resulting from a<br />
release of weapons-grade B. anthracis from a military<br />
facility. [3] In 1993, the Aum Shinrikyo cult attempted to<br />
disperse anthrax spores from a high-rise building in<br />
Tokyo. [4] There have also been instances of the intentional
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
release of biological agents in the United States. In 1985,<br />
the Bhagwan Shree Rajneesh cult used Salmonella to<br />
infect 752 patients in Oregon. [5] In 2001, the United<br />
States experienced the first intentional anthrax release in<br />
this country. [6] Bacillus anthracis has the ability to form<br />
resilient spores, making it a prime candidate for a biological<br />
weapon; the spores are highly resistant to extreme<br />
conditions and can be dispersed easily. [7],[8]<br />
The use of chemical agents as weapons of mass destruction<br />
has also been well documented. Chemical weapons<br />
include nerve agents, blister agents, asphyxiants, and<br />
pulmonary irritants. [9] In 1936, Japan used chemical<br />
weapons (hydrogen cyanide, mustard gas, phosgene)<br />
during their invasion of China. [10] Between 1980-1988,<br />
Iraq used mustard gas and nerve agents against Iran and<br />
Iraqi Kurds. [10] In 1995, the Aum Shinrikyo cult released<br />
sarin, a toxic nerve agent, in the subway system in<br />
Tokyo. [4]<br />
The development of early detection technologies that are<br />
able to identify both biological and chemical warfare<br />
agents quickly is important to homeland defense agencies,<br />
national health care systems, and first responders.<br />
Such a technology should be robust, field-deployable,<br />
and inexpensive. It should yield fast and sensitive identification<br />
of an unknown sample on location, and it should<br />
function autonomously. [11]<br />
Scientists have adapted molecular biology techniques to<br />
detect B. anthracis using DNA-based, antibody-based,<br />
and mass spectrometry analysis approaches. These tests<br />
vary greatly in sensitivity, response time, cost, availability,<br />
and complexity of use. Polymerase chain reaction<br />
(PCR) and other nucleic acid-based methods have been<br />
widely examined for the detection of anthrax. [12]-[22]<br />
Several antibody-based methods have also been examined<br />
for anthrax spore recognition. [23]-[29]<br />
Currently, several chemical detectors are being investigated<br />
to rapidly identify Bacillus spores. Virtually all gas<br />
chromatograph (GC) detectors, such as the widely used<br />
flame ionization detector (FID), produce a signal indicating<br />
the presence of a compound eluted from the column;<br />
however, they lack the specific information required for<br />
unambiguous compound identification. An expedient<br />
and simple method to identify unknown analytes<br />
requires a detector to provide an orthogonal set of information<br />
for each chromatographic peak. The mass<br />
spectrometer (MS) is generally considered one of the<br />
most definitive detectors for compound identification, as<br />
it generates a fingerprint pattern of fragment ions for<br />
each GC elutant. Mass spectrometric information is often<br />
sufficient for sample identification through comparison<br />
to compound libraries, and has been used to identify<br />
some Bacillus species. [30]-[34]<br />
While GCs are continuously being miniaturized and<br />
reduced in cost, [35] mass spectrometers are still very<br />
expensive, between $50,000 and $75,000. They are relatively<br />
large, making field deployment difficult. They<br />
also require operation at low pressures and their spectra<br />
can be difficult to interpret. An alternative miniaturized<br />
technology developed at <strong>Draper</strong> <strong>Laboratory</strong>, known as<br />
the differential mobility spectrometer (DMS), is available<br />
as a hand-held mobile unit that operates at ambient temperature<br />
and pressure. Our micromachined DMS sensor<br />
is capable of rapid, sensitive detection of many chemicals.<br />
[36]-[39] For example, it has been shown to<br />
distinguish o-, p-, and m-xylene isomers; [40] these isomers<br />
cannot be separated and distinguished using<br />
traditional chemical analysis techniques (such as GC-<br />
MS). We tested the ability of the DMS to detect simulants<br />
of chemical and biological weapons agents. We have previously<br />
shown its capability to distinguish dipicolinic<br />
acid, picolinic acid, and pyridine – three chemical components<br />
present in high concentrations in spores as<br />
verified by GC-MS. [41] Here, we show evidence of spectrum-wide<br />
biological markers that distinguish pyrolyzed<br />
spores from background materials, and the ability to<br />
detect chemical weapons agent simulants.<br />
Material and Methods<br />
DMS Operating Principles<br />
<strong>Draper</strong> <strong>Laboratory</strong> developed the micromachined DMS<br />
as a novel Microelectromechanical System (MEMS) sensor<br />
for biological and chemical agent detection. When<br />
incorporated into a detection system, the DMS provides<br />
quantitative, sensitive detection down to the parts-pertrillion<br />
(ppt) range. [40] It offers the advantage of<br />
operating in air at atmospheric pressure and ambient<br />
temperature. It filters ions based on their nonlinear<br />
mobility in high-strength RF electric fields. The miniaturized<br />
DMS has enhanced sensitivity and resolution<br />
due to the close spacing of the RF plates in the drift<br />
tube region. These features make the device suitable for<br />
use as a quantitative, relatively low-cost, portable<br />
detector.<br />
DMS operation measures the differential mobility of<br />
ions as they travel through the air in response to an<br />
applied electric field, as shown in Figure 1. Differential<br />
mobility is dependent on the charge, size, and mass of<br />
an ion as it experiences high and low electric fields. A<br />
solid or liquid sample is first pyrolyzed, breaking apart<br />
the chemical bonds by thermal energy, and converting<br />
the sample to a gaseous form of smaller and more<br />
volatile fragments. These gaseous sample fragments are<br />
then introduced into the spectrometer where they are<br />
ionized by either a radioactive source or ultraviolet<br />
(UV) radiation. This paper presents data using a 63Ni<br />
33
ionization source. Once ionized, the sample is transported<br />
by a carrier gas through an ion filter toward the<br />
detecting electrodes (Faraday plates). The electric field<br />
that is applied between the parallel ion filter electrodes<br />
can be adjusted to tune the ion filter. Two fields are<br />
applied across the parallel plates: an asymmetric waveform<br />
electric field alternating between high- and<br />
low-strength fields, and a low-strength DC compensation<br />
voltage. The asymmetric field amplitude is held<br />
constant while the compensation voltage levels are<br />
adjusted to allow a particular ion species to pass<br />
through the ion filter. Once the ion species passes<br />
through the ion filter, it is detected upon collision with<br />
the detector electrodes as an ion current. Uncompensated<br />
ions are scattered toward the ion filter<br />
electrodes, neutralized, and swept out by the carrier<br />
gas. By noting the applied field conditions (voltages)<br />
and the current level at the detector electrode, various<br />
ion species can be identified. The DMS can also be programmed<br />
to sweep through a range of compensation<br />
voltages over a specified amount of time (seconds). This<br />
is beneficial as it allows simultaneous detection of various<br />
ion species that originate from the same sample. An<br />
example of a DMS chip is shown in Figure 2(a); a development<br />
platform (Sionex Corporation, Waltham, MA)<br />
designed for proof of principle and user evaluation is<br />
shown in Figure 2b.<br />
34<br />
Sample<br />
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
Pyrolysis<br />
∆H<br />
67-keV<br />
Ionization<br />
63Ni<br />
Figure 1. Schematic of pyrolysis-DMS operation. A solid or liquid sample is pyrolyzed, and the fragments are then ionized<br />
as they flow past a radioactive nickel source. The ions are then swept into the drift tube of the DMS where they<br />
are separated based on their mobility in the applied electric fields. Under the appropriate compensation voltage,<br />
an ion will leave the drift tube and strike a Faraday plate that detects the strike based on the charge transfer.<br />
+<br />
+<br />
Drift Tube<br />
+<br />
Compensation<br />
Electric Field<br />
Drift Tube<br />
V<br />
Detector<br />
RF<br />
Electric Field<br />
Detector<br />
(+)<br />
Figure 2. (a) Picture of a DMS chip. The drift tube region<br />
and detectors are shown with arrows. (b)<br />
Photograph of a hand-held Sionex Corporation,<br />
Inc. DMS prototype unit.<br />
Bacillus Spore Preparation<br />
B. subtilis spores were selected as a surrogate for B.<br />
anthracis to evaluate the ability of the DMS to detect these<br />
spores. B. subtilis strain SMY, a wild-type, prototrophic,<br />
Marburg strain (obtained from P. Schaeffer), [42] was pregrown<br />
overnight at 30°C on a plate of tryptose blood agar<br />
base (Difco Laboratories, Franklin Lakes, NJ) and used to<br />
inoculate 2-L of DS medium [43] in a 6-L Erlenmeyer flask.<br />
The flask was incubated with shaking (200 rpm) at 37°C<br />
for 48 hours. The cells were harvested by centrifugation at<br />
13,000 x g for 20 min at 4°C, washed four times with 100ml<br />
sterile, deionized water, and resuspended in 20-ml<br />
sterile water. The suspension was estimated to contain<br />
95% mature, refractile spores by phase contrast<br />
(-)
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
microscopy. The spore titer was determined by assaying<br />
colony formation on DS agar plates after heating to 80°C<br />
for 10 min. Spores were diluted in sterile water when<br />
lower concentrations were required for testing.<br />
The Effect of Pyrolysis on B. subtilis Spores as<br />
Determined by SDS-PAGE<br />
Pyrolysis was used as the sample introduction technique<br />
tested to volatilize the spores before entering the DMS. The<br />
Pyroprobe 1000 from CDS Analytical, Inc. (Oxford, PA) was<br />
used. This unit is fabricated to interface with a GC inlet<br />
port. A slurry containing spores resuspended in sterile water<br />
was loaded into a quartz tube (2.5 mm outer diameter, 1.95<br />
mm inner diameter, 25.5 mm length) that was placed in the<br />
pyrolysis probe. The probe was then loaded into the pyrolysis<br />
unit, which has a nitrogen environment. The sample<br />
can then be pyrolyzed, and if there is gas flow, it will be carried<br />
out of the pyrolysis unit and into a GC column. For<br />
these experiments, we used a 0.5-m deactivated fused silica<br />
column only (Agilent Technologies, Palo Alto, CA).<br />
To better understand the effect of pyrolysis on B. subtilis<br />
spores, the spores were pyrolyzed using various conditions<br />
and examined using denaturing polyacrylamide gel electrophoresis<br />
(SDS-PAGE). Fifteen µg of spores<br />
(approximately 1.5e8 spores) resuspended in water were<br />
exposed initially to 110°C interface temperature and then<br />
pyrolyzed at a ramping rate of 0.01°C/ms under each of the<br />
following conditions: 250°C, 5 s; 250°C, 10 s; 350°C, 5 s;<br />
350°C, 10 s; 450°C, 5 s; 450°C, 10 s; 550°C, 5 s; and<br />
550°C, 10 s; 600°C, 5 s; 650°C, 5 s; 700°C, 5 s; 750°C, 5 s;<br />
800°C, 5 s; 850°C, 5 s; and 900°C, 5 s. The pyrolyzed samples<br />
were recovered with Laemmli sample buffer (Bio-Rad),<br />
and dithiothreitol (DTT) was added to a final concentration<br />
of 300 mM to denature the proteins prior to SDS-PAGE<br />
analysis. The samples were then placed in boiling water for<br />
5 min and run on a gradient 4-20% tris-glycine polyacrylamide<br />
gel (Bio-Rad, Hercules, CA). A sample of 1.5 µg of<br />
spores that were not pyrolyzed or exposed to the 110°C<br />
interface temperature was included as a control.<br />
As the gradient gels consist of a gradually increasing density<br />
of polyacrylamide matrix, proteins can often stain<br />
unevenly. We found that the larger proteins in the more permeable<br />
matrix in the top area of the gel are able to stain<br />
more efficiently than those that are smaller and found in the<br />
thicker polyacrylamide matrix at the bottom of the gel. To<br />
minimize the disproportionate staining, we found that it<br />
was best to prestain the gel initially with Coomassie Blue,<br />
destain the gel, and then stain with silver to produce uniform<br />
results. Other groups have used the technique of<br />
consecutive Coomassie Blue and silver staining in the examination<br />
of proteins separated by electrophoresis in gradient<br />
gels, [44] and silver staining reveals proteins at lower concentration<br />
levels than other analytical methods.<br />
The gels were stained in a 40% methanol/10% acetic acid/<br />
0.01% Coomassie Blue R-250 solution for 1 h, and then<br />
destained in a 40% methanol/10% acetic acid solution<br />
overnight. The gels were then prepared for silver staining by<br />
washing twice in deionized water for 30 min. The Silver<br />
Stain Plus Kit (Bio-Rad) was used to silver stain the gels for<br />
20 min, with the silver stain solution made per manufacturer<br />
instructions. The staining reaction was stopped by<br />
exposing the gel to a 5% acetic acid solution for 20 min, and<br />
the gels were then washed with deionized water for an additional<br />
20 min before recording the results by digital<br />
photography.<br />
Bacillus Spore Analysis Using DMS<br />
Pyrolysis was the sample introduction method for DMS<br />
analysis of spores. The pyrolysis unit was attached to an<br />
inlet port of an HP 5890 GC. The sample was pyrolyzed and<br />
swept into 0.5 m of a fused silica guard column with the<br />
carrier gas, helium (He), at 150 ml/min flow. This flow exited<br />
the GC oven and joined 50 ml/min nitrogen (N2) flow to<br />
enter the DMS. Mass flow controllers (MKS Instruments,<br />
Andover, MA) were used to ensure constant volumetric flow<br />
of the carrier gases. The GC oven temperature was set to a<br />
250°C constant temperature.<br />
The DMS was programmed so that the compensation electric<br />
field swept through a range of voltages from -40 to 10 V<br />
every 1.6125 s. The RF field was set to 1200 V. The detection<br />
of ions at the various compensation voltages over time<br />
is recorded on a laptop computer that is connected to the<br />
DMS unit. Ten 4-µl sterile water samples were pyrolyzed<br />
and their spectra from the DMS recorded. These spectra give<br />
the background signal of the water in which the spores were<br />
resuspended. Next, 10 samples of 120,000 spores (3 x 108 spores/ml slurry) were pyrolyzed and the DMS spectra<br />
recorded. Finally, 10 samples of 40,000 spores (1 x 107 spores/ml slurry) were pyrolyzed. The pyrolysis unit was<br />
held at a 150°C interface temperature, and the samples were<br />
pyrolyzed at 550°C for 100 s, with a 20°C/ms ramping rate.<br />
The spectra were analyzed using OriginPro 7.0 and the data<br />
graphed as the signal intensity vs. the compensation voltage.<br />
Each plotted line is the signal average of 25 voltage sweeps<br />
after spore pyrolysis in a single experiment. The baseline of<br />
all data signals was shifted to zero, and the maximum<br />
absolute value across each scan was calculated. Each spectral<br />
sweep was then normalized at the highest peak.<br />
Simulant Nerve and Blister Agent DMS Detection<br />
Solutions of methyl salycilate (MS) (Kintek, La Marque,<br />
TX), a simulant nerve agent, were made at the following<br />
concentrations: 344 parts-per-billion (ppb), 57 ppb, 9.5<br />
ppb, 1.6 ppb, 0.27 ppb, and 0.045 ppb. These solutions<br />
were analyzed with the DMS. The intensity of the signal was<br />
examined as a function of concentration. Next, solutions of<br />
MS and dimethyl methyl phosphonate (DMMP) (Kintek, La<br />
35
Marque, TX), a simulant blister agent, were analyzed<br />
using the DMS. The spectrum for each compound was<br />
first recorded individually. Then, a mixture of DMMP<br />
and MS was prepared, and the spectrum of this mixture<br />
was recorded.<br />
RESULTS<br />
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
Pyrolysis was used to volatilize the samples for introduction<br />
into the DMS. The SDS-PAGE results show that<br />
when the spores are pyrolyzed at higher temperatures,<br />
they are fragmented into molecules smaller than the<br />
limit of detection of this assay (Figure 3). The lower<br />
limit of band resolution on the 4-20% tris-glycine gels is<br />
10 kDa, although proteins as small as 1-2 kDa can still<br />
be detected. At the lower pyrolysis temperatures (lanes<br />
2-4), the band patterns of the pyrolyzed spore samples<br />
look similar to the spores that did not undergo pyrolysis<br />
(lanes 1 and 10), indicating little, if any, disruption to<br />
the spores when exposed to these temperatures. The<br />
bands seen here are likely due to slight fragmenting of<br />
the spores when exposed to the denaturing conditions<br />
during SDS-PAGE sample preparation. The intermediate<br />
temperatures (lanes 5-7) cause proteins of high molecular<br />
weight to be released, indicating that the spores have<br />
been disrupted but not efficiently volatilized. At temperatures<br />
of 650°C and greater held for 5 s (lanes 13-18)<br />
and at 550°C when held for 10 s (lane 9), the recovered<br />
material from the pyrolyzed spores does not show up on<br />
the gel, indicating that the fragments in the sample are<br />
not retained in the gel and so are well under 10 kDa.<br />
Figure 3. Effect of pyrolysis on B. subtilis spores. 4-20%<br />
tris-glycine SDS-PAGE of Bacillus subtilis spores,<br />
stained with Silver Stain Plus Kit (Bio-Rad). All<br />
pyrolyzed samples are 15 µg of B. subtilis spores<br />
exposed initially to 110°C interface temperature<br />
and with a ramping rate of 0.01 C/ms. Lanes: 1,<br />
1.5 µg B. subtilis spores; 2, pyrolyzed spores at<br />
250°C, 5 s; 3, pyrolyzed spores at 250°C, 10 s; 4,<br />
pyrolyzed spores at 350°C, 5 s; 5, pyrolyzed<br />
spores at 350°C, 10 s; 6, pyrolyzed spores at<br />
450°C, 5 s; 7, pyrolyzed spores at 450°C, 10 s; 8,<br />
pyrolyzed spores at 550°C, 5 s; 9, pyrolyzed<br />
spores at 550°C, 10 s. Middle: Precision Plus<br />
Protein Standard (Bio-Rad). Lanes: 10, 1.5 µg B.<br />
subtilis spores; 11, pyrolyzed spores at 550°C, 5 s;<br />
36<br />
1 2 3 4 5 6 7 8 9 MW<br />
kDa<br />
250<br />
150<br />
100<br />
75<br />
50<br />
37<br />
25<br />
20<br />
15<br />
10<br />
10 11 12 13 14 15 16 17 18<br />
12, pyrolyzed spores at 600°C, 5 s; 13, pyrolyzed<br />
spores at 650°C, 5 s; 14, pyrolyzed spores at<br />
700°C, 5 s; 15, pyrolyzed spores at 750°C, 5 s; 16,<br />
pyrolyzed spores at 800°C, 5 s; 17, pyrolyzed<br />
spores at 850°C, 5 s; 18, pyrolyzed spores at<br />
900°C, 5 s.<br />
B. subtilis spore samples were then analyzed using the<br />
DMS. They were pyrolyzed and carried though a 0.5-m<br />
deactivated fused silica column with helium as a carrier<br />
gas, and then joined with nitrogen to flow into the DMS.<br />
The compensation field, time, and abundance of ions<br />
striking the detector were recorded. The results for the<br />
sterile water and for the two different concentrations of<br />
spores are shown in Figure 4. The plots show DMS signal<br />
amplitude on the y-axis and the compensation<br />
voltage on the x-axis. Each line represents a different<br />
sample’s spectra averaged over time. Major biomarker<br />
peaks associated with the presence of spores are seen at<br />
-29 V, -12 V, and -3 V (peaks 1, 3, and 4, respectively).<br />
The peaks are concentration-dependent and are not<br />
detected in the water controls. Many smaller biomarkers<br />
are seen between -12 V and -3 V, although they are not<br />
consistent in amplitude between experimental trials.<br />
(a)<br />
(b)<br />
DMS Signal Amplitude (V)<br />
2<br />
1<br />
1 2 3 4 5<br />
0<br />
-40<br />
2<br />
-30 -20 -10 0 10<br />
1<br />
(c) 2<br />
0 -40 -30 -20 -10 0 10<br />
1<br />
0<br />
-40 -30 -20 -10 0 10<br />
Compensation Voltage (V)<br />
Figure 4. Major and minor DMS spectral biomarkers from<br />
B. subtilis spore pyrolysis: (a) water only controls,<br />
(b) 40,000 pyrolyzed spores, (c) 120,000<br />
pyrolyzed spores. Biomarkers 1, 3, and 4 appear<br />
to correlate with the presence of spores, and the<br />
amplitude is concentration-dependent.<br />
MS, a nerve agent simulant, was used in six concentrations<br />
ranging from 344 ppb to 45 ppt to determine the<br />
ability of the DMS to detect the compound, as well as its<br />
limit of detection. Figures 5(a) and 5(b) show these<br />
results; the DMS detected part-per-trillion concentrations<br />
of MS with very high signal-to-noise ratios at<br />
concentrations well below those of clinical significance.
Intensity (a.u.)<br />
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
Methyl Salycilate<br />
(a) 389 ng (344 ppb)<br />
(b)<br />
65 ng (57 ppb)<br />
10.8 ng (9.5 ppb)<br />
1.8 ng (1.6 ppb)<br />
0.3 ng (0.27 ppb)<br />
0.05 ng (0.045 ppb)<br />
0<br />
-25 -20 -15 -10 -5 0 5 10 -12 -7<br />
Compensation Voltage (V)<br />
Log of Sample Amount (gr)<br />
Figure 5. Simulant chemical weapon detection with the DMS. (a) DMS spectra of varying concentrations of MS. The signal<br />
intensity is shown as a function of the compensation voltage. (b) DMS signal peak area as a function of mass of<br />
MS introduced. Amount of compound introduced corresponds to concentrations ranging from 344 ppb to 45 ppt.<br />
Also shown is the molecular structure of MS.<br />
Additionally, the ability of the DMS to simultaneously<br />
detect two chemical weapon agent simulants was examined.<br />
For these studies, single and dual-compound<br />
solutions of DMMP and MS were employed. Figure 6(a)<br />
shows the DMS spectra collected for the single compounds.<br />
As can be seen, the DMS produces a unique<br />
spectral signature for each compound. Figure 6(b) shows<br />
the spectrum for a DMMP-MS mixture (red line) superimposed<br />
on the spectrum for MS (green line). The resultant<br />
spectrum shows features of each of the two compounds. In<br />
the positive ion spectrum, the DMMP peak at about -5 V<br />
and the MS peak at about -2 V are observed. In the negative<br />
ion spectrum, only the MS peak at -5 V shows up<br />
clearly. This demonstrates that simultaneous detection of<br />
the two simulants can be achieved with the DMS.<br />
(a)<br />
DMMP<br />
DMMP<br />
MS<br />
MS<br />
(b)<br />
Compensation Voltage (V)<br />
DISCUSSION<br />
Bacillus anthracis is an example of a biological agent that<br />
could be used as a biological weapon, as the spores can be<br />
dispersed easily, are highly resistant to extreme conditions,<br />
and are highly pathogenic. In this study, we demonstrate<br />
the ability of the DMS to detect biomarkers of B. subtilis,<br />
a surrogate for B. anthracis, chosen for its ready availability<br />
and lack of pathogenicity. As we have now<br />
demonstrated that pyrolysis coupled with DMS is a useful<br />
method for the analysis of spore components, we can<br />
hypothesize that different markers would be found in B.<br />
anthracis spores compared with B. subtilis spores, as the<br />
spores differ considerably in their coat proteins and in<br />
some cytoplasmic proteins. In fact, for this reason, it is<br />
unlikely that the same peaks would be found if examining<br />
Figure 6. Simulant chemical weapon detection with the DMS. (a) DMS spectra of MS and DMMP. Spectra were collected<br />
sequentially. (b) Simultaneous detection of nerve and blister agent simulants DMMP and MS. A mixture of DMMP<br />
and MS is graphed with signal intensity as a function of compensation voltage (green line). The spectrum for MS<br />
(red line) is shown again for reference.<br />
Log of Peaks Area<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
45 ppt<br />
MS<br />
DMMP+MS<br />
MS<br />
DMMP+MS<br />
344 ppb<br />
Compensation Voltage (V)<br />
0<br />
0<br />
H0<br />
Positive Ions<br />
Negative Ions<br />
-30 -20 -10 0 10 -30 -20 -10 0 10<br />
37
B. anthracis spores. This would be beneficial as it would<br />
allow distinction between B. subtilis or other spores from<br />
B. anthracis. Additionally, we further demonstrate the ability<br />
of the DMS to detect chemical weapon agent simulants.<br />
Pyrolysis was used as the sample introduction method for<br />
the DMS, which uses thermal energy to break apart chemical<br />
bonds. [45] The fragmentation of the sample will be<br />
characteristic of the relative strengths of the bonds in the<br />
original molecule. The maximum known detectable particle<br />
size for the DMS is about 500 Daltons (0.5 kDa) based<br />
on previous results (data not shown). We tested several<br />
pyrolysis methods to ensure that the spores were broken<br />
down sufficiently for detection by the DMS. Based on the<br />
SDS-PAGE results shown, the spores were fragmented<br />
completely at temperatures above 650°C. However, it is<br />
important to note that the spores were also completely<br />
fragmented at the lower temperature 550°C when this<br />
temperature was held for 10 s. The fragmentation thus<br />
seems to be dependent not only on temperature, but also<br />
on the time at which this temperature is held. This allows<br />
flexibility in determining an optimal pyrolysis condition to<br />
use for DMS analysis. The knowledge of various pyrolysis<br />
conditions that ensure adequate fragmentation for DMS<br />
analysis is important not only for these experiments, but<br />
for any field setups in the future that use pyrolysis as a<br />
sample introduction method.<br />
We examined the response of the DMS to pyrolyzed<br />
spores. We chose to use a lower pyrolysis temperature and<br />
extend it over a longer period of time in an attempt to<br />
ensure fragmentation of the spores to particles well under<br />
the detection limit of the gel, but also without complete<br />
destruction of any characteristics of these molecular fragments.<br />
The DMS spectra shown in Figure 4 demonstrate the ability<br />
of the DMS to distinguish the spores from the sterile<br />
water in which they are suspended. Significant peaks are<br />
marked with numbers 1-5. Peaks 1, 3, and 4 appear to be<br />
correlated with the presence of spores and could potentially<br />
mark the presence of specific biomarkers. In fact, it<br />
seems that these peaks may also be concentration-dependent,<br />
as their magnitude appears to increase proportionally<br />
to the number of spores present in the sample. The spectra<br />
for the spores can be distinguished from that of the<br />
water in which they were resuspended.<br />
We also demonstrate the use of the DMS for chemical<br />
weapons agents. MS was used as a simulant nerve agent.<br />
The DMS produces a clear spectrum for MS at concentration<br />
levels as low as 45 ppt, as shown in Figures 5a and 5b.<br />
Another compound, DMMP, was used as a simulant blister<br />
agent. A unique DMS spectrum is obtained for this compound<br />
when compared with the one obtained for MS.<br />
When a mixture of these two compounds was analyzed, the<br />
resultant spectrum showed features of both compounds.<br />
38<br />
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
This experiment shows that the specificity of the DMS<br />
enables the detection of multiple chemical warfare agents<br />
simultaneously.<br />
CONCLUSIONS<br />
In view of the growing concern over terrorism, there is a<br />
need for a sensor that can detect trace amounts of chemical<br />
or biological warfare agents to minimize the potential<br />
impact of such a release on the public. Such a sensor<br />
would aid the government in the defense against chemical<br />
and biowarfare attacks; an early alert to such an attack<br />
could save many lives. It would also aid the public health<br />
sector in the treatment of biowarfare victims by speeding<br />
diagnosis based on the identification of the particular<br />
agent delivered. To service both industries, such a sensor<br />
must be portable, inexpensive, sensitive, and able to detect<br />
multiple biological or chemical agents.<br />
<strong>Draper</strong> <strong>Laboratory</strong> has designed a sensor that fits these<br />
characteristics, the DMS. Sionex Corporation is a <strong>Draper</strong><br />
spin-off company created to commercialize and market<br />
this new device. The DMS has several advantages over<br />
other types of detectors. It is small, extremely sensitive,<br />
and relatively inexpensive. Not only are conventional ion<br />
mobility spectrometers much larger and more expensive,<br />
they also operate with short pulses of ions. In comparison,<br />
the DMS analyzes continuously-introduced ions with<br />
nearly 100% of the “tuned” ions reaching the detector.<br />
Also, as the electric fields required to filter the ions are on<br />
the order of 10,000 V/cm, the DMS actually benefits from<br />
miniaturization; by reducing the gap dimensions to the<br />
order of 500 µm, the voltages required for ion filtering are<br />
easily achievable. Mass spectrometers are larger and more<br />
expensive. They also require an inert environment for<br />
detection, whereas the DMS operates at atmospheric pressure.<br />
A portable hand-held DMS unit is already a reality<br />
and further miniaturization is possible.<br />
We have demonstrated the utility of the DMS to detect<br />
potential biological and chemical warfare agents. The DMS<br />
is a sensitive, hand-held device that can detect multiple<br />
biological and chemical agents, even in the presence of<br />
interferents. Furthermore, these devices can be manufactured<br />
using mass production techniques, significantly<br />
lowering their cost. The DMS identified a unique, repeatable<br />
spectrum for B. subtilis spores, used as a surrogate for<br />
B. anthracis spores. The concentrations shown here as<br />
being detected easily are very low for the reagentless and<br />
fast (less than 5 min) class of sensors. Even at such low<br />
concentrations, the spores can be distinguished from the<br />
water in which they are resuspended. This sensor with the<br />
detection capacity currently shown could be used initially<br />
as a trigger device, where a warning would be given if a<br />
signal that looks similar to that resulting from the presence<br />
of spores is discovered. In the future, however, if we can<br />
show the ability to detect lower concentrations than
shown here, we could move toward a detector that can<br />
actually make class decisions based on species and amount<br />
present, and would therefore offer alarms only in the event<br />
of an actual release. Furthermore, we have shown the<br />
capabilities of the DMS in the detection of chemical<br />
weapons agents. The sensitivity of the DMS enabled detection<br />
of a simulant nerve agent at a concentration of 45 ppt.<br />
The specificity of the sensor was demonstrated by the<br />
simultaneous detection of two chemical weapons simulants.<br />
Future experiments will test the ability of the DMS to distinguish<br />
between closely-related biological and chemical<br />
agents. We also intend to try these experiments using various<br />
interferents to further demonstrate the specificity of<br />
the sensor.<br />
ACKNOWLEDGMENTS<br />
We would like to express our thanks to the following<br />
<strong>Draper</strong> employees for their assistance on this project:<br />
Joung-Mo Kang, Andrew D. Dineen, Henry A.<br />
Raczkowski, and J. Ryan Prince. The authors would also<br />
like to thank Drs. Jeffrey A. Gelfand and Michael V.<br />
Callahan for fruitful discussions on chemical and biological<br />
weapons detection.<br />
This project was sponsored by the Department of the<br />
Army, Cooperative Agreement DAMD-99-2-9001. The<br />
content of this paper does not necessarily reflect the position<br />
or the policy of the government, and no official<br />
endorsement should be inferred.<br />
REFERENCES<br />
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
[1] Klietmann, W.F. and K.L. Ruoff, “Bioterrorism: Implications for<br />
the Clinical Microbiologist,” Clinical Microbiology Reviews, Vol.<br />
14, 2001, pp. 364-81.<br />
[2] Harris, S., “Japanese Biological Warfare Research on Humans: A<br />
Case Study on Microbiology and Ethics,” Ann. N.Y. Acad. Sci.,<br />
Vol. 666, 1992, pp. 21-52.<br />
[3] Abramova, F.A., L.M. Grinberg, O.V. Yampolskaya, D.H. Walker,<br />
“Pathology of Inhalation Anthrax in 42 Cases from the<br />
Sverdlovsk Outbreak of 1979,” Proc. Natl. Acad. Sci., Vol. 90,<br />
1993, pp. 2291-2294.<br />
[4] Olson, K.B., “Aum Shinrikyo: Once and Future Threat?” Emerg.<br />
Infect. Dis., Vol. 5, 1999, pp. 513-6.<br />
[5] Torok, T.J., R.V. Tauxe, R.P. Wise, J.R. Livengood, R. Sokolow, S.<br />
Mauvais, K.A. Birkness, M.R. Skeels, J.M. Horan, L.R. Foster, “A<br />
Large Community Outbreak of Salmonellosis Caused by<br />
Intentional Contamination of Restaurant Salad Bars,” JAMA,<br />
Vol. 278, 1997, pp. 389-95.<br />
[6] Jernigan, J.A., D.S. Stephens, D.A. Ashford, C. Omenaca, M.S.<br />
Topiel, M. Galbraith, “Bioterrorism-Related Inhalational<br />
Anthrax: the First 10 Cases Reported in the United States,”<br />
Emerg. Infect. Dis., Vol. 7, 2001, pp. 933-44.<br />
[7] Mock, M. and A. Fouet, “Anthrax,” Annu. Rev. Microbiol., Vol.<br />
55, 2001, pp. 647-71.<br />
[8] Broussard, L.A., “Biological Agents: Weapons of Warfare and<br />
Bioterrorism,” Molecular Diagnostics, Vol. 6, 2001, pp. 323-33.<br />
[9] Bozeman, W.P., D. Dilbero, J.L. Schauben, “Biological and<br />
Chemical Weapons of Mass Destruction,” Emerg. Med. Clin.<br />
North Am., Vol. 20, 2002, pp. 975-93.<br />
[10] Evison, D., D. Hinsley, P. Rice, “Chemical Weapons,” BMJ, Vol.<br />
324, 2002, pp. 332-5.<br />
[11] Parker, M., P. Emanuel, J. Wilson, P. Estacio, M. Callahan, D.<br />
Cullin, Proceedings from The Consensus Conference on the Role<br />
of Biosensors in the Detection of Agents of Bioterrorism,<br />
Telemedicine & Advanced Technology Research Center<br />
(TATRC), 2003, pp. 1-15.<br />
[12] Turnbull, P.C., R.A. Hutson, M.J. Ward, M.N. Jones, C.P. Quinn,<br />
N.J. Finnie, C.J. Duggleby, J.M. Kramer, and J. Melling,<br />
“Bacillus anthracis but Not Always Anthrax,” J. Appl. Bacteriol.,<br />
Vol. 72, 1992, pp. 21-8.<br />
[13] Patra, G., P. Sylvestre, V. Ramisse, J. Therasse, J.L. Guesdon,<br />
“Isolation of a Specific Chromosomic DNA Sequence of Bacillus<br />
anthracis and Its Possible Use in Diagnosis,” FEMS Immunol.<br />
Med. Microbiol., Vol. 15, 1996, pp. 223-31.<br />
[14] Cooney, S., “Rapid Anthrax Test Development,” Nature<br />
Medicine, Vol. 7, pp. 1265.<br />
[15] Dubiley, S., E. Kirillov, A. Mirzabekov, “Polymorphism Analysis<br />
and Gene Detection by Minisequencing on an Array of Gel-<br />
Immobilized Primers,” Nucleic Acids Res., Vol. 27, 1999, pp. e19.<br />
[16] Higgins, J.A., M.S. Ibrahim, F.K. Knauert, “Sensitive and Rapid<br />
Identification of Biological Threat Agents,” Ann. N.Y. Acad. Sci.,<br />
Vol. 894, 1999, pp. 130-48.<br />
[17] Beyer, W., S. Pocivalsek, R. Bohm, “Polymerase Chain Reaction-<br />
ELISA to Detect Bacillus anthracis from Soil Samples<br />
-Limitations of Present Published Primers,” J. Appl. Microbiol.,<br />
Vol. 87, 1999, pp. 229-36.<br />
[18] Bruno, J.G. and J.L. Kiel, “In Vitro Selection of DNA Aptamers<br />
to Anthrax Spores with Electrochemiluminescence Detection,”<br />
Biosens. Bioelectron., Vol. 14, 1999, pp. 457-64.<br />
[19] Lee, M.A., G. Brightwell, D. Leslie, H. Bird, A. Hamilton,<br />
“Fluorescent Detection Techniques for Real-Time Multiplex<br />
Strand-Specific Detection of Bacillus anthracis Using Rapid<br />
PCR,” J. Appl. Microbio., Vol. 87, 1999, pp. 218-23.<br />
[20] Strizhkov, B.N., A.L. Drobyshev, V.M. Mikhailovich, A.D.<br />
Mirzabekov, “PCR Amplification on a Microarray of Gel-<br />
Immobilized Oligonucleotides: Detection of Bacterial Toxinand<br />
Drug-Resistant Genes and Their Mutations,” Biotechniques,<br />
Vol. 29, 2000, pp. 844-848.<br />
[21] Makino, S.I., H.I. Cheun, M. Watarai, I. Uchida, K. Takeshi,<br />
“Detection of Anthrax Spores from the Air by Real-Time PCR,”<br />
Lett. Appl. Microbiol., Vol. 33, 2001, pp. 237-40.<br />
[22] Uhl, J.R., C.A. Bell, L.M. Sloan, M.J. Espy, T.F. Smith, J.E.<br />
Rosenblatt, F.R. Cockerill, “Application of Rapid-Cycle Real-<br />
Time Polymerase Chain Reaction for the Detection of Microbial<br />
Pathogens: the Mayo-Roche Rapid Anthrax Test,” Mayo Clin.<br />
Proc., Vol. 77, 2002, pp. 673-80.<br />
[23] Phillips, A.P. and K.L. Martin, “Quantitative Immunofluorescence<br />
Studies of the Serology of Bacillus anthracis<br />
Spores,” Appl. Environ. Microbiol., Vol. 46, 1983, pp. 1430-2.<br />
[24] Phillips, A.P., K.L. Martin, M.G. Broster, “Differentiation<br />
Between Spores of Bacillus anthracis and Bacillus cereus by a<br />
Quantitative Immunofluorescence Technique,” J. Clin.<br />
Microbiol., Vol. 17, 1983, pp. 41-7.<br />
[25] Phillips, A.P. and K.L. Martin, “Investigation of Spore Surface<br />
Antigens in the Genus Bacillus by the Use of Polyclonal<br />
Antibodies in Immunofluorescence Tests,” J. Appl. Bacteriol.,<br />
Vol. 64, 1988, pp. 47-55.<br />
39
[26] Yu, H., “Comparative Studies of Magnetic Particle-Based Solid<br />
Phase Fluorogenic and Electrochemiluminescent Immunoassay,”<br />
J. Immunol. Methods., Vol. 218, 1998, pp. 1-8.<br />
[27] Longchamp, P. and T. Leighton, “Molecular Recognition<br />
Specificity of Bacillus anthracis Spore Antibodies,” J. Appl.<br />
Microbiol., Vol. 87, 1999, pp. 246-9.<br />
[28] De, B.K., S.L. Bragg, G.N. Sanden, K.E. Wilson, L.A. Diem, C.K.<br />
Marston, A.R. Hoffmaster, G.A. Barnett, R.S. Weyant, T.G.<br />
Abshire, J.W. Ezzell, T. Popovic, “A Two-Component Direct<br />
Fluorescent-Antibody Assay for Rapid Identification of Bacillus<br />
anthracis,” Emerg. Infect. Dis., Vol. 8, 2002, pp. 1060-5.<br />
[29] Zhou, B., P. Wirsching, K.D. Janda, “Human Antibodies Against<br />
Spores of the Genus Bacillus: a Model Study for Detection of<br />
and Protection Against Anthrax and the Bioterrorist Threat,”<br />
Proc. Natl. Acad. Sci. USA, Vol. 99, 2002, pp. 5241-6.<br />
[30] Shute, L.A., C.S. Gutteridge, J.R. Norris, R.C. Berkeley, “Curie-<br />
Point Pyrolysis Mass Spectrometry Applied to Characterization<br />
and Identification of Selected Bacillus Species,” J. Gen.<br />
Microbiol., Vol. 130, 1984, pp. 343-55.<br />
[31] Fox, A., G.E. Black, K. Fox, S. Rostovtseva, “Determination of<br />
Carbohydrate Profiles of Bacillus anthracis and Bacillus cereus<br />
Including Identification of O-Methyl Methylpentoses Using Gas<br />
Chromatography-Mass Spectrometry,” J. Clin. Microbiol., Vol.<br />
31, 1993, pp. 887-94.<br />
[32] Hathout, Y., P.A. Demirev, Y.P. Ho, J.L. Bundy, V. Ryzhov, L.<br />
Sapp, J. Stutler, J. Jackman, C. Fenselau, “Identification of<br />
Bacillus Spores by Matrix-Assisted Laser Desorption Ionization-<br />
Mass Spectrometry,” Appl. Environ. Microbiol., Vol. 65, 1999,<br />
pp. 4313-9.<br />
[33] Demirev, P.A., J. Ramirez, C. Fenselau, “Tandem Mass<br />
Spectrometry of Intact Proteins for Characterization of<br />
Biomarkers from Bacillus cereus T Spores,” Anal. Chem., Vol.<br />
73, 2001, pp. 5725-31.<br />
[34] Elhanany, E., R. Barak, M. Fisher, D. Kobiler, Z. Altboum,<br />
“Detection of Specific Bacillus anthracis Spore Biomarkers by<br />
Matrix-Assisted Laser Desorption/Ionization Time-of-Flight<br />
Mass Spectrometry,” Rapid Commun. Mass Spectrom., Vol. 15,<br />
2001, pp. 2110-6.<br />
[35] Mowry, C., C. Morgan, Q. Baca, R. Manginell, R. Kotenstette, P.<br />
Lewis, G. Frye-Mason, “Rapid Detection of Bacteria with<br />
Miniaturized Pyrolysis-GC Analysis,” Proceedings of SPIE,<br />
Boston, MA, 2001.<br />
40<br />
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
[36] Miller, R.A., G.A. Eiceman, E.G. Nazarov, A.T. King, “A Novel<br />
Micromachined High-Field Asymmetric Waveform-Ion<br />
Mobility Spectrometer,” Sensors and Actuators, Vol. 67, 2000,<br />
pp. 300-6.<br />
[37] Eiceman, G.A., B. Tadjikov, E. Krylov, E.G. Nazarov, R.A. Miller,<br />
J. Westbrook, P. Funk, “Miniature Radio-Frequency Mobility<br />
Analyzer as a Gas Chromatographic Detector for Oxygen-<br />
Containing Volatile Organic Compounds, Pheromones, and<br />
Other Inset Attractants,” J. Chromatography, Vol. 917, 2001,<br />
pp. 205-17.<br />
[38] Miller, R.A., G.A. Eiceman, E.G. Nazarov, A.T. King, “A MEMS<br />
Radio-Frequency Ion Mobility Spectrometer for Chemical Agent<br />
Detection,” presented at Solid-State Sensor and Actuator<br />
Workshop, Hilton Head Island, SC, 2000.<br />
[39] Krylov, E., E.G. Nazarov, R.A. Miller, B. Tadjikov, G.A. Eiceman,<br />
“Field Dependence of Mobilities for Gas-Phase-Protonated<br />
Monomers and Proton-Bound Dimers of Ketones by Planar<br />
Field Asymmetric Waveform Ion Mobility Spectrometer<br />
(PFAIMS),” J. Phys. Chem., Vol. 106, 2002, pp. 5437-44.<br />
[40] Miller, R.A., E.G. Nazarov, G.A. Eiceman, A.T. King, “A MEMS<br />
Radio-Frequency Ion Mobility Spectrometer for Chemical Vapor<br />
Detection,” Sensors and Actuators, Vol. 91, 2001, pp. 307-18.<br />
[41] Davis, C.E., J.M. Kang, C.E. Dubé, J.T. Borenstein, E.G.<br />
Nazarov, R.A. Miller, A.M. Zapata, “Spore Biomarker Detection<br />
Using a MEMS Differential Mobility Spectrometer,” presented at<br />
Transducers, Solid-State Sensors, Actuators and Microsystems,<br />
2003.<br />
[42] Schaeffer, P., J. Millet, and J.P. Aubert, “Catabolic Repression of<br />
Bacterial Sporulation,” Proc. Natl. Acad. Sci. USA, Vol. 54,<br />
1965, pp. 704-711.<br />
[43] Fouet, A. and A.L. Sonenshein, “A Target for Carbon Source-<br />
Dependent Negative Regulation of the citB Promoter of Bacillus<br />
subtilis,” J. Bacteriol., Vol. 172, 1990, pp. 835-44.<br />
[44] Kirschstein, M., R. Jensen, R. Schroder, K. Sack, “Proteinuria<br />
after Renal Transplantation: Diagnosis with Highly Sensitive<br />
Silver Stain in Sodium Dodecylsulphate-Polyacrylamide<br />
Gradient Gel Electrophoresis,” Klin. Wochenschr, Vol. 69, 1991,<br />
pp. 847-52.<br />
[45] Wampler, T.P., Analytical Pyrolysis: An Overview, Marcel<br />
Dekker, Inc., New York, 1995.
Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology<br />
(l-r) Angela M. Zapata, Melissa D. Krebs<br />
Melissa D. Krebs is a Member of the Technical Staff at <strong>Draper</strong> <strong>Laboratory</strong>. Her Masters thesis work involved the study of<br />
protein interactions in the cellulosome of Clostridium thermocellum to investigate the mechanism for the efficient degradation<br />
of cellulose. Her current research interests include various application areas for DMS sensor technology, including<br />
biowarfare detection and clinical diagnostics. She received BS and MS degrees in Chemical Engineering from the University<br />
of Rochester (2002 and 2003, respectively).<br />
Angela M. Zapata is a Senior Member of the Technical Staff in the Biomedical Engineering Group at <strong>Draper</strong> <strong>Laboratory</strong>. She<br />
is responsible for developing novel technologies for solutions in clinical diagnostics and therapeutics, environmental monitoring<br />
and remediation, and homeland defense. In 2002, she joined the Cambridge Rindge and Latin School, Cambridge,<br />
where she started a 4-year school-to-work biotechnology training program for high school students. She rejoined <strong>Draper</strong><br />
on a full-time basis in the fall of 2004. She holds 2 patents and 14 publications in the areas of field-deployable and MEMSbased<br />
analytical devices. Dr. Zapata received a BS in Chemistry from Worcester State College (1992) and a PhD in Analytical<br />
Chemistry from Tufts University (2000).<br />
Erkinjon G. Nazarov is a Chief Scientist at Sionex Corporation, Waltham, MA. He is a pioneer in DMS technology and has<br />
extensive experience in the design, fabrication, and evaluation of differential mobility spectrometry-based sensors. He<br />
received a BS from Leningrad Polytechnic Institute, Leningrad, Russia, a PhD from Ioffe Physical Technical Institute,<br />
Leningrad, and Dr.Phys. and Math.Sci. degrees from St. Petersburg Polytechnic University, St. Petersburg, Russia (1981 and<br />
1992, respectively).<br />
Raanan A. Miller is the Founder, Vice President of Technology, and Chief Technical Officer of Sionex Corporation,<br />
Waltham, MA. He has over 12 years experience in MEMS technology development, including sensor and actuator design,<br />
modeling and process design, and fabrication. He has developed novel electromagnetic optical switches and beam-steering<br />
systems incorporating magnetic thin films and worked on incorporating novel materials such as lead zirconate titanate<br />
(PZT) thin films into MEMS acoustic sensors and actuators. A leading expert in differential mobility spectrometry, he is<br />
responsible for the MEMS microDMS sensor development and much of the Sionex intellectual property. Dr. Miller received<br />
a BS from Boston University and MS and PhD degrees from the California Institute of Technology, Pasadena.<br />
Isaac S. Costa was previously a member of the MEMS and Bioengineering groups at <strong>Draper</strong> <strong>Laboratory</strong>, where he worked<br />
for 5 years. His research included applying DMS to spore detection and process development for the manufacture of the<br />
micro-canary sensor, the nonvolatile residue sensor, and the all-silicon oscillating accelerometer. His primary interest is in<br />
advanced process development for silicon MEMS devices based on silicon-on-insulator technology. He received a BS in<br />
Interdepartmental Sciences from the American International College, Springfield, MA.<br />
Abraham L. Sonenshein is a Professor of Molecular Biology and Microbiology at the Tufts University School of Medicine,<br />
and has studied the regulation of gene expression and sporulation in Bacillus subtilis for more than 30 years. He was the<br />
Editor of the Journal of Bacteriology from 1990 to 1996, and served as the Editor-in-Chief of “Bacillus subtilis and Other<br />
Gram-Positive Bacteria: Biochemistry, Physiology and Molecular Genetics and Bacillus subtilis” and “Its Closest Relatives:<br />
from Genes to Cells,” both published by ASM Press. Dr. Sonenshein was elected to fellowship in the American Academy of<br />
Microbiology in 1993 and has served on its Board of Governors since 1999. He received a BS in Biochemical Sciences from<br />
Princeton University (1965) and a PhD in Biology from MIT (1970).<br />
Cristina E. Davis is a Principal Member of the Technical Staff and Group Leader of Bioengineering at <strong>Draper</strong> <strong>Laboratory</strong>.<br />
Her work focuses on biological detection and analysis of biomarkers using the DMS sensor technology. Previous research<br />
includes high-throughput electrophysiology systems, membrane biophysics, signal transduction pathways in cell volume<br />
regulation, voltage-sensitive fluorescent dyes, and MEMS biosensor research. Dr. Davis received a BS from Duke University,<br />
Durham, NC, with a double major in Mathematics and Biology (1994), and MS and PhD degrees in Biomedical Engineering<br />
from the University of Virginia (1996 and 1999, respectively).<br />
41
Autonomous Mission Management for Spacecraft Rendezvous<br />
Using an Agent Hierarchy<br />
Mark Jackson, Christopher D'Souza, Hobson Lane<br />
Copyright © 2005 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc. Presented at Infotech@Aerospace. Arlington, VA,<br />
September 26-29, 2005<br />
INTRODUCTION<br />
An agent-based autonomous mission manager for spacecraft is developed and applied to a spacecraft<br />
rendezvous problem. The mission manager monitors, plans, and executes rendezvous activities<br />
for a chaser satellite with low-thrust and high-thrust effectors. The software is based on a hierarchy of<br />
“activity agents,” each of which plans, monitors, executes, and replans an activity. These activity agents<br />
are instantiated and scheduled by a software framework capable of reconfiguring the agent hierarchy<br />
when replanning occurs. Targeting and planning algorithms are incorporated into the agent hierarchy<br />
to orchestrate a rendezvous with an orbiting spacecraft. These algorithms make use of low-thrust guidance<br />
to arrive in the neighborhood of the target, followed by a series of high-thrust burns to arrive at<br />
a specified offset point at a specified time. During both the low-thrust and high-thrust activities, the<br />
mission manager monitors subsystem status as well as the relative navigation state of the chaser. When<br />
low-thrust failures occur, the manager completes the rendezvous by scheduling additional high-thrust<br />
burns. The mission manager, along with low- and high-thrust guidance algorithms, is implemented in<br />
simulation. A demonstration of several replanning events – both failure and error driven – is provided<br />
with 3-dimensional (3D) graphics as well as an “agent activity” display that depicts the active agents<br />
and shows the hierarchy reconfiguration process when replans occur.<br />
The next 10 to 20 years are likely to see an increasing need<br />
for spacecraft capable of autonomous rendezvous with<br />
other spacecraft or space objects. The National<br />
Aeronautics & Space Administration (NASA) is currently<br />
developing a set of exploration vehicles, both manned and<br />
unmanned, to accomplish several missions, including<br />
International Space Station (ISS) crew exchange, ISS<br />
resupply, lunar transit and landing, lunar return, and<br />
ultimately, Mars exploration. These missions will all<br />
require spacecraft rendezvous and docking in various<br />
42<br />
orbital environments, including low Earth orbit, lunar<br />
orbit, Martian orbit, and possibly, rendezvous and docking<br />
at or near Lagrange points. Additionally, there is significant<br />
military and commercial interest in satellite rendezvous for<br />
resupply, intelligence, and other operations.<br />
Many of these operations will be conducted by unmanned<br />
spacecraft in arenas for which close ground contact and<br />
supervision are neither practical nor desirable. Spacecraft<br />
will be required to autonomously plan rendezvous maneuvers,<br />
monitor navigation and subsystem status, diagnose
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
problems with the existing maneuver plans, and replan<br />
when required. To accomplish this, software will be<br />
required with capabilities above and beyond classic guidance<br />
algorithms.<br />
Toward this end, Northrop Grumman Corporation in collaboration<br />
with <strong>Draper</strong> <strong>Laboratory</strong> has applied an<br />
agent-hierarchy approach developed by <strong>Draper</strong> to a satellite<br />
rendezvous problem in simulation. The satellite incorporates<br />
Northrop Grumman’s low-thrust rendezvous guidance.<br />
In addition, it is equipped with high-thrust actuators and a<br />
sensor suite consisting of Global Positioning System (GPS)<br />
for absolute navigation, relative GPS, and light detection and<br />
ranging (LIDAR) for close-in rendezvous.<br />
Three sections follow. First, some background is provided<br />
on the general principles of mission management as they<br />
apply to spacecraft. This section discusses the role of the<br />
mission manager in spacecraft and discusses mission management<br />
principles and processes. The second section<br />
applies these principles to the rendezvous problem mentioned<br />
above. Finally, some simulation results are presented<br />
that demonstrate the capability of the mission manager to<br />
react to subsystem failures and navigation state updates.<br />
Background on Mission Management from Spacecraft<br />
Role of the Mission Manager<br />
For space vehicles, the role of autonomous mission management<br />
software is to configure guidance, navigation, and<br />
control (GN&C) and subsystem components to meet mission<br />
objectives (Figure 1). The inputs to the mission<br />
manager are the states and status of the vehicle and its subsystems<br />
from failure detection, isolation, and recovery<br />
(FDIR) and from navigation, as well as the current mission<br />
objectives. The outputs are commands to guidance and<br />
control (G&C), and modes and configurations for subsystems.<br />
The mission manager, therefore, multiplexes a great<br />
deal of state and status data to create the required commands.<br />
The principal challenge to the practical<br />
implementation of autonomous systems in spacecraft is<br />
not in determining the optimal path or best course of<br />
action for a given situation (or set of inputs), but rather, it<br />
is the complexity that arises from the range of possible failure<br />
situations and vehicle configurations that the software<br />
must handle. A structured approach to the design and<br />
implementation of an autonomous mission manager is<br />
absolutely critical. An ad-hoc design will quickly become<br />
brittle and may be virtually untestable.<br />
Crew/Mission Control<br />
Objectives<br />
Flight Computer Subsystems<br />
FDIR<br />
Nav<br />
Mission<br />
Manager G&C<br />
Sensors<br />
Effectors<br />
Other<br />
Figure 1. Mission manager interfaces.<br />
Much of the modern literature acknowledges four basic<br />
functions that the mission manager must perform to<br />
accomplish this multiplexing task. We shall refer to these<br />
as: monitoring, diagnosis, planning, and execution. These<br />
correspond to the observe, orient, decide, and act functions<br />
that are prevalent in the military literature. [1] We<br />
review these here since they figure prominently in our discussion<br />
of hierarchical architectures and their application<br />
to spacecraft rendezvous.<br />
Monitoring is the process of collecting state and status<br />
inputs and checking for failures or problems with the current<br />
plan. The primary job of the monitoring function is to<br />
answer the question: Given the current state and status,<br />
will the current plan and vehicle configuration meet my<br />
objectives within constraints? Often, for space applications,<br />
answering this question involves propagating the<br />
current state of the vehicle forward in time to the end of a<br />
current activity or mission phase.<br />
Diagnosis is the process of identifying problems with the<br />
current plan or configuration and deciding whether<br />
replanning or reconfiguration is necessary. Note that this<br />
diagnosis function is focused on identifying problems with<br />
plans or configurations, not on finding mechanical problems<br />
within the subsystems. Although these diagnosis jobs<br />
may overlap, we generally consider the diagnosis of subsystem<br />
faults to be part of the FDIR software.<br />
Planning is the process of creating a series of commands or<br />
configurations that achieve the mission objectives within<br />
constraints. When primary mission objectives cannot be<br />
met, planners must be capable of selecting alternate objectives.<br />
In the worst case, alternate objectives may include<br />
mission abort or spacecraft safing. Planning may occur at<br />
the beginning of a mission phase or activity or whenever<br />
diagnosis determines that replanning is necessary.<br />
Finally, the execution software enacts the plan by sequencing<br />
commands to G&C algorithms and vehicle<br />
subsystems. This sequencing may be time or event-driven<br />
depending on the nature of the plan.<br />
Note that these four functions form a continuous loop with<br />
monitoring and execution running continuously, and diagnosis<br />
and planning happening when failures occur or<br />
modeling errors accrue. The next section starts with planning<br />
and shows how a hierarchical planning process can be<br />
used to make decisions or optimize a path or configuration.<br />
Hierarchical Planning<br />
Consider the problem of moving a vehicle from point A to<br />
point C in Figure 2. We assume that subplanner 1 is capable<br />
of finding an optimal path from A to B, while<br />
subplanner 2 is capable of finding the optimal path from B<br />
to C, and that the coordinates of the intermediate point, B<br />
are B1 and B2. These coordinates are known as interface<br />
variables. This problem may be solved by a simple hierarchy<br />
of three planning entities as shown. The supervisor<br />
planner invokes subplanner 1 by requesting it to solve a<br />
problem consisting of the start state, A, the goal state, B,<br />
43
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
and any applicable constraints. Subplanner 1 returns the<br />
cost of the optimal trajectory from A to B, call it J1.<br />
Similarly, the supervisor planner may request subplanner<br />
2 to solve the B to C problem and receives the B to C<br />
cost, J2. The total cost, J1 + J2, is plotted topographically<br />
on the right as a function of the interface states<br />
(coordinates of B). The supervisor optimizes the A to C<br />
trajectory by searching for the optimal coordinates of B.<br />
Depending on the attributes of the problem, the supervisor<br />
may find the optimal solution by using one of any<br />
number of optimization techniques, gradient descent for<br />
example.<br />
Figure 2. Optimization or decision-making via hierarchical<br />
planners.<br />
On the other hand, the supervisor may not be required to<br />
find the optimal solution at all. It may instead search<br />
through a predefined set of interface states (red Xs in the<br />
figure) choosing the option with the lowest cost. When<br />
the supervisor evaluates a set of options, we shall call the<br />
process decision-making. When an iterative search is<br />
used, the process may be an optimization process. The<br />
multilevel optimization technique described here is<br />
referred to as interaction coordination. Other techniques,<br />
such as price coordination are also available. [2]<br />
Of course, this iterative process is not new to those familiar<br />
with optimization. Many optimization algorithms<br />
work by iterating on a set of control variables to minimize<br />
a cost function. But diagramming the problem this<br />
way highlights some of the advantages of thinking about<br />
planning in a hierarchical sense. Each subplanner may be<br />
an expert at its piece of the problem. Subplanner 2 need<br />
not have any knowledge of subplanner 1’s problem and<br />
vice versa. Further, the supervisor need not have all the<br />
details of the subproblems to solve the overall problem.<br />
Second, once the vehicle has passed point B, there is no<br />
longer any need to consider the A to B problem, and subplanner<br />
1 (and all its associated data and memory) may<br />
be deactivated. Finally, note that the time horizon for<br />
each planner gets longer as we move up the hierarchy<br />
with the supervisor having the lowest level of detail while<br />
planning over the longest time horizon.<br />
It is easy to imagine an extension of the two-level hierarchy<br />
described above in which each subplanner in turn<br />
solves its problem by invoking subplanners of its own.<br />
44<br />
A<br />
Subplanner<br />
1<br />
Cost = J1<br />
Supervisor<br />
Planner<br />
B = interface<br />
state<br />
B<br />
Subplanner<br />
2<br />
Cost = J2<br />
Total Cost, J = J1 + J2<br />
(out of page)<br />
B2<br />
J=50 J=40<br />
B1<br />
C<br />
Discrete decision points<br />
J=30<br />
J=50<br />
Optimum<br />
This sort of hierarchical planning breakdown has one<br />
other very important advantage. The same process that<br />
creates the planning tree results in a natural breakdown<br />
for all the functions required of a mission manager. In<br />
fact, the process of invoking subplanners and sub-subplanners<br />
can be used to create a hierarchy of agents, each<br />
responsible for monitoring, diagnosing, planning, and<br />
executing a section of the mission. We shall refer to these<br />
agents as activity agents. Activity agents that are connected<br />
to a superior agent in the hierarchy are referred to as<br />
children of the superior, while the superior agent is the<br />
parent of each child. Agents without children (bottom<br />
level nodes) are referred to as leaf agents, or leaf nodes.<br />
Figure 3 shows an example for a spacecraft rendezvous<br />
and docking mission. Each of the activity agents has four<br />
sub-boxes, corresponding to monitoring, diagnosis,<br />
planning, and execution. Just as for planning, the time<br />
horizons for each activity increases with its level in the<br />
hierarchy, while the detail of its functions decreases.<br />
This process of breaking down a complex problem or<br />
mission into activities or phases is referred to as temporal<br />
decomposition. The process of defining the functionality of<br />
each activity agent is referred to here as functional decomposition.<br />
In this context, the functional decomposition<br />
consists of defining the monitoring, diagnosis, planning,<br />
and execution functions required by each<br />
activity.<br />
Rendez<br />
D P<br />
M E<br />
Burn<br />
D P<br />
M E<br />
D P<br />
M E<br />
Survey<br />
D P<br />
M E<br />
Acquire Insert<br />
D P<br />
M E<br />
Mission<br />
D P<br />
M E<br />
Maneuver<br />
D P<br />
M E<br />
Approach<br />
D P<br />
M E<br />
Maintain Init Final<br />
D P D P D P<br />
M E M E M E<br />
Figure 3. Example hierarchy of activity agents.<br />
The interfaces between parent and child activities in<br />
Figure 3 consist of problem specifications from parent to<br />
child and solution information from child to parent. For<br />
vehicle applications, the problem specification usually<br />
consists of a start state for the child activity, a goal state,<br />
and a set of constraints. Sometimes, the goal state is only<br />
partially specified or constrained, so the child must<br />
return the final state in the solution information.<br />
Solution information may also contain sensitivity data,<br />
resource data, and cost data, etc. At the lowest level, the<br />
interfaces are commands to the GN&C and subsystems.<br />
These commands may be thought of as problem specifications<br />
to the vehicle.<br />
The implementation of these hierarchical mission management<br />
architectures requires a software framework
capable of instantiating the agent hierarchy during the<br />
planning process, sequencing the monitoring and diagnosis<br />
functions, and rebuilding the hierarchy from any<br />
level when replanning is required. The framework provides<br />
the designer with built-in activity interfaces,<br />
activity code templates, and locations for user-defined<br />
monitoring, diagnosis, planning and execution functions.<br />
The framework used in the example application<br />
below is the All Domain Execution and Planning<br />
Technology (ADEPT) developed by <strong>Draper</strong> <strong>Laboratory</strong>. [2]<br />
More details on the ADEPT framework will be provided<br />
with the examples.<br />
Design Process<br />
In the development of complex mission management<br />
applications, the design process is as important as the<br />
software architecture. Figure 4 illustrates the recommended<br />
design process. The process starts with an<br />
analysis of vehicle requirements and operations concepts.<br />
Once these are understood, a temporal<br />
decomposition is performed to create the activity hierarchy.<br />
The rendezvous application described below<br />
provides an example of a temporal decomposition. Rules<br />
of thumb are provided with the example to assist designers<br />
in creating a manageable breakdown without<br />
overdividing the problem.<br />
REQUIREMENTS<br />
Concept of<br />
Operations<br />
Operational<br />
Requirements<br />
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
DECOMPOSITION<br />
Temporal<br />
Functional<br />
Algorithm<br />
Selections<br />
Mathematical<br />
Problems<br />
Algorithm<br />
Knowledge Base<br />
PEM<br />
Schematic<br />
Selected<br />
Algorithms<br />
Populate ADEPT<br />
Architecture<br />
Figure 4. ADEPT design process.<br />
Next, a functional decomposition of each activity is performed<br />
to define requirements for each of the four<br />
activity functions. These requirements are posed as<br />
mathematical problems that may be solved by appropriate<br />
decision-making or optimization algorithms. These<br />
algorithms are then used to populate the framework of<br />
activity agents. Once the temporal and functional<br />
decompositions are complete, designers should step<br />
through the agent responses to hypothetical scenarios or<br />
“use cases” to identify any gaps in the activity hierarchy<br />
and to further identify algorithm requirements.<br />
The algorithms used by the mission manager may be<br />
divided into two broad categories: domain-specific algorithms<br />
and decision-making/optimization algorithms.<br />
The domain-specific algorithms act on vehicle state and<br />
status data to provide information about future consequences<br />
of actions. These may include state propagation<br />
routines, environmental models, and vehicle or subsystem<br />
models. The decision-making algorithms use this<br />
information about consequences to make decisions, to<br />
optimize, or to plan. These techniques may range from<br />
simple logical comparisons to sophisticated optimization<br />
or artificial intelligence algorithms.<br />
The next sections apply this design process to a spacecraft<br />
rendezvous problem.<br />
Application: Prototype Autonomous Mission<br />
Manager<br />
The hierarchy of activity agents described above was<br />
applied to an autonomous spacecraft rendezvous mission.<br />
This Prototype Autonomous Mission Manager<br />
(PAMM) was implemented in simulation aboard a chaser<br />
spacecraft whose mission was to rendezvous with a<br />
dynamically passive target. For this application, the target<br />
spacecraft was assumed to have GPS capability. This<br />
allowed the chaser navigation system to update the target<br />
location via ground uplinks and to utilize relative GPS<br />
techniques for mid-field rendezvous. The chaser effector<br />
models included a low-thrust propulsion system as well<br />
as high-thrust engines capable of actuating nearly impulsive<br />
burns. Chaser sensor models included a GPS system<br />
for absolute inertial position, a relative GPS model to<br />
provide relative state information within about 25 km,<br />
and a LIDAR rendezvous sensor capable of returning<br />
range, azimuth, and elevation information for close-in<br />
operations (
Y eci (me)<br />
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
Figure 5. Low-thrust portion of rendezvous.<br />
The low-thrust guidance algorithm works by optimizing<br />
the attitude profile during long periods of constant thrust<br />
to accomplish rendezvous, minimizing time and fuel consumption<br />
for the rendezvous.<br />
The low-thrust portion of the rendezvous may take several<br />
days, each consisting of dozens of orbits. This means<br />
that the mission manager must be capable of planning<br />
high-thrust burns to complete the rendezvous from a variety<br />
of failure conditions, resulting in a range of phase<br />
angles, altitude differentials, and lateral errors.<br />
If all goes nominally, the low-thrust guidance delivers the<br />
satellite to a point approximately 150 km behind and 30<br />
km below the RSO. Figure 6 shows the final 2500 km of<br />
the rendezvous. In the figure, the RSO is at the origin and<br />
is moving to the left. The solid green trace is the relative trajectory<br />
of the chaser in the orbit plane. As the figure shows,<br />
in the nominal case, two final high-thrust burns precisely<br />
complete the trajectory to the desired offset point.<br />
Altitude Difference (km)<br />
Figure 6. Chaser downrange and altitude errors from RSO<br />
(+Vbar left).<br />
46<br />
x107 1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
-0.2<br />
-0.4<br />
-0.6<br />
-0.8<br />
X eci (me) x106 -1<br />
-8 -6 -4 -2 0 2 4 6 8<br />
10<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
-60<br />
-70<br />
-80<br />
Target offset point<br />
Nominal Mission (No Failures)<br />
Low-thrust cutoff<br />
Downrange (km)<br />
Green: Chaser<br />
Red: RSO<br />
RSO Orbit<br />
Chaser lowthrust<br />
spiral<br />
trajectory<br />
High-thrust burns<br />
Chaser lowthrust<br />
trajectory<br />
-90 0 500 1000 1500 2000 2500<br />
When failures occur, the high-thrust system has enough<br />
impulse to accomplish a rendezvous from approximately<br />
100 to 2500 km behind the RSO. Should the failure occur<br />
outside this region, the mission manager waits until the<br />
chaser has “lapped” the RSO and arrived back in the region<br />
“behind” the RSO prior to scheduling high-thrust burns.<br />
In testing an autonomous rendezvous concept, it is important<br />
to include navigation errors, particularly in the<br />
knowledge of the target vehicle (RSO) state. Filter convergence<br />
and target state updates test the mission manager<br />
response at several levels. Smaller corrections cause<br />
replanning at lower levels in the hierarchy since they tend<br />
to affect only the current or upcoming activities. Large corrections<br />
can cause a complete replan of a trajectory or<br />
burn sequence. Also, target state estimation updates are<br />
very similar, from the mission manager point of view, to<br />
target maneuvers.<br />
The chaser navigation software includes a dual inertial<br />
state extended Kalman filter. This means that the filter<br />
tracks and propagates the states of both the chaser and<br />
RSO. During far field rendezvous, the chaser inertial state<br />
is updated by the onboard GPS sensor, while the RSO state<br />
is updated by ground uplinks. Once inside communication<br />
range, relative GPS provides continuous relative RSO<br />
location information, which is used to update the filter’s<br />
inertial RSO state estimate. Navigation state updates and<br />
sensor acquisitions can cause the mission manager to execute<br />
replanning.<br />
The above discussion of the CONOPS has prepared us to<br />
list a set of objectives explicitly for the rendezvous portion<br />
of the PAMM mission. These will be enacted by the agent<br />
hierarchy as described in the sections on temporal and<br />
functional decomposition below.<br />
These mission objectives are simple and clear and may be<br />
stated as:<br />
1. Achieve the target offset point at the specified time<br />
using low thrust to 150 km and high-thrust burns<br />
thereafter.<br />
2. If #1 cannot be achieved (possibly due to low-thrust<br />
system failures outside 150 km, target maneuvers, or<br />
large target state estimation errors), achieve the target<br />
offset point at the specified time using the high-thrust<br />
system from the point of failure.<br />
3. If the target offset point cannot be achieved at the specified<br />
time, adjust the arrival time to an achievable value<br />
and continue the mission.<br />
4. If the target offset point cannot be achieved with<br />
onboard resources, abort the rendezvous and enter a<br />
circular orbit.<br />
5. If failures or resource limitations prevent any of the<br />
above, enter safe mode.
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
PAMM Mission Temporal Decomposition<br />
The temporal decomposition process starts with the<br />
CONOPS and the mission objectives. Once these are<br />
understood, a nominal mission timeline can be developed<br />
with the goal of identifying mission activities and<br />
their interface states. As a rule of thumb, activity boundaries<br />
are set whenever a new subsystem or guidance<br />
mode must be invoked to continue the mission. From<br />
the CONOPS then, the PAMM temporal decomposition<br />
will include a low-thrust activity and a high-thrust activity<br />
(Figure 7). During the low-thrust rendezvous, the<br />
mission manager invokes the low-thrust agent and<br />
assigns it a target state relative to the RSO. The lowthrust<br />
agent continuously uses the optimal guidance<br />
scheme described above to assess the vehicle’s capability<br />
to reach the low-thrust target state given current<br />
resources. The agent also monitors the low-thrust<br />
propulsion system for failure indications and resource<br />
usage. The low-thrust agent can correct for small errors<br />
that may accrue in the trajectory due to navigation errors<br />
or thrust control errors. If for some reason the low-thrust<br />
agent determines that the target state cannot be achieved<br />
from the current vehicle state, given the current<br />
resources (propellant) and health status, then it reports<br />
this failure to the mission activity agent. The mission<br />
activity agent may change the low-thrust target state to a<br />
state that is both reachable by the low-thrust system and<br />
from which a high-thrust rendezvous may be completed<br />
to the target point. The low-thrust target then, is the<br />
interface state between the two agents as described previously<br />
in the Hierarchical Planning section.<br />
The high-thrust rendezvous nominally consists of two<br />
burn events with coasting flight between the burns as<br />
described in the CONOPS section. The first burn raises<br />
the orbital apogee to the altitude of the desired RSO offset<br />
Low-Thrust<br />
Rendezvous<br />
High-Thrust<br />
Rendezvous<br />
Mission<br />
Figure 7. Temporal hierarchy for PAMM mission.<br />
point, while the second burn places the chaser in an orbit<br />
that is co-elliptic and co-altitude with the RSO. The second<br />
burn is timed to cause the chaser to arrive at the<br />
offset point at the desired time. As we shall see later,<br />
additional burns may be required to achieve the offset<br />
point if the high-thrust agent is invoked for any reason at<br />
a starting state other than close to the nominal low-thrust<br />
termination state.<br />
The high-thrust maneuvers include periods of thrusting<br />
and periods of nonthrusting attitude hold or “pointing.”<br />
In contrast, the segments of low-thrust maneuvering<br />
involve continuous thrusting while pointing of the space<br />
vehicle (and thrust vector) to effect the solution to the<br />
minimum-time rendezvous problem according to an<br />
adaption of the Kechechian solutions. Because the lowthrust<br />
trajectory employs a constant thrust approach, the<br />
minimum time solution is also the minimum fuel solution.<br />
To employ this low-thrust algorithm effectively, the<br />
mission manager must account for these periods of constrained<br />
attitude requirements in order to ensure the<br />
satisfaction of other mission constraints such as power<br />
generation, target sensing, and navigation. During the<br />
high-thrust rendezvous, the burns will be short and long<br />
periods will be spent in the pointing activity. The mission<br />
planner may schedule pointing to optimize solar power<br />
generation, improve thermal conditions, acquire the RSO<br />
with sensors, etc. Pointing activities could be planned<br />
onboard, scheduled on the ground, and tasked as mission<br />
objectives, or ground-planned activities could be<br />
treated as constraints to the mission manager. Since the<br />
focus for this activity was on the rendezvous problem,<br />
pointing and attitude hold activities were not implemented,<br />
and the mission manager was implemented in a<br />
3-degree-of-freedom (DOF) simulation. The implemented<br />
activity agents are shaded in Figure 7. Reference [3]<br />
Burn<br />
Burn Burn<br />
Thrust Point Height Point CoCo-<br />
Point Safe<br />
Raising<br />
ellipticelliptic Abort<br />
Proximity<br />
Operations<br />
Docking<br />
47
details many of the considerations for planning the attitude<br />
profile of a satellite during rendezvous.<br />
Once the activities have been identified for the nominal<br />
mission, failure and off-nominal conditions are considered<br />
to complete the problem decomposition. We<br />
already have the capability to handle many failure types<br />
since the mission activity agent can request the highthrust<br />
system to take over should low-thrust failures<br />
occur, and the low-thrust agent can replan the lowthrust<br />
profile should burn errors or navigation state<br />
updates require it. However, when low-thrust failures<br />
occur very early in the rendezvous, or if high-thrust failures<br />
prevent completion of the rendezvous, then it may<br />
not be possible to achieve the desired offset point. When<br />
this occurs, the mission agent may invoke an “abort”<br />
agent to place the spacecraft in a safe orbit co-elliptic<br />
with the RSO, or if this is not possible, to safe the spacecraft<br />
subsystems. Note that the abort agent utilizes the<br />
same burn co-elliptic agent used by the high-thrust<br />
agent to accomplish its safe orbit objective. Only the<br />
agents required to accomplish the current mission<br />
objectives are actually invoked by the mission agent and<br />
instantiated by the ADEPT framework, so the abort<br />
agent does not appear in the successful cases shown in<br />
the results section below.<br />
Once the desired offset point has been achieved, the<br />
PAMM spacecraft may conduct proximity operations,<br />
docking operations, or other RSO-relative maneuvers.<br />
The proximity operations and docking agents are not<br />
implemented here, but are displayed in the results section<br />
as placeholders for future development. Reference<br />
[3] describes some considerations and techniques for<br />
autonomous proximity operations and docking.<br />
PAMM Functional Decomposition<br />
Once the hierarchy of agents has been determined<br />
through the temporal decomposition process, we can<br />
begin the process of defining the required functionality<br />
of each agent by defining its monitoring, diagnosis,<br />
planning, and execution functions. For the sake of<br />
brevity, this section will describe only the more interesting<br />
functions associated with the shaded elements in<br />
Figure 7.<br />
Mission Agent Planning<br />
Beginning with the mission activity agent, we start by<br />
discussing planning, since the planning process actually<br />
creates and reconfigures the agent hierarchy. The mission<br />
agent planning function looks at the current relative navigation<br />
state of the chaser with respect to the RSO, along<br />
with subsystem state and status information. If subsystems<br />
are nominal and the altitude differential between<br />
the vehicles is large, then the mission agent will spawn a<br />
low-thrust child and assign it the objective of reaching a<br />
nominal target state below and behind the RSO (refer<br />
48<br />
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
again to Figure 6). The low-thrust child agent in turn<br />
invokes the optimal guidance algorithm to plan a trajectory<br />
to the assigned relative state. If low-thrust planning<br />
is successful, the mission agent creates a high-thrust<br />
child, assigns it the low-thrust termination state as an<br />
initial condition and the desired offset point as an objective.<br />
The high-thrust agent in turn calculates a series of<br />
burns to achieve the desired state by invoking the corresponding<br />
burn agents. If high-thrust planning is<br />
successful, then the mission agent terminates planning<br />
and begins the monitoring cycle.<br />
If low-thrust planning is unsuccessful, the mission agent<br />
attempts to accomplish the remaining rendezvous by<br />
assigning the high-thrust agent the current state as initial<br />
condition, rather than the predicted low-thrust<br />
termination state. As part of the high-thrust problem<br />
specification, the mission agent may also relax the<br />
arrival time requirement. If the high-thrust agent is successful,<br />
the mission is continued with high thrust only;<br />
if not, an abort is commanded by invoking the abort<br />
agent.<br />
Note that no assumptions were made about the initial<br />
vehicle conditions for the planning process. This means<br />
that the replanning process is exactly the same as the initial<br />
planning process, although the resulting plan – and<br />
corresponding agent hierarchy – may be different.<br />
Mission Agent Monitoring<br />
Once the rendezvous plan is complete, the mission agent<br />
begins to monitor subsystem status, as well as the status<br />
from all its executing children. The job of the monitoring<br />
function is to determine whether the mission objectives<br />
may be accomplished given the current plan and the current<br />
vehicle state and status. Two types of conditions may<br />
cause the mission agent to trigger a replan by executing its<br />
diagnosis function. First, a child agent (e.g., low thrust)<br />
may report that it can no longer achieve the objective<br />
assigned by the mission agent, and second, a subsystem<br />
failure may be detected that directly triggers the mission<br />
diagnosis function. The distinction between these two<br />
types of failure conditions becomes important in the diagnosis<br />
process.<br />
Mission Agent Diagnosis<br />
Diagnosis is executed whenever monitoring detects a<br />
condition that prevents achieving the mission objective<br />
with the current plan (set of activities). The job of diagnosis<br />
is to determine the reason for the failure of the<br />
plan in order to select a replanning strategy. If, for example,<br />
diagnosis is triggered because the low-thrust agent<br />
has determined (through its own monitoring and diagnosis<br />
process) that it can no longer reach its commanded<br />
state, then the mission agent may attempt to assign a<br />
new goal to the low-thrust child agent. This could occur<br />
for example if low-thrust propellant usage was larger
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
than expected, or if a navigation state update placed the<br />
RSO further out of the orbit plane than previously<br />
thought.<br />
If, on the other hand, the reason that diagnosis was triggered<br />
was due to a low-thrust engine failure, then the<br />
replanning strategy would be to attempt to use only the<br />
high-thrust system.<br />
Mission Agent Execution<br />
The execution portion of the mission agent, as for all nonleaf<br />
nodes, is simply to command the next activity when a<br />
child activity is complete. For the leaf nodes, the execution<br />
functions send configuration and desired state commands<br />
to GN&C and other vehicle subsystems.<br />
Low-Thrust Agent Planning<br />
The low-thrust planning algorithm uses a trajectory optimization<br />
technique initially developed by J.A.<br />
Kechechian. [4],[5] One of the keys to the simplifications<br />
accomplished in Kechechian’s low-thrust transfer algorithm<br />
is the formulation of the orbital differential<br />
equations in a polar coordinate frame that is nonsingular.<br />
The singularities in conventional orbital element<br />
approaches tend to frustrate optimization attempts.<br />
Likewise, Cartesian position-velocity formulations lead<br />
to highly nonlinear cost functions and inaccurate numerical<br />
integration due to the cyclical and rapidly varying<br />
characteristics of these state variables. Kechechian avoids<br />
these pitfalls by using nonsingular equinoctial orbital<br />
elements. The Northrop Grumman adaptation of this<br />
approach corrects several errors in the published<br />
Kechechian formulation and extends the technique to<br />
include optimization of mean anomaly (for rendezvous)<br />
along with the other five orbital elements optimized by<br />
Kechechian (for orbit transfer). The resulting nonlinear<br />
optimization solution outputs a time sequence of spacecraft<br />
attitude that accomplishes the desired rendezvous<br />
in minimum time given the thrust constraints of the EP<br />
system. This profile is then executed and monitored<br />
within the PAMM heirarchy.<br />
High-Thrust Agent Planning<br />
The high-thrust agent planning algorithm is responsible<br />
for generating a set of burns that causes the spacecraft<br />
to arrive at the desired RSO offset point at a specified<br />
time, given an initial condition at the termination or<br />
failure of low thrust. The approach to solving this problem<br />
is to generate a series of co-elliptic orbits that<br />
stair-step the chaser to the offset point (Figure 8). A coelliptic<br />
orbit keeps the chaser at an approximately<br />
constant altitude differential with the RSO. Since the<br />
altitude difference determines the rate of closure with<br />
the RSO, the time of arrival at the offset point may be<br />
controlled by varying the height of each co-elliptic, as<br />
well as the amount of time spent on each. Although a<br />
single co-elliptic could be used to control the arrival<br />
time, multiple co-elliptics are used to prevent long burn<br />
durations, as well as to cause the closure rate to diminish<br />
with relative range. The algorithm adds co-elliptic<br />
orbits until arrival time objectives and burn constraints<br />
are met.<br />
Altitude Difference (km)<br />
10<br />
0<br />
Low-thrust failure 15000 s prior to transition<br />
-10<br />
High-thrust burns<br />
-20<br />
-30<br />
-40<br />
Co-elliptic orbits<br />
-50<br />
-60<br />
-70<br />
-80<br />
Low-thrust failure<br />
-90<br />
0 500 1000 1500 2000 2500<br />
Downrange (km)<br />
Figure 8. Multiple co-elliptic trajectory.<br />
The chaser and RSO states are propagated using an<br />
Encke-Nystom integration scheme with J2, J3, and J4<br />
zonal gravitational coefficients. The Encke-Nystrom integration<br />
scheme takes into account the second-order<br />
nature of the equations of motion, requiring fewer function<br />
evaluations (only three for a fourth-order scheme).<br />
The advantage of the Encke-Nystrom integrators is that it<br />
allows large step sizes because it evaluates the trajectory<br />
along the Keplerian orbit and computes the perturbations<br />
from the Keplerian orbit. Thus, for low Earth orbit,<br />
step sizes on the order of 200 s can be taken while maintaining<br />
numerical precision.<br />
Burn Agent Planning and Monitoring<br />
In the process of generating the burn plan, the highthrust<br />
agent creates child burn agents of two types:<br />
height raising and co-elliptic. The height-raising agents<br />
are responsible for raising apogee to the height of the<br />
next co-elliptic, while the co-elliptic agents cause burns<br />
at a desired co-elliptic altitude to place the chaser at a<br />
constant altitude differential with the RSO. The final<br />
transfer to the Vbar is a 135-deg transfer to correct for<br />
out-of-plane errors that have built up during the final coelliptic<br />
orbit and to take advantage of improved<br />
navigation toward the end of the rendezvous sequence.<br />
The interface states between the burn agents are the<br />
states just prior to the next burn. The height-raising<br />
agent, for example plans to place the chaser at the<br />
desired altitude at the ignition time of the co-elliptic<br />
burn. Similarly, the co-elliptic agent plans to keep the<br />
chaser at the current altitude until the next height-raising<br />
burn.<br />
49
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
The type of the first burn in the sequence depends on the<br />
initial state. The high-thrust agent checks to see if the initial<br />
state is already close to a co-elliptic and, if so,<br />
schedules the first burn as a height-raising burn at some<br />
later time (Figure 8 is an example of this case). In this, the<br />
burn plan will include an even number of burns, since a<br />
pair of burns is required to transition from one co-elliptic<br />
to another. On the other hand, if the chaser is not initially<br />
in a co-elliptic orbit, an immediate burn will be scheduled<br />
to place it in one – resulting in an odd number of burns.<br />
When small navigation state updates occur, it is desirable<br />
to avoid replanning the entire burn sequence, so the active<br />
burn agent continuously monitors the error between the<br />
desired and predicted states at the next burn’s ignition<br />
time. If this error exceeds a threshold, the burn is<br />
replanned to achieve the desired altitude differential. Only<br />
if the burn magnitude exceeds the engine on-time constraint<br />
does the burn agent diagnoser deliver a failed status<br />
to the high-thrust agent, who initiates a replan of the<br />
entire burn schedule.<br />
PAMM Simulation<br />
The PAMM was implemented in a 2-body, 3-DOF<br />
Simulink-based simulation. Figure 9 shows the block<br />
diagram of the chaser vehicle in the simulation with<br />
ADEPT-based mission manager accepting state and status<br />
data from GN&C and subsystems and returning modes<br />
and commands. The ADEPT framework software is Clanguage<br />
code written in an object-oriented style to ease<br />
integration into real-time systems. The agent hierarchy<br />
and agent functions described above were implemented<br />
within the framework and interfaced to the simulation<br />
via a Simulink S-function. The ADEPT framework also<br />
supplies a developer interface (Figure 10) that displays<br />
current activities in green, future activities in blue, activities<br />
that are replanning in red, and deactivated, or<br />
“pruned” activities in grey.<br />
50<br />
Mission Objectives<br />
State<br />
and<br />
Status<br />
Telemetry_Chaser<br />
Telemetry_RSO<br />
Planet_Data<br />
Relative_State<br />
Telemetry_RSO<br />
PAMM<br />
GN&C<br />
Vehicle<br />
Models &<br />
Dynamics<br />
Mode<br />
and<br />
Commands<br />
Figure 9. PAMM chaser simulation.<br />
Telemetry_Chaser<br />
Figure 10. Agent hierarchy display.<br />
The GN&C block in the PAMM chaser vehicle included<br />
a dual-inertial state, nonlinear filter to provide realistic<br />
estimates of RSO and chaser inertial positions and velocities.<br />
The filter interfaced with the sensor models<br />
described in the CONOPS section. An RSO ground<br />
update model was provided to emulate the process<br />
whereby a ground station tracking the RSO would send<br />
the chaser an updated RSO state in order to better plan<br />
the rendezvous. The accuracy (covariance) of the RSO<br />
ground update was also uplinked to the chaser in order<br />
to allow the navigation filter to accurately process the<br />
measurement (Figure 11).<br />
4000<br />
2000<br />
0<br />
Chaser Nav Target Position Error (m)<br />
Target GPS Uplink<br />
-2000<br />
Relative GPS Acq<br />
-4000<br />
0 0.5 1 1.5 2 2.5 3 3.5<br />
4<br />
2<br />
0<br />
-2<br />
MissionAct<br />
MissionAct<br />
LowThrRend HighThrRend ProxOps Dock<br />
Burn Burn Burn Burn<br />
Chaser Nav Target Velocity Error (m/s)<br />
x10 4<br />
x10 4<br />
0 0.5 1 1.5 2 2.5 3 3.5<br />
x10<br />
Figure 11. Effects of target state uplink and relative GPS<br />
acquisition on target state error.<br />
The PAMM control laws were very simple as this was a 3-<br />
DOF implementation. Steering logic converted velocity<br />
change commands to thrust commands and sequencing<br />
logic timed the high-thrust burns. Low- and high-acceleration<br />
engines with representative thrust were included in<br />
the effector modeling. The dynamics included J2 gravity<br />
for both chaser and RSO bodies.<br />
The simulation also included graphical capability for visualization<br />
of the rendezvous geometry.<br />
4<br />
-4
Error (m)<br />
RESULTS<br />
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
The objective of this application was to demonstrate the<br />
mission manager response to two types of events: subsystem<br />
failures and navigation state updates. This section<br />
shows results from two example runs with low-thrust failures.<br />
Both cases also include replanning events from<br />
navigation state updates. For reference to the nominal<br />
case, see the CONOPS section. Some comments on the<br />
results of all the PAMM testing are included at the end of<br />
this section.<br />
Example 1: Early Failure of Low-Thrust System<br />
The intent of this example was to demonstrate the mission<br />
manager response to a low-thrust failure as well as a<br />
change in estimated target position. The plot in Figure 12<br />
is the in-plane relative position plot (orbital velocity points<br />
left) that summarizes the vehicle trajectory. The figure also<br />
shows the planner response to the failure. The top diagram<br />
Altitude Difference (km)<br />
10000<br />
Chaser Nav Target Position Error (m)<br />
0 0.5 1 1.5 2 2.5 3 3.5 4<br />
Chaser Nav Target Velocity Error (m/s) x104 -5000<br />
5<br />
Error (m/s)<br />
5000<br />
0<br />
0<br />
10<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
-60<br />
-70<br />
-80<br />
Low-thrust failure 15000 s prior to transition<br />
-90 0 500 1000 1500 2000 2500<br />
Downrange (km)<br />
Figure 12. Low-thrust failure triggers replan event.<br />
x10 4<br />
Relative<br />
GPS Acq<br />
-5<br />
0 0.5 1 1.5 2<br />
Time (s)<br />
2.5 3 3.5 4<br />
High-thrust burns<br />
Chaser position<br />
Low-thrust failure<br />
10<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
-60<br />
-70<br />
-80<br />
Relative<br />
GPS Acq<br />
Figure 13. GPS acquisition triggers replan event.<br />
shows the agent hierarchy at the moment that the mission<br />
agent’s diagnoser has triggered a replan event due to<br />
the low-thrust failure. The bottom diagram shows the<br />
state of the hierarchy immediately after the replan is<br />
complete. The blocks on the left (labeled MissionAct and<br />
LowThrRend) represent the agents that were executing<br />
just prior to the replan. These will be deactivated on the<br />
next pass of the PAMM. Since this is a complete failure of<br />
the low-thrust system, the mission agent spawns only a<br />
high-thrust child. Note that the high-thrust agent’s planner<br />
perceives the relative state at failure to be close to a<br />
co-elliptic trajectory. For this reason, no immediate burn<br />
is required. The final result is that four burns are planned<br />
to effect a double co-elliptic path to the desired offset<br />
point.<br />
Figure 13 (left plot) shows the chaser navigation estimate<br />
errors for the RSO inertial state. In this example, no<br />
MissionAct<br />
MissionAct<br />
LowThrRend HighThrRend ProxOps Dock<br />
Burn Burn<br />
-90<br />
0 500 1000 1500 2000 2500<br />
Downrange (km)<br />
MissionAct<br />
LowThrRend HighThrRend ProxOps Dock<br />
High-thrust burns<br />
Chaser position<br />
Burn Burn Burn Burn<br />
MissionAct<br />
MissionAct<br />
HighThrRend ProxOps Dock<br />
Burn Burn Burn<br />
MissionAct<br />
LowThrRend HighThrRend ProxOps Dock<br />
Burn Burn Burn Burn<br />
51
ground updates are received prior to relative GPS acquisition,<br />
so the errors have grown to over 5 km when relative<br />
GPS is acquired.<br />
Throughout this period, the monitor function within the<br />
height-raising burn agent is continuously predicting the<br />
relative state just prior to the final burn at the target offset<br />
point. The navigation state update causes this prediction<br />
to exceed an allowable threshold, so the height-raising<br />
agent replans the burn to achieve the offset point objective<br />
(Figure 13, center plot). Since the objective is achieved<br />
successfully by the new burn parameters, no replanning is<br />
required at any higher levels.<br />
The right side of Figure 13 shows the planner state after<br />
the replan. All prior activities are pruned and the current<br />
activities are displayed in green. The formerly<br />
active burn agent is displayed in gray. This run completes<br />
with the chaser arriving at the offset point at the<br />
desired time.<br />
Example 2: Late Failure of Low-Thrust System<br />
Figure 14 summarizes this case, which shows the effect<br />
of a low-thrust failure close to the targeted low-thrust<br />
termination point. The failure occurs with significant<br />
vertical relative velocity so an odd number of burns were<br />
planned – in this case three. The first burn placed the<br />
chaser in a co-elliptic orbit (see the left plot of the figure).<br />
The burn was timed to control the relative height of<br />
the co-elliptic in order to control the arrival time. The<br />
right side of Figure 14 shows the mission manager’s<br />
response to the replan. Note that the original two-burn<br />
solution was replaced by three burn agents.<br />
52<br />
Altitude Difference (km)<br />
10<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
-60<br />
-70<br />
-80<br />
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
Low-thrust failure 15000 s prior to transition<br />
Low-thrust failure<br />
Downrange (km)<br />
High-thrust burns<br />
Chaser position<br />
-90 0 500 1000 1500 2000 2500<br />
Relative GPS acquisition occurred just prior to the final<br />
height-raising burn (as in Figure 13) and, as in the previous<br />
example, the height-raising burn was replanned to<br />
arrive at the offset point at the required time.<br />
Additional Testing Results Comments<br />
The mission manager performance was tested for several<br />
dozen cases with low-thrust failures between 100 and<br />
2500 km behind the target on the final portion of the<br />
low-thrust rendezvous. In all these cases, the agent hierarchy<br />
was able to achieve a high-thrust burn solution to<br />
arrive at the offset point, although sometimes the time of<br />
arrival was not achievable.<br />
As shown in Figure 5, the nominal low-thrust trajectory<br />
“spiraled” outward toward the RSO, lapping it several<br />
times as the altitude difference was reduced. When lowthrust<br />
failures occurred outside the tested downrange<br />
window, the mission agent waited until the chaser was<br />
inside this window, and then activated the high-thrust<br />
agent. The success of the high-thrust replan depended on<br />
the phase angle at the time of failure and the total<br />
remaining altitude differential. When the high-thrust<br />
replan was not successful, the mission would have been<br />
aborted. No attempt was made to quantify the region of<br />
successful high-thrust replanning; this is left for future<br />
work.<br />
CONCLUSIONS<br />
Figure 14. Late low-thrust failure triggers replan event.<br />
A hierarchy of activity agents has been applied to solve a<br />
satellite rendezvous problem. The nominal rendezvous<br />
profile included a low-thrust and high-thrust phase. The<br />
MissionAct<br />
MissionAct<br />
LowThrRend HighThrRend ProxOps Dock<br />
Burn Burn<br />
MissionAct<br />
LowThrRend HighThrRend ProxOps Dock<br />
Burn Burn Burn
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
agent hierarchy was developed to handle failures of the<br />
low-thrust system, as well as updates to the chaser<br />
knowledge of the RSO state.<br />
The agent hierarchy was implemented within a software<br />
framework that instantiates the appropriate number and<br />
types of agents when planning or replanning occurred.<br />
The framework also provides handles to user-definable<br />
functions for the monitoring, diagnosis, planning, and<br />
execution the functions required of each agent.<br />
The agent hierarchy included low-thrust and high-thrust<br />
agents. The low-thrust agent utilized trajectory planning<br />
algorithms provided by Northrop Grumman Corp.,<br />
which generated an optimal trajectory to the vicinity of<br />
the RSO. From there, a high-thrust agent planned a series<br />
of near-impulsive burns to arrive at a desired RSO offset<br />
point at a desired time.<br />
The PAMM was implemented in simulation (Figure 15)<br />
and tested for low-thrust failures and navigation state<br />
updates. The agent hierarchy was able to predict the consequences<br />
of failures and/or changes in navigation states,<br />
diagnose problems with the current plan, replan a new<br />
set of activities when mission objectives were not met,<br />
and execute the new set of activities to accomplish the<br />
rendezvous.<br />
Figure 15. Visualization of PAMM satellite arriving at offset<br />
point.<br />
When failures or state updates occurred within the region<br />
described above, rendezvous replanning was successful.<br />
Outside this region, success depended on the phase angle<br />
and altitude differential at the time of the failure. If planning<br />
was not successful, a mission abort was commanded.<br />
Future work should include more testing to determine the<br />
region of a successful rendezvous capability for a set of<br />
spacecraft parameters. Additionally, the capability to handle<br />
other system failures (such as sensor failures) should<br />
be added and tested within the PAMM simulation.<br />
ACKNOWLEDGMENTS<br />
Portions of this work were supported by the Northrop<br />
Grumman Agile Space Vehicle Research and Development<br />
program and by the <strong>Draper</strong> ADEPT development team led<br />
by Richard Hildebrant. The authors also wish to acknowledge<br />
Jennifer Hamelin, who led the PAMM project through<br />
the concept development phase.<br />
REFERENCES<br />
[1] Boyd, J.R., “The Essence of Winning and Losing,” Excerpts in<br />
presentation format at http://www.defense-and-society.org/fcs/<br />
ppt/boyds_ooda_loop.ppt, January 1996.<br />
[2] Ricard, M. and S. Kolitz, “The ADEPT Framework for<br />
Intelligent Autonomy,” VKI Lecture Series on Intelligent Systems<br />
for Aeronautics, von Karman Institute, Brussels, Belgium, May<br />
2002.<br />
[3] Zimpfer, D., P. Spehar, F. Clark, C. D’Souza, M. Jackson,<br />
“Autonomous Rendezvous and Capture Guidance, Navigation,<br />
and Control,” 2005 Flight Mechanics Symposium, NASA<br />
Goddard Space Flight Center, October 2005.<br />
[4] Kechechian, J.A., “Trajectory Optimization Using Nonsingular<br />
Orbital Elements and True Longitude,” Journal of Guidance,<br />
Control, and Dynamics, Vol. 20, No. 5, September-October<br />
1997.<br />
[5] Kechechian, J.A., “Minimum-Time Constant Acceleration Orbit<br />
Transfer with First-Order Oblateness Effect,” Journal of<br />
Guidance, Control, and Dynamics, Vol. 23, No. 4, July-August<br />
2000.<br />
[6] Hu, H.C, T. Straube, J. Madsen, M. Ricard, “Shuttle Abort Flight<br />
Management (SAFM) – An Onboard Real-time Abort<br />
Determination and Assessment Application for Crew Safety and<br />
Awareness,” 2002 Core Technologies for Space Systems<br />
Conference, November 2002.<br />
[7] Jackson, M.C., T. Straube, T.J. Fill, S. Nemeth, “Onboard<br />
Determination of Shuttle Glide Capability for Shuttle Abort<br />
Flight Manager (SAFM),” 2002 Core Technologies for Space<br />
Systems Conference, November 2002.<br />
[8] Hu, H., R. Proud, J. Hart, “Spacecraft Mission Assessment and<br />
Replanning Tool (SMART) – a Real-Time, Intelligent<br />
Autonomous Flight Management (AFM) System for Increased<br />
Safety and Performance of Human Spaceflight Vehicles,” AIAA-<br />
2004-946, 42nd AIAA Aerospace Sciences Meeting and<br />
Exhibition, Reno, NV, January 5-8, 2004.<br />
[9] Abramson, M., M. Cleary, S. Kolitz, “Steps Toward Achieving<br />
Robust Autonomy,” Proceedings of the AAAI Spring Symposium,<br />
Stanford, CA, March 2001.<br />
53
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
[10] Adams, M.B., O.L. Deutsch, J.V. Harrison, “A Hierarchical<br />
Planner for Intelligent Systems,” Proceedings of SPIE - The<br />
International Society for Optical Engineering, Vol. 548,<br />
Arlington, VA, April 9-11, 1985.<br />
[11] Albus, J.S., R. Lumia, H. McCain, “Hierarchical Control of<br />
Intelligent Machines Applied to Space Station Telerobots,” IEEE<br />
Transactions on Aerospace and Electronic Systems, Vol. 24, No.<br />
5, September 1988.<br />
[12] Wade, M., “Kistler K-1,” article on the Web at http://www.astronautix.com/lvs/kislerk1.htm,<br />
2003.<br />
[13] Jackson, M. and R. McDonald, “<strong>Draper</strong> Simulation Analysis<br />
Tool (DSAT): Graphical Object Simulation Techniques and<br />
Tools for Simulink,” 2004 AIAA Conference on Simulation<br />
Tools and Techniques, August 2004.<br />
[14] Jackson, M. “Prototype Autonomous Mission Manager and<br />
Simulation,” The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc., July 25,<br />
2004.<br />
54
Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy<br />
(l-r) Christopher D’Souza and Mark Jackson<br />
Mark Jackson is a Principal Member of the Technical Staff in the Mission Design and Analysis Group of<br />
the Algorithms and Software Directorate. He is currently located at the <strong>Draper</strong> field site office at the<br />
Johnson Space Center in Houston, TX. He holds a BS in Systems Engineering from the U.S. Naval<br />
Academy (1982) and an MS in Aerospace Engineering from MIT (1994).<br />
Christopher D’Souza is currently employed in the Autonomous Flight Systems Office of the Aeroscience<br />
and Flight Mechanics Division of the Johnson Space Center. He is responsible for orbital analysis (rendezvous,<br />
cislunar, and lunar operations) for the CEV. Prior to that, he was employed at <strong>Draper</strong> <strong>Laboratory</strong><br />
from 1996 to 2005. His research areas and specialties are guidance and navigation systems for spacecraft<br />
and missiles as well as hybrid optimal control systems. Dr. D’Souza received BS and MS degrees from the<br />
University of Illinois (Urbana) and a PhD from the University of Texas in Austin, all in Aerospace<br />
Engineering.<br />
Hobson Lane is a Section Head within the Control Systems Department at Northrop Grumman Space<br />
Technology. At the time of the research described in this paper, he was the Principal Investigator of the<br />
Agile Space Vehicle Technology Development program at Northrop Grumman. More recently, he managed<br />
the Autonomous Walking Inspection and Maintenance Robot program under NASA’s exploration initiative,<br />
and he is currently the System Engineering Team Lead on a satellite laser communication system technology<br />
demonstration program. Mr. Lane received a double BS from Vanderbilt University in Physics and<br />
English (1993) and an MS degree from Georgia Tech (1996).<br />
55
Fluid Effects in Vibrating Micromachined Structures<br />
Peter Y. Kwok, Marc S. Weinberg, Kenneth S. Breuer<br />
Copyright © 2005 by The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc. Printed with permission in Journal of Microelectromechanical Systems,<br />
Vol. 14, No. 4, August 2005, pp. 770 – 781<br />
INTRODUCTION<br />
Squeeze-film damping and hydrodynamic lift for a micromechanical perforated proof mass are<br />
calculated and measured. This work has resulted in closed-form expressions that can be used to design<br />
accelerometers, tuning-fork gyroscopes, and other micromechanical devices. The fluid damping and<br />
lift are determined using finite-element analyses of the normalized and linearized governing equations<br />
where the boundary condition of the pressure relief holes is derived using pipe flow analysis. The<br />
rarefaction of gas is incorporated in the governing equations based on slip flow condition. As a further<br />
check, a one-dimensional network model is developed to account for the boundary condition of the<br />
holes on a tilted proof mass. Both closed-form and numerical solutions are compared against experimental<br />
data over a range of pressures.<br />
Fluid damping can be an important consideration in various<br />
microelectromechanical system (MEMS) designs.<br />
Typical motions in a MEMS device involve oscillations of a<br />
proof mass in the horizontal or vertical directions, and the<br />
fluid (gas) flow between the proof mass and the substrate<br />
can be described by classic Couette flow or Poiseuille<br />
flow, [1] respectively. When the mean free path of the fluid is<br />
comparable to the characteristic length of the MEMS<br />
device, the gas does not behave entirely as a continuum,<br />
but rather exhibits some of the characteristics associated<br />
56<br />
with discrete particles. This phenomenon is known as rarefaction,<br />
and it often occurs when the MEMS unit is placed<br />
in a sealed package at very low pressure in order to minimize<br />
the fluid damping. As the mean free path of the fluid<br />
increases, the continuum assumptions of fluid begin to<br />
break down, and the flow transitions into a slip flow regime<br />
and ultimately, a free molecular flow regime. The structural-fluid<br />
interaction is further complicated when pressure<br />
relief holes are present in the proof mass, resulting in flow<br />
both beneath and through the moving structure.
A micromechanical tuning-fork gyroscope (TFG) is used<br />
to study the effect of rarefied gas on micromachined structures.<br />
It consists of two proof masses with evenly<br />
distributed square holes, as shown in Figure 1. The proof<br />
masses are attached to anchors through elastic beams, and<br />
separated from the substrate by a small gap. The design<br />
parameters are depicted in Figure 2. The proof masses<br />
oscillate in the x direction when voltage is supplied at the<br />
outer combs. Any angular rate applied in the y-axis will<br />
cause the proof masses to also oscillate in the z direction<br />
due to Coriolis acceleration. By measuring the differential<br />
capacitance of the proof masses based on the vertical displacements,<br />
the angular rate can be determined.<br />
Figure 1. Photograph of a micromechanical TFG with the<br />
coordinate system defined.<br />
y<br />
b<br />
z<br />
y<br />
x<br />
Elastic Beam<br />
L<br />
x gc<br />
Lc<br />
wc<br />
Ltc<br />
Inner Combs<br />
Proof Mass Combs<br />
Base Beam<br />
Figure 2. Illustration of the TFG design parameters.<br />
Proof mass oscillation is damped by the fluid surrounding<br />
the structure. This is known as squeeze-film damping for<br />
the vertical oscillation since the gas film is squeezed by the<br />
vibrating plate. When the proof mass is oscillating horizontally,<br />
shear forces act on its surfaces. If the proof mass<br />
Fluid Effects in Vibrating Micromachined Structures<br />
Lp<br />
Lp<br />
Lh<br />
Anchor<br />
Outer Combs<br />
is unintentionally tilted during fabrication, hydrodynamic<br />
lift will be generated from the pressure built up underneath<br />
the proof mass as it moves horizontally. This is<br />
known as “surfboarding.”<br />
TFG performance is related to squeeze-film damping and<br />
possible hydrodynamic lift from surfboarding. High<br />
squeeze-film damping degrades gyro bias stability (lowfrequency<br />
variations of output with no actual input<br />
angular rate), and it also leads to high hydrodynamic lift,<br />
which is discussed in the “Lift from Surfboarding Effect”<br />
section. For some gyro mechanizations, good performance<br />
requires that the TFG response not lag the proof mass<br />
oscillation because of phase shifts induced by squeeze-film<br />
damping. In addition, the holes are important in determining<br />
bias stability since they typically reduce damping<br />
and hydrodynamic lift by about 100X (compared with no<br />
holes). Therefore, in order to improve the TFG performance,<br />
it is imperative to understand the detailed flow<br />
characteristics within the gyroscope unit.<br />
Squeeze-film motion and the surfboarding effect are common<br />
in many MEMS designs. Previous work [2]-[4] has been<br />
performed based on the Reynolds equation to analyze<br />
squeeze-film damping on a solid flat plate. The equation<br />
had been modified [5]-[7] to account for rarefaction using<br />
slip flow conditions. Similar approaches were also performed<br />
on flow analysis in small channels. [8],[9] References<br />
[10] and [11] are recent analyses of squeeze-film damping<br />
in perforated plates with finite thickness. The approach in<br />
Reference [10] is heavily numerically based and does not<br />
give a simple closed-form equation suitable for trade-off<br />
analyses for the squeeze damping. Reference [11] offers a<br />
closed-form approximation to the damping of a finite<br />
thickness, perforated plate. The derivation given in the<br />
“Closed-Form Approximation for Squeeze-Film Damping<br />
and Surfboarding Lift” section is a more direct, simpler<br />
alternative that is valid for most cases of practical interest.<br />
The derivation of Reference [11] is also discussed further<br />
in that section. Neither References [10] nor [11] consider<br />
rarefaction or hydrodynamic lift (surfboarding) effects.<br />
The hydrodynamic lift for a tilted solid flat plate has been<br />
evaluated based on Reynolds lubrication theory. [12] Since<br />
the pressure built up underneath the proof masses from<br />
squeeze-film motion and surfboarding is vented through<br />
the pressure relief holes in the proof masses, each hole<br />
serves as a “chimney” and generates additional damping<br />
resistance.<br />
The goal of this work is to develop easy-to-use, closedform<br />
solutions for squeeze-film damping and surfboarding<br />
lift for a MEMS design with holes on the proof mass. The<br />
numerical models of squeeze-film damping and surfboarding<br />
and the experimental results are used to validate the<br />
solutions. The derivation of closed-form solutions for<br />
squeeze-film damping and hydrodynamic lift with numerical<br />
and experimental verification at various pressure levels<br />
provides very simple and useful tools for trade-off studies<br />
57
for gyroscopes, oscillators, and other high mechanical<br />
quality factor devices. Our experimental data also offer<br />
valuable information that is not readily available in literature<br />
for comparison with various analytical models.<br />
This paper’s unique contributions include:<br />
1. A closed-form solution for the squeeze-film damping in<br />
a finite thickness perforated plate. This offers a simple<br />
alternative to the results of Reference [11].<br />
2. The previous work on thin plates assumed a roughly<br />
hexagonal configuration of circular holes. This paper<br />
uses a pattern of square holes and spacing.<br />
3. The model is extended to rarefied gases and to hydrodynamic<br />
lift, situations of practical importance to<br />
gyroscope design.<br />
4. Test results.<br />
5. Lumped parameter and finite-element verifications of<br />
closed-form models. The lumped parameter model<br />
offers an alternative to the finite-element method<br />
(FEM) and does not include the assumption that all<br />
flow in a cell goes through the nearest hole. This<br />
assumption is embedded in the Reynolds equation<br />
approach. [11],[13]<br />
This paper begins with the derivation of the governing<br />
equation for squeeze-film damping with rarefaction<br />
included, followed by the development of the chimney<br />
boundary condition in the following section. The governing<br />
equation is normalized, linearized, and solved using<br />
FEM. The significance of the various design parameters on<br />
squeeze-film damping is also illustrated.<br />
The third section begins with evaluating the shear damping<br />
based on a flat proof mass that is parallel to the<br />
substrate. Next, the surfboarding analysis is carried out<br />
based on the slider bearing equation [12] with rarefaction<br />
included. The equation is then solved numerically to estimate<br />
the upper and lower bounds of the hydrodynamic<br />
lift. A one-dimensional network model is developed to<br />
include the chimney boundary condition.<br />
In fourth section, the closed-form solution for squeezefilm<br />
damping is derived with comb teeth considered. The<br />
closed-form estimation of the hydrodynamic lift from the<br />
squeeze-film damping for small proof mass tilts is also<br />
illustrated.<br />
Lastly, an experimental validation is presented and the test<br />
data are compared with the analytical and computational<br />
results.<br />
Fluid Under the Effect of Squeeze-Film Damping<br />
The pressure gradient of gas between the proof mass and<br />
the substrate is related to the velocity profiles in the x and<br />
y directions when the proof mass is undergoing squeezefilm<br />
motion. The flow in the x direction can be described<br />
by Navier-Stokes equation [1]<br />
58<br />
Fluid Effects in Vibrating Micromachined Structures<br />
where<br />
p = pressure<br />
u = fluid velocity in the x direction<br />
µ = gas viscosity<br />
x,y,z = Cartesian coordinates defined in Figure 1.<br />
Equation (1) is valid under the assumption that (a) the<br />
gap height is always much smaller than the lateral extent<br />
of the proof mass plate; (b) the motion is sufficiently slow<br />
that the unsteadiness can be neglected; (c) the gas obeys<br />
the ideal gas law; and (d) the system is isothermal, with<br />
the density proportional to the pressure. Using a slip<br />
boundary condition [14] u = λ((∂u)/(∂z)) at z = 0 and<br />
u = –λ((∂u)/(∂z)) at z = h, where h is the gap height and<br />
λ is the mean free path of gas, the governing equation<br />
becomes<br />
or<br />
considering the flow in both x and y directions. [15]<br />
Equation (3) is the Reynolds equation with the rarefaction<br />
term, 6h 2λ. It is then normalized [16] according to the<br />
parameters listed in Table 1.<br />
(1)<br />
(2)<br />
(3)<br />
Table 1. Normalized Parameters for Squeeze-Film<br />
Equation.<br />
Parameter Symbol Normalization<br />
Normalized Pressure Ψ p/Pa<br />
Normalized Gap Height H h/ho<br />
Normalized Coordinate X, Y x/L, y/b<br />
Normalized Time T ωzτ<br />
where<br />
Pa = ambient pressure<br />
ho = initial gap height<br />
ωz = frequency of oscillation in z direction<br />
τ = time<br />
L,b = length and width of a proof mass<br />
In addition, Knudsen number (Kn) is defined as the ratio<br />
between λ and the characteristic length, h; and Kno is<br />
defined as the Knudsen number at the initial condition,<br />
i.e., λo/ho. The normalized governing equation becomes
where σ is the squeeze number, [3] (12µωzL2)/(Pah2 o).<br />
Equation (4) can be linearized using perturbation expansion.<br />
[2] Assuming the variation of gap height is small<br />
compared with its mean value and the variation in pressure<br />
is also small compared with the ambient pressure<br />
level, the normalized pressure and gap height can be<br />
expressed as<br />
(4)<br />
Ψ = 1 + εΨ^ (5)<br />
H = 1 + ε cos T (6)<br />
where ε = δ/ho and δ is the maximum vertical displacement<br />
of the proof mass. Note that Eq. (6) sets the proof<br />
mass at its highest position at T = 0, thereby begins<br />
squeezing initially. Substituting Eqs. (5) and (6) into Eq.<br />
(4), the equation of order ε becomes<br />
Then, assuming the solution is in the form [3]<br />
(7)<br />
Ψ^ = Ψ0 sin T + Ψ1 cos T (8)<br />
the time dependence can be eliminated<br />
(9)<br />
(10)<br />
One can consider the term, (σ/(1 + 6Kn)), to be an effective<br />
squeeze number with rarefaction included.<br />
Blech [3] has solved the coupled linearized squeeze film<br />
Eqs. (9) and (10) with Kn = 0 for a rectangular plate. The<br />
nondimensional damping (f0) and spring forces (f1) can be<br />
obtained through<br />
(11)<br />
(12)<br />
where A is the total area acted on by pressure. For a viscously-damped<br />
free vibration, [17] the damping<br />
coefficient, c, is determined by (f0 . Pa . A)/W, where W<br />
is the vertical velocity of the proof mass. The critical<br />
damping coefficient, Cc, is defined as 2mωn, where m is<br />
the mass and ωn is the natural frequency of the system.<br />
The quality factor, Qsqueeze, defined as Cc/2c, can be<br />
expressed as<br />
Fluid Effects in Vibrating Micromachined Structures<br />
(13)<br />
As the plate moves down, the fluid underneath the proof<br />
mass is forced out through the pressure relief holes. This<br />
“chimney” flow can be modeled as pipe flow, [18] with the<br />
pipe length being the thickness of the proof mass. The<br />
pressure gradient (∇Ψ0 and ∇Ψ1) is zero at the line of<br />
symmetry between the holes. The schematic of the chimney<br />
flow is shown in Figure 3. A circular cross section is<br />
assumed for simplicity. The entrance length of the flow is<br />
about 25% of the total chimney length (using a Reynolds<br />
number of order of 10 -1), and thus, the entrance length<br />
effect can be neglected as a first approximation. The governing<br />
equation of the fluid through the chimney is<br />
therefore<br />
(14)<br />
where r is the radial coordinate and R is the radius of the<br />
chimney.<br />
z<br />
Symmetry<br />
of Flow<br />
x<br />
t<br />
Ambient<br />
Pressure, Pa<br />
Proof Mass<br />
Figure 3. Schematic of “chimney” flow.<br />
Solving for the velocity profile with slip boundary condition<br />
u = –λ((∂u)/(∂r)) at r = R and (∂u)/(∂r) = 0 at r = 0,<br />
and integrating over the cross section to obtain the flow<br />
rate, the average flow velocity is then determined by dividing<br />
the flow rate by the cross sectional area, πR2 (15)<br />
From energy conservation, [18] the pressure drop along the<br />
chimney is balanced by the friction loss from the flow; also<br />
substituting Lh as the hydraulic diameter of a square cross<br />
section, [19] the pressure drop is given by<br />
(16)<br />
Again, after normalizing the pressure based on Table 1 and<br />
using the perturbation analysis, with Ψ = 1 + εΨ^, the<br />
equation for the pressure becomes<br />
Lp<br />
Lh<br />
h P<br />
w = dh<br />
dτ<br />
59
(17)<br />
Since the squeeze number is low, the flow can be assumed<br />
to be incompressible, and Vchimney can be represented<br />
based on conservation of mass (volume) by<br />
(18)<br />
where ωzδ is the proof mass velocity W. Substituting Eq.<br />
(18) into Eq. (17) and rearranging the terms, the normalized<br />
pressure at the chimney boundary condition can be<br />
derived as<br />
or<br />
(19)<br />
(20)<br />
Equation (20) shows that the boundary condition of the<br />
hole can be represented by the product of the squeeze<br />
number, a geometric term, and a rarefaction term. The<br />
ratios h/Lp and Lh/Lp are particularly important due to the<br />
exponential dependences on their value. A contour plot of<br />
log(Kgeometry) with various h/Lp and Lh/Lp ratios is shown<br />
in Figure 4.<br />
Figure 4. Contour plot of log(Kgeometry) with various h/Lp<br />
and Lh/Lp ratios.<br />
60<br />
Lh/Lp<br />
Log10(Kgeometry)<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9<br />
h/Lp<br />
Fluid Effects in Vibrating Micromachined Structures<br />
4<br />
3<br />
2<br />
1<br />
0<br />
-1<br />
-2<br />
-3<br />
With the chimney boundary condition defined, a numerical<br />
simulation is performed using PDEase, [20] a<br />
finite-element solver capable of solving field equations in<br />
an arbitrary two-dimensional domain. As mentioned previously,<br />
the geometric symmetry allows the problem to<br />
be defined only on a square “cell” domain with a pressure<br />
relief hole at the center. The total pressure can be<br />
obtained by multiplying the solution to the number of<br />
cells in the proof mass. Figure 5 illustrates the problem<br />
definition with the appropriate boundary conditions.<br />
Based on the dimensions of the design parameters for the<br />
TFG tabulated in Table 2, Kgeometry is 1.4 with h/Lp = 0.3<br />
and Lh/Lp = 0.45, and Krarefaction is 5.5e-5 with Pa = 5<br />
mTorr.<br />
∇Ψ^ = 0<br />
Kgeometry × Krarefaction × σ<br />
∇Ψ^ = 0<br />
∇Ψ^ = 0<br />
∇Ψ^ = 0<br />
Figure 5. Problem definition for squeeze-film damping<br />
with rarefaction effect.<br />
Table 2. Dimensions of the Design Parameters of the<br />
Selected TFG.<br />
Parameter Dimension (µm)<br />
Proof Mass Length (L) 450<br />
Proof Mass Width (b) 400<br />
Proof Mass Thickness (t) 10<br />
Gap Height (h) 3<br />
Hole Width (Lh) 4.5<br />
Hole Spacing (Lp) 10<br />
Combs Teeth Length (Ltc) 50<br />
Combs Overlap Length (Lc) 25<br />
Combs Teeth Width (wc) 2<br />
Combs Gap (gc) 3<br />
A solution contour of Ψ0 from the numerical simulation is<br />
shown in Figure 6. Since the simulation is only carried out<br />
on a single cell, the length in the squeeze number calculation<br />
would be using Lp (10 µm). The effective squeeze<br />
number with rarefaction included is 0.0304, based on Pa =<br />
5 mTorr, µ = 1.85e-5 N-s/m 2 at 300 K, ωz = 27 kHz × 2π,<br />
and Kn = 3000.
y<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
0 0.2 0.4 0.6 0.8 1<br />
x<br />
Figure 6. Solution contour of nondimensional damping<br />
pressure (Ψ0) with σ = 0.0304.<br />
Lift from Surfboarding Effect<br />
The proof masses will oscillate in the x direction when<br />
excited at their horizontal vibration mode frequency. Shear<br />
will be exerted by the fluid between the stationary substrate<br />
and the moving proof mass. The fluid flow between<br />
the proof mass and the substrate can be described by<br />
Couette flow, [1],[21] where the pressure is constant along<br />
the x direction. The shear damping force, fshear, acting on<br />
the proof mass under horizontal oscillation is<br />
(21)<br />
where U is the horizontal velocity of the proof mass (1.57<br />
m/s for the selected TFG).<br />
In addition, the oscillation motion also induces Couette<br />
flow in between the comb teeth. The average shearing area<br />
Acomb = Lct is used for the calculation. Neglecting the<br />
squeezing of the fluid by the tip of each comb at the end,<br />
the resulting shear is similar to Eq. (21). Hence, the total<br />
quality factor becomes<br />
(22)<br />
where nc is the number of comb teeth (nc = 80).<br />
It can be seen from Eq. (22) that as pressure becomes really<br />
low, in which λ >> h and λ >> gc, the quality factor can<br />
be approximated [9] by<br />
(23)<br />
The quality factor becomes independent of the gap<br />
heights; hence, the shear stress on the top surface should<br />
also be included.<br />
The pressure is constant along the x and y directions when<br />
the gap height is uniform across the proof mass. However,<br />
the pressure underneath the proof mass will build up and<br />
Fluid Effects in Vibrating Micromachined Structures<br />
generates lift if the proof mass is tilted at an angle, and<br />
such flow can be modeled by a stationary tilted plate with<br />
the substrate moving horizontally at velocity U, as depicted<br />
in Figure 7.<br />
z<br />
x<br />
Figure 7. Fluid flow model for surfboarding analysis.<br />
The governing equations for the fluid flow are the same as<br />
Eq. (1), in both x and y directions. Proceeding with slip<br />
boundary conditions: [5] u = U + λ((∂u)/(∂z)) and v =<br />
λ((∂v)/(∂z)) at z = 0 and u = –λ((∂u)/(∂z)) and v =<br />
–λ((∂v)/(∂z)) at z = h, and assuming the density is proportional<br />
to the pressure for an isothermal flow, the equation<br />
of motion becomes<br />
(24)<br />
This is simply the Reynolds equation for slider bearing<br />
with rarefaction effect. [22] The equation is then normalized<br />
using the same nondimensional parameters from<br />
Table 1<br />
X<br />
h1<br />
H 3<br />
Y H3<br />
X + 6H2 Kn X +<br />
Y + 6H2 Kn Y =<br />
(25)<br />
where Λ is the bearing number, (6µLU)/(Pah 2). The gap<br />
height is defined as<br />
and, therefore, the normalized H is<br />
( H)<br />
X<br />
(26)<br />
(27)<br />
where h1 and h2 are the gap heights at higher and lower<br />
end, respectively, and ε = (h1 – h2)/(h1).<br />
After substituting in the perturbation expansion using Ψ =<br />
1 + εΨ^ and neglecting higher-order terms, the normalized<br />
equation of order ε becomes<br />
L<br />
U<br />
h2<br />
(28)<br />
61
Equation (28) can be solved over the proof mass, however,<br />
the boundary conditions of the pressure relief holes<br />
are not uniform due to the varying gap heights, hence,<br />
the “cell simulation” approach may not provide the most<br />
accurate result. At the same time, evaluating the boundary<br />
condition on each hole and simulating the entire<br />
proof mass surface (1800 holes) may be impractical. As a<br />
result, FEM is used to provide the upper and lower<br />
bounds for the hydrodynamic lift, based on the results<br />
from proof mass with no holes and perfectly vented<br />
holes, respectively. The problem domain and definition<br />
are similar to those in Figure 5, except with governing<br />
Eq. (28). The upper bound estimate assumes a flat proof<br />
mass with no hole for pressure relief; and the lower<br />
bound estimate of hydrodynamic lift is determined with<br />
ambient pressure at the hole boundaries. The results are<br />
shown later in Figure 14.<br />
In addition, a one-dimensional network model is developed<br />
using Matlab [23] to account for the chimney effect<br />
and the varying gap heights. The network model consists<br />
of pressure nodes underneath the proof mass and the<br />
holes. Each pressure node is connected by “pressure resistances,”<br />
as shown in Figure 8.<br />
Figure 8. One-dimensional flow network for surfboarding<br />
analysis.<br />
For n holes, there are 2n+1 pressure nodes (Pi), n chimney<br />
resistances (Rc), and 2n+2 proof mass resistances<br />
62<br />
Pa<br />
Pa<br />
Lp<br />
Lh<br />
Rc Rc<br />
Pa<br />
P1 P 2 P3 P 4<br />
P5 P6 Pi-1 Pi Pa<br />
R1 R2 R3 R4 R5 R6 Ri Ri+1<br />
l<br />
Fluid Effects in Vibrating Micromachined Structures<br />
(Ri). l is defined as Lp – Lh. For tutorial purposes, the<br />
resistances are derived using a single “block” from the first<br />
column (width = Lp). From Kirchoff’s law [24]<br />
(29)<br />
with Pa set to zero. The driving flow rates, Q1 and Q2, are<br />
Q1 = (Lph1U)/(2) and Q2 = (Lph2U)/(2), respectively.<br />
Solving for Ri<br />
(30)<br />
with the modified viscosity to account for the rarefaction<br />
effect, [4] i.e.<br />
(31)<br />
The pressure resistance of chimney can be derived similarly<br />
as Eq. (16)<br />
(32)<br />
A set of equations can therefore be written based on conservation<br />
of mass<br />
(33)<br />
for n is odd (pressure node under the proof mass), and<br />
(34)<br />
for n is even (pressure node under the hole). Arranging the<br />
2n+1 equations into a matrix form, we have Eq. (35).<br />
(35)
Solving for the pressure nodes simultaneously, the hydrodynamic<br />
lift is obtained by integrating the pressure<br />
distribution over the area of the proof mass and multiplying<br />
the number of columns.<br />
To verify the network model, Lh is set to 0.1 µm and t is<br />
set to 1020 m, the effect of pressure relief holes is therefore<br />
minimized by the large chimney resistance. The network<br />
solution is then compared against the one-dimensional flat<br />
plate solution [25]<br />
(36)<br />
where h = h(x) as defined in Eq. (26).<br />
The comparison of the pressure distributions of a 450-µm<br />
flat plate is shown in Figure 9, with Kn = 17.5, and ε =<br />
0.0645 (h1 – h2 = 0.2 µm). The network solution with infinitely<br />
long (1020 m) and small (0.1 µm) chimney agrees<br />
well with the flat plate solution.<br />
Pressure (Pa)<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
0 100 200 300 400 500<br />
X (µm)<br />
Figure 9. Comparison of pressure distributions between<br />
analytical solution and network model with “infinite”<br />
chimney resistance.<br />
Closed-Form Approximation for Squeeze-Film<br />
Damping and Surfboarding Lift<br />
In this section, a closed-form solution for estimating the<br />
squeeze-film damping is derived. The rarefaction effect is<br />
embedded in the air viscosity term, µ. Consider the shaded<br />
trapezoid in Figure 10, which represents the fluid<br />
flow in a single cell. The flow passing through line x is<br />
given by<br />
Fluid Effects in Vibrating Micromachined Structures<br />
Analytical Solution for Flat Plate<br />
Network Model Solution<br />
(37)<br />
Symmetry Lines<br />
Figure 10. Illustration of fluid flow in a single cell under<br />
squeeze-film motion.<br />
With a cross section of h by x, the pressure gradient along<br />
the flow direction is<br />
Hence, the pressure at position x is<br />
(38)<br />
(39)<br />
The pressure drop from chimney can be approximated by<br />
fully developed circular pipe flow with Lh as the hydraulic<br />
diameter, [18] i.e.<br />
(40)<br />
The vertical damping force for the entire cell (8 trapezoids)<br />
is obtained by integrating the pressure under the proof mass<br />
(41)<br />
and the squeeze-film damping coefficient from the proof<br />
mass is therefore<br />
where η = Lp/Lh, nh = number of holes (nh = 1800).<br />
y<br />
Hole<br />
x<br />
Lh/2<br />
Lp/2<br />
x<br />
(42)<br />
63
However, when the damping calculated with chimneys<br />
exceeds that calculated from flat plate, the flat plate solution<br />
[3] should be used. In this limit, fluid is squeezing out<br />
from the sides rather than through the chimneys.<br />
The damping contributed by the combs is also estimated.<br />
Each comb tooth is assumed to be infinitely long so that<br />
the flow is in the short direction across the comb tooth.<br />
Using the overlap comb bottom surface area, Ltcwc/2, the<br />
flow in one gap is given by<br />
(43)<br />
The last term in Eq. (42) is the increase in damping<br />
caused by the finite length of the holes. Using the formula<br />
Eq. (38) for fully developed flow in a wide rectangular<br />
channel – the gap between the fixed and moving comb<br />
teeth – the pressure at the corner of the comb tooth is<br />
determined by<br />
(44)<br />
Assuming that Ltc >> wc, the damping coefficient from the<br />
comb teeth is<br />
(45)<br />
The first term in Eq. (45) represents the squeeze film<br />
beneath the comb teeth, and the second term represents<br />
squeezing the flow through the gaps between moving and<br />
fixed teeth. This gap occurs along only one half the tooth<br />
length. The total damping coefficient, cz, is therefore the<br />
sum of Eqs. (42) and (45).<br />
Next, the hydrodynamic lift is estimated based on the similarity<br />
between the squeeze film and surfboarding<br />
equations. To begin, incompressible gas is assumed so that<br />
density drops from the formulation, thus, the Reynolds<br />
equations are simplified and the ambient pressure for the<br />
boundary conditions can be set to zero. Since typical<br />
solved pressures are within 0.1% of the ambient pressure,<br />
the constant density assumption is acceptable.<br />
When the constant density assumption is applied, the<br />
pressure equation with both vertical motion and surfboarding<br />
for low Reynolds number laminar flow is<br />
64<br />
Fluid Effects in Vibrating Micromachined Structures<br />
(46)<br />
where U, V, and W are the proof mass velocities in x, y, and<br />
z directions, respectively.<br />
For a flat plate, the derivatives with respect to x and y are constants.<br />
For this study, ∂h/∂y = 0. When expanding to<br />
first-order terms, the parentheses on the left side become zero<br />
so that Eq. (24) results with the rarefaction condition added.<br />
The right-hand side represents constant driving terms.<br />
For the linear analysis, the pressures from squeeze-film<br />
damping, the response to W, and from surfboarding,<br />
response to U, are proportional. This result is important<br />
since one can rather easily envision the squeeze-damping<br />
process and the assumption that for a large plate, the<br />
flow from each cell (hole plus surrounding land) can be<br />
calculated individually.<br />
Because the equation for pressure, Eq. (46), is linear, the<br />
hydrodynamic lift can be calculated directly from the<br />
fluid-caused sense axis damping by<br />
(47)<br />
Applying this equation on the case from Table 2 and<br />
Figure 6 where P = 5 mTorr, µ = 1.85e-5 N-s/m 2 at 300<br />
K, ωz = 27 kHz × 2π, Kn = 3000, U = 1.57 m/s, and<br />
assuming the tilt, h1 – h2, is 0.2 µm, the hydrodynamic<br />
lift per hole determined from the network model is<br />
2.56e-12 N, and the closed-form solution is 2.31e-12<br />
N.<br />
In Reference [11], the Reynolds equation for flat plates is<br />
modified by including a term for the vertical flow per unit<br />
area. This term is developed by considering the flow in a<br />
closed, round cell with a vertical cylinder. Where<br />
Reference [11] continues onto the modified Reynolds<br />
equations, the closed-form approach herein uses the<br />
results of a square cell directly; thus, the mathematical<br />
derivation is simplified considerably.<br />
Experimentation and Result Comparison<br />
Experiments are performed on TFG samples with the<br />
dimensions listed in Table 2. The fluid damping from<br />
squeeze-film motion is obtained and compared with the<br />
analytical results in terms of quality factor. The lift force<br />
from surfboarding is also determined and compared<br />
with the numerical and closed-form predictions.<br />
A TFG sample unit is placed inside a bell jar with controlled<br />
pressure. The temperature of the TFG unit is<br />
monitored by a thermistor. The proof masses begin to<br />
oscillate vertically with excitation at their vertical resonant<br />
frequency. The decay response of the TFG is<br />
recorded in an oscilloscope after switching off the<br />
power supplied. A sample decaying signal is shown in<br />
Figure 11.
Output Voltage (V)<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
-0.2<br />
-0.4<br />
-0.6<br />
Figure 11. A decaying signal recorded at pressure = 200<br />
mTorr and temperature = 305 K.<br />
Each response has approximately 450 data points, and its<br />
envelope exhibits exponential decay in accordance to free<br />
vibration with viscous damping. The upper envelope of<br />
the decaying signal is extrapolated and the damping ratio<br />
is determined by curve fitting. The noise floor of the signal<br />
is eliminated to ensure more accurate curve fitting. The<br />
measurement and curve fitting procedure is carried out for<br />
a range of pressures at room temperature.<br />
Five sample units are tested at pressures ranging from 1<br />
mTorr to 1 Torr. The Knudsen number is determined<br />
based on the pressure and temperature recorded. The data<br />
show that as the pressure decreases, the fluid damping<br />
decreases and structural damping becomes dominant, and<br />
the quality factor eventually becomes the structural quality<br />
factor. The structural damping can be determined by<br />
curve-fitting the data. [15] -0.8<br />
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8<br />
Time (s)<br />
The fluid damping can be isolated<br />
from the total damping using<br />
(48)<br />
The quality factor is isolated from structural damping<br />
using Eq. (48), and a selected sample is plotted against the<br />
numerical results from finite-element analysis for a range<br />
of Knudsen number in Figure 12. It can be seen that the<br />
quality factor from fluid damping follows a linear trend<br />
with pressure. The quality factor measurements show reasonable<br />
agreement with the prediction from the<br />
finite-element and closed-form solutions. The average discrepancy<br />
between the measured data and the<br />
finite-element solution is 59%, and 39% for the closedform<br />
results. This is considered to be good agreement,<br />
since the solution spans several orders of magnitude. In<br />
fact, many calculations [26],[27] have shown a factor of four<br />
higher in predicting squeeze-film damping in other MEMS<br />
devices. These calculations neglected the pressure buildup<br />
at the holes, which further illustrates the significance of<br />
chimney resistance. The continuum flow model is also<br />
Fluid Effects in Vibrating Micromachined Structures<br />
1.00E+05<br />
1.00E+04<br />
1.00E+03<br />
Qsqueeze 1.00E+06<br />
1.00E+02<br />
1.00E+01<br />
Test Data<br />
Closed Form<br />
FEM<br />
Continuum<br />
1.00E+00<br />
0.01 0.1 1 10 100 1000 10000 100000<br />
Knudsen Number<br />
Figure 12. Comparison of measured quality factors with<br />
continuum and slip flow predictions for<br />
squeeze-film damping.<br />
plotted to illustrate the difference due to rarefaction. Note<br />
that the solutions merge as the Knudsen number<br />
approaches continuum regime. In addition, the comb<br />
teeth only contribute about 0.03% to the overall quality<br />
factor calculation, hence, not a significant source for<br />
squeeze-film damping.<br />
The experimental process is also carried out for horizontal<br />
oscillation. The decay response of the TFG is again<br />
recorded in an oscilloscope when the power supplied is<br />
switched off. The same post-processing method is used<br />
to evaluate the quality factor in the x-axis. Four sample<br />
TFG units are tested at pressures ranging from 1 mTorr<br />
to 1 Torr. The fluidic quality factor is isolated from the<br />
structural quality factor using Eq. (48).<br />
For the Knudsen number range considered, the shear<br />
areas A and Acombs should be doubled for the quality factor<br />
calculations as discussed previously. For illustration<br />
purposes, the quality factor curves from single-surface<br />
and double-surface calculations are shown in the log<br />
Qshear vs. log Kn plot in Figure 13. In single-surface calculation,<br />
only the bottom surface of the proof mass is<br />
used to calculate quality factor, whereas in double-surface<br />
calculation, both top and bottom surfaces of the<br />
proof mass are used as the effective area. As mentioned<br />
in Eq. (23), the shear on the top surface of the proof mass<br />
should also be included when the mean free path is<br />
much larger than the gap height, and the quality factor<br />
becomes independent of the gap heights.<br />
The quality factor measurements show reasonable agreement<br />
with the prediction from slip flow model. The<br />
average discrepancies between the data and analytical<br />
results are 58% and 29%, for single and double area calculations,<br />
respectively. As with the squeeze film results,<br />
we consider the agreement here to be excellent, considering<br />
the fact that the solution spans several orders of<br />
magnitude. Based on the design parameters of the selected<br />
TFG, the contribution to the total quality factor from the<br />
65
combs is about 39%. Thus, the combs shearing can be an<br />
important source of damping in horizontal oscillation.<br />
Again, the continuum flow model is also plotted to illustrate<br />
the difference due to rarefaction.<br />
For surfboarding, the experimental setup combines the<br />
horizontal and vertical oscillations. The TFG unit is<br />
placed inside the bell jar and oscillated at resonant horizontal<br />
frequency; at the same time, the vertical output<br />
signal is monitored. With no angular rate input, the lift<br />
force is determined from the sensor output, which<br />
includes the motor and sense coupling forces that are<br />
also in-phase with the bias readout. These other in-phase<br />
biases included in the output can cause the measured lift<br />
to be 2 to 4 times higher than the actual lift generated<br />
from the surfboarding effect. Unlike squeeze film and<br />
shear damping measurements, each TFG unit may have<br />
different tilt orientation, and thus, only a sample TFG<br />
unit with prominent bias output is tested and the measurement<br />
is plotted against the typical bias and tilt range<br />
observed. This TFG has the same dimensions as those<br />
listed in Table 2, except for the proof mass thickness,<br />
which is 8 µm instead of 10 µm.<br />
Figure 14 shows the comparison of surfboarding lift evaluated<br />
from measurements and the various solutions over<br />
a range of pressures (Kn). The typical bias of this TFG<br />
design measured at the sealed pressures of the production<br />
units ranges between 0.01 to 0.05 deg/s/mTorr. The<br />
typical hydrodynamic lift over the range of pressures<br />
(Kn) is then estimated by inverse pressure scaling, represented<br />
by the shaded region. As seen in the figure,<br />
measurements at very low pressure can be difficult as the<br />
instruments are approaching their resolution limits, thus,<br />
the data do not remain a constant slope at high Kn.<br />
Knowing that the typical differential tilt of the proof<br />
masses for this TFG design is about 0.2 to 0.3 µm, the<br />
measured data that are mostly above the shaded region<br />
indicate that this particular tested sample should have a<br />
tilt about 0.4 µm.<br />
66<br />
1.00E+08<br />
1.00E+07<br />
1.00E+06<br />
1.00E+05<br />
1.00E+04<br />
Quality Factor in X-Axis 1.00E+09<br />
1.00E+03<br />
Test Data<br />
Slip Flow (bottom surface)<br />
Slip Flow (both surfaces)<br />
Continuum Flow<br />
1.00E+02<br />
0.01 0.1 1 10 100 1000 10000 100000 1000000<br />
Knudsen Number<br />
Figure 13. Comparison of measured quality factors with<br />
continuum and slip flow predictions for horizontal<br />
shear damping.<br />
Fluid Effects in Vibrating Micromachined Structures<br />
Hydrodynamic Lift (N)<br />
1.00E+06<br />
1.00E+07<br />
1.00E+08<br />
1.00E+09<br />
1.00E+10<br />
1.00E+11<br />
1.00E+12<br />
1.00E+13<br />
Test Data<br />
Upper Bound<br />
Lower Bound<br />
Closed Form<br />
Network Model<br />
1.00E+14<br />
10 100 1000<br />
Knudsen Number<br />
10000 100000<br />
Figure 14. Comparison of measured surfboarding lift with<br />
various solutions over a range of pressures with<br />
h1 – h2 = 0.4 µm.<br />
To justify the approximation of the tilt, the upper and<br />
lower bounds are established based on the PDEase solutions<br />
with ε = 0.1333, or a gap height difference of 0.4 µm.<br />
The one-dimensional network model solution and the<br />
closed-form solution are also plotted based on the same<br />
tilt. The results show that the lift calculated from upper and<br />
lower bounds differs by 3 orders of magnitude, illustrating<br />
the significance of the chimney resistance. As mentioned,<br />
the lift determined from the bias readout also contains<br />
other in-phase bias elements, such as motor coupling force.<br />
The closed-form and network solutions show nice agreement,<br />
and they are within a factor of 2 after considering<br />
the influence of the motor coupling force. With the uncertainty<br />
in the measurement and in build parameters, the<br />
models describe the phenomena.<br />
Comparison with Molecular Flow Model<br />
It is important to note that differences in the velocity profiles<br />
can be found between our first-order slip model and<br />
other molecular flow models in the transitional and free<br />
molecular flow regimes, even for simple two parallel<br />
plates. [28] One way to obtain a better understanding is by<br />
performing a molecular dynamic simulation on a simple<br />
case with two parallel plates and comparing the results. In<br />
fact, for the two parallel-plate case, a continuum equation<br />
similar to Navier-Stokes can be formed that is applicable<br />
for the whole Knudsen regime, however, with modified<br />
viscosity and boundary models.<br />
Another way is to compare the shear stress per unit velocity<br />
for two parallel plates. This allows our model to be<br />
compared directly to a molecular flow model without a<br />
large effort. The comparison also addresses the issue that,<br />
although the velocity distributions may differ between the<br />
slip flow and molecular models, the agreement of the<br />
shear stresses in the molecular flow region is suitable for<br />
many engineering calculations.<br />
Recall Eq. (21), the shear stress per unit velocity for parallel<br />
plates is given by
(49)<br />
where µ is 1.9e-5 N-s/m2 for air at 310 K, λ is 5.3e-3 m at<br />
310 K and 10 mTorr, and h is 3e-6 m and omitted for large<br />
mean free path.<br />
Taken from Eq. (7) of Reference [28], which referred to<br />
Reference [29], the shear stress per relative plate velocity<br />
in the free molecular flow regime is given by<br />
(50)<br />
where ρ is 1.45 × 10-5 kg/m3 for air at 310 K and 10<br />
mTorr, M is 28 gm/gm-mole, and Ru is 8.31 N-m/gmmole-K.<br />
The various models for the shear stress cited in<br />
Eqs. (8)-(11) of Reference [28] all converge to Eq. (50) for<br />
high Knudsen numbers.<br />
In the limit of large mean free path, both Eqs. (49) and<br />
(50) are proportional to pressure. If one assumes that viscosity<br />
is proportional to the square root of temperature,<br />
both Eqs. (49) and (50) are inversely proportional to the<br />
square root of temperature. For simple slip flow, Eq. (49),<br />
the stress per relative velocity is 0.0018 N-s/m3; for the<br />
idealized molecular flow, Eq. (50), the stress per relative<br />
velocity is 0.0035 N-s/m3, based on the parameters of air<br />
at T = 310 and P = mTorr. Calculated by greatly different<br />
techniques and noting that the velocity distributions<br />
differ, [28] the two shear stress results differ by less than a<br />
factor of two. This calculation corroborates our experimental<br />
result that the slip flow model gives usable<br />
engineering estimates for damping over a wide pressure<br />
range, even into the molecular flow region.<br />
SUMMARY AND DISCUSSION<br />
Closed-form solutions for squeeze-film damping and surfboarding<br />
effects on a micromechanical TFG have been<br />
developed with rarefaction and perforation considerations.<br />
The solutions are in good agreement with numerical solutions<br />
and experimental data. The methodology can be<br />
applied to other micromachined structures such as<br />
accelerometers.<br />
Sample TFG units are tested for the squeeze-film damping<br />
and horizontal shear damping. The solutions show an<br />
average discrepancy within 40%, even though the flow<br />
belongs to a free molecular flow regime based on the<br />
Knudsen number. This is a very reasonable approximation<br />
considering that the range of quality factors covers few<br />
orders of magnitude.<br />
It is found that for the pressure range considered (1 Torr<br />
to 1 mTorr), the rarefaction effect could contribute up to<br />
four orders of magnitude increase in quality factor over the<br />
solution obtained with the continuum assumption. This is<br />
the same for shear damping from horizontal oscillation. In<br />
the case of TFG with an operating pressure ranging from 1<br />
Fluid Effects in Vibrating Micromachined Structures<br />
to 50 mTorr and a gap height of 3 µm, the Knudsen number<br />
is on the order of 10 3. Although the Knudsen number<br />
value indicates free molecular flow – far beyond the<br />
regime where this first-order slip flow model is valid, the<br />
squeeze-film damping and surfboarding agrees quite well<br />
with experiment. The agreement between theory and<br />
measurement over such a wide range of Kn is surprising<br />
and far better than one would expect. One must be cautious,<br />
and we emphasize that this does not imply that<br />
first-order rarefaction theory is uniformly valid at such low<br />
pressures. However, for this particular flow geometry, the<br />
agreement can be explained to some extent. If one<br />
expands the equations of motion to include higher-order<br />
corrections for rarefaction (e.g., the Burnett equations),<br />
one finds that the additional terms drop out for simple<br />
geometries such as parallel (or nearly parallel) plates. For<br />
this reason, the approximation used here is formally accurate<br />
at least to second order and perhaps even further. We<br />
also note that the better-than-expected agreement between<br />
theory and experiment at high Knudsen numbers has also<br />
been found in other related investigations. [13],[30],[31]<br />
Finally, a TFG unit with prominent bias was tested for<br />
surfboarding lift. The test results represent a bias roughly<br />
four times the contribution from the real surfboarding lift<br />
because of other in-phase effects such as motor coupling,<br />
and show that the amount of tilt is between 3 to 5 µm over<br />
the 450-µm proof mass length. Hydrodynamic lift from<br />
the numerical models and the closed-form solution are<br />
evaluated based on a 4-µm tilt. The closed-form and<br />
lumped parameter network solutions are able to predict<br />
the real surfboarding within a factor of 2.<br />
In summary, the closed-form solutions for squeeze-film<br />
damping and surfboarding lift agree with more complicated<br />
FEM and lumped parameter models and with measured data.<br />
ACKNOWLEDGMENT<br />
The authors would like to thank Mr. Greg Kirkos and Mr.<br />
Eli Weinberg for their contribution to the finite-element<br />
model and network flow model, and Mr. Richard Elliot for<br />
assisting with the experimental setup.<br />
REFERENCES<br />
[1] Fay, J.A., Introduction to Fluid Mechanics, MIT Press, 1994.<br />
[2] Griffin, W.S., H.H. Richardson, S. Yamanami, “A Study of Fluid<br />
Squeeze-Film Damping,” Journal of Basic Engineering, June<br />
1966, pp. 451-456.<br />
[3] Blech, J.J., “On Isothermal Squeeze Films,” Journal of<br />
Lubrication Technology, Vol. 105, October 1983, pp. 615-620.<br />
[4] Veijola, T., H. Kuisma, et al., “Equivalent-Circuit Model of the<br />
Squeeze Gas Film in a Silicon Accelerometer,” Sensors and<br />
Actuators A, Vol. A48, 1995, pp. 239-248.<br />
[5] Burgdorfer, A., “The Influence of the Molecular Mean Free Path<br />
on the Performance of Hydrodynamics Gas Lubricated Bearing,”<br />
Journal of Basic Engineering, Vol. 81, 1959, pp. 94-99.<br />
[6] Schaaf, S.A. and F.S. Sherman, “Skin Friction in Slip Flow,”<br />
Journal of Aeronautical Sciences, Vol. 21, No. 2, 1953, pp. 85-90.<br />
67
[7] Rohsenow, W. and H. Choi, Heat, Mass, and Momentum<br />
Transfer, Prentice-Hall, 1961.<br />
[8] Arkillic, E. and K.S. Breuer, “Gaseous Slip Flow in Long<br />
Microchannels,” J. Microelectromechanical Systems, Vol. 6, No.<br />
2, 1997, pp. 167-178.<br />
[9] Cho, Y-H., B.M. Kwak, A.P. Pisano, R.T. Howe, “Slide Film<br />
Damping in Laterally Driven Microstructures,” Presented at<br />
MEMS’99, Orlando, Florida, January 1999.<br />
[10] Yang, Y-J and C-J Yu, “Macromodel Extraction of Gas Damping<br />
Effects for Perforated Surfaces with Arbitrarily-Shaped<br />
Geometries,” Proc. MSM 2002, pp. 178-181.<br />
[11] Bao, M., H. Yang, et al., “Modified Reynolds Equation and<br />
Analytical Analysis of Squeeze-Film Air Damping of Perforated<br />
Structures,” Journal of Micromech. Microeng., Vol. 13, No. 6, 2003.<br />
[12] Gross, W.A., Gas Film Lubrication, John Wiley and Sons, 1962.<br />
[13] Veijola, T. and T. Mattila, “Compact Squeezed-Film Damping<br />
Model for Perforated Surface,” Proceedings of Transducers ‘01,<br />
Munchen, Germany, June 2001, pp. 1506-1509.<br />
[14] Arkilic, E.B., M.A. Schmidt, K.S. Breuer, “Gaseous Slip Flow in<br />
Long Microchannels,” Journal of Microelectromechanical<br />
Systems, Vol. 6, No. 2, June 1997.<br />
[15] Kwok, P., Fluid Effects in Vibrating Micromachined Structure,<br />
MS Thesis, Mechanical Engineering, MIT, Cambridge, MA,<br />
1999.<br />
[16] Langlois, W.E., “Isothermal Squeeze Films,” Quarterly Applied<br />
Mathematics, Vol. XX, No. 2, 1962, pp. 131-150.<br />
[17] Nashif, A.D., D.I. Jones, J.P. Henderson, Vibration Damping,<br />
John Wiley & Sons, 1985.<br />
[18] Munson, B.R., D.F. Young, and T.H. Okiishi, Fundamentals of<br />
Fluid Mechanics, John Wiley & Sons, 1994.<br />
[19] Potter, M.C. and D.C. Wiggert, Mechanics of Fluids, Prentice-<br />
Hall, Inc., 1991.<br />
[20] Macsyma Inc., PDEase2D Reference Manual, 3rd Edition,<br />
November 1996.<br />
68<br />
Fluid Effects in Vibrating Micromachined Structures<br />
[21] Ye, W., X. Wang, et al., “Air Damping in Laterally Oscillating<br />
Microresonators: A Numerical and Experimental Study,”<br />
JMEMS, Vol.12, No. 5, 2003, pp. 557-566.<br />
[22] Cercignani, C. and A. Daneri, “Flow of a Rarefied Gas Between<br />
Two Parallel Plates,” Journal of Applied Physics, Vol. 34, No. 12,<br />
December 1963, pp. 3509-3513.<br />
[23] MathWorks Inc., Matlab® Function Reference, Vol. 1-3, Version<br />
5, October 1997.<br />
[24] Standard Handbook for Mechanical Engineers, 7th Edition,<br />
Baumeister, T. and L.S. Marks, Editors, McGraw-Hill, New<br />
York, NY, 1967.<br />
[25] Sherman, F.S., Viscous Flow, McGraw-Hill Publishing<br />
Company, 1990.<br />
[26] Kim, E., Y. Cho, M. Kim, “Effect of Holes and Edges on the<br />
Squeeze-Film Damping of Perforated Micromechanical<br />
Structures,” MEMS’99, Orlando, FL, January, 1999.<br />
[27] Bhave, S.A., J.I. Seeger, et al., “An Integrated, Vertical-Drive, In-<br />
Plane-Sense Microgyroscope,” Transducers ‘03, Boston, MA,<br />
June 9-12.<br />
[28] Bahukudumbi, P., J.H. Park, A. Besok, “A Unified Engineering<br />
Model for Steady and Quasi-Steady Shear-Driven Gas<br />
Microflows,” J. Microscale Thermophysical Engineering, Vol. 7,<br />
2003, pp. 291-315.<br />
[29] Kogan, M.N., Rarefied Gas Dynamics, Plenum Press, New York,<br />
1969.<br />
[30] Arkilic, E., M.A. Schmidt, K.S. Breuer “Slip Flows and<br />
Tangential Momentum Accommodation in Micromachined<br />
Channels” Journal of Fluid Mechanics, Vol. 437, 2001, pp. 29-<br />
44.<br />
[31] Veijola, T., H. Kuisma, J. Lahdenpera, “The Influence of Gas-<br />
Surface Interaction on Gas-Film Damping in a Silicon<br />
Accelerometer,” Sensors and Actuators, A, Vol. A66, 1998, pp.<br />
83-92.
Fluid Effects in Vibrating Micromachined Structures<br />
Marc S. Weinberg<br />
Peter Y. Kwok was a <strong>Draper</strong> Fellow from 1998-99, after which he joined the Volpe National<br />
Transportation Systems Center. In 2001, he joined Foster-Miller, Waltham, MA as a Project Engineer,<br />
working on structural analysis. He is a Member of the American Society of Mechanical Engineers (ASME).<br />
He received an MS in Mechanical Engineering from MIT (1999).<br />
Marc S. Weinberg served in the United States Air Force Aeronautical System Division, Wright-Patterson<br />
AFB during 1974 and 1975, where he applied modern and classical control theory to design turbine<br />
engine controls, and at the Air Force Institute of Technology, where he taught gas dynamics and feedback<br />
control. At <strong>Draper</strong> <strong>Laboratory</strong>, he has been responsible for the design and testing of a wide range of micromechanical<br />
gyroscopes, accelerometers, hydrophones, microphones, angular displacement sensors, chemical<br />
sensors, and biomedical devices. He holds 24 patents with 12 additional in application. Dr. Weinberg<br />
has been a Member of ASME since 1971. He received BS, MS, and PhD degrees in Mechanical Engineering<br />
from MIT (1971, 1971, and 1974, respectively), where he also held a National Science Foundation<br />
Fellowship.<br />
Kenneth S. Breuer was in the Department of Aeronautics and Astronautics, MIT, from 1990 to 1999,<br />
before moving to Brown University where he is currently a Professor of Engineering. His research interests<br />
are in the fluid mechanics of micron and nanometer-scale systems. He is a member of ASME. Dr.<br />
Breuer received a PhD from MIT (1988).<br />
69
A Reusable Design for Precision Lunar Landing Systems<br />
Proceeding of the 2005 International Lunar Conference<br />
Toronto, Canada, September 2005<br />
Abramson, M.; Bryant, C.; Carr, F.; Kolitz, S.; Kahn, A.;<br />
Key, J.; Sakamoto, P.<br />
Robust Planning Under Uncertainty<br />
American Institute of Aeronautics and Astronautics<br />
(AIAA) Infotech@Aerospace, September 26-29, 2005,<br />
Crystal City, VA. Sponsored by: AIAA<br />
Abramson, M.; Carter, D.; Collins, B.; Kolitz, S.; Scheidler,<br />
P.; Strauss, C.<br />
Improving the Value of EO-1 Observations<br />
Proceedings of the 2005 Earth Science Technology<br />
Conference, College Park, MD, June 2005<br />
Adams, M.; Manobianco, J.; Bickford, J.; Manobianco, D.M.<br />
In Situ Atmospheric Profiling Using Mobile Ad Hoc<br />
Sensor Networks<br />
Nano Science and Technology Institute (NSTI)<br />
Nanotechnology Conference and Trade Show, Anaheim,<br />
CA, May 8-12, 2005. Sponsored by: NSTI<br />
70<br />
2005 Published Papers<br />
Artist’s Concept of the Lunar Surface Access Module (LSAM)<br />
Image courtesy of NASA/John Frassanito & Associates<br />
Adams, M.; Lepanto, J.; Hanson, M.<br />
Mixed Initiative Command and Control of<br />
Autonomous Air Vehicles<br />
AIAA Journal of Computing, Information, and<br />
Communication, Vol. 2, February 2005<br />
Ardini, J.<br />
A High-Performance Radix-2 FFT in ANSI-C for RTL<br />
Generation<br />
8 th Military and Aerospace Programmable Logic Devices<br />
(MAPLD), Washington, D.C., September 7-9, 2005.<br />
Sponsored by: MAPLD<br />
Ardini, J.<br />
Demand and Penalty-Based Resource Allocation for<br />
Reconfigurable Systems with Runtime Partitioning<br />
8 th Military and Aerospace Programmable Logic Devices<br />
Conference, Washington, D.C., September 7-9, 2005.<br />
Sponsored by: MAPLD
Barton, G.; Marinis, T.; Pryputniewicz, D.<br />
A Multichip Module Implementation of a Radiation-<br />
Tolerant Microprocessor for Space Applications<br />
32 nd International Microelectronics and Packaging<br />
Society (IMAPS) New England Annual Symposium and<br />
Exhibition, Boxborough, MA, May 17, 2005. Sponsored<br />
by: IMAPS<br />
Barton, G.; Marinis, T.; Pryputniewicz, D.<br />
A Multichip Module Implementation of a Radiation-<br />
Tolerant Microprocessor for Space Applications<br />
38 th International Symposium on Microelectronics,<br />
Boxboro, MA, September 25-29, 2005. Sponsored by:<br />
IMAPS<br />
Bedrossian, N.; Jang, J-W; Alaniz, A.; Johnson, M.;<br />
Sebelius, K.; Mesfin, Y.<br />
International Space Station U.S. GN&C Attitude<br />
Hold Controller Design for Orbiter Repair Maneuver<br />
Guidance, Navigation, and Control Conference and<br />
Exhibit, San Francisco, CA, August 15-18, 2005.<br />
Sponsored by: AIAA<br />
Bickford, J.; Adams, M.; Manobianco, J.<br />
GEMS: A Mobile Wireless Network for Atmospheric<br />
Sampling<br />
Defense and Security Symposium, Orlando, FL, March<br />
28-April 1, 2005. Sponsored by: International Society<br />
for Optical Engineering (SPIE)<br />
Bickford, J.; George, S.; Manobiance, J.; Adams, M.;<br />
Manobianco, D.<br />
Large-Scale Deployment and Operation of Distributed<br />
Sensor Assets Optimized for Robust Mars Exploration<br />
Evolvable Hardware Conference, Washington, D.C.,<br />
June 29-July 1, 2005. Sponsored by: National<br />
Aeronautics and Space Administration/Department of<br />
Defense (NASA/DoD)<br />
Bloom, I.K.<br />
Aliquots of Adhesives Presentation<br />
15 th Alternatives to Toxic Materials in Industrial<br />
Processes, Phoenix, AZ, January 10-13, 2005<br />
2005 Published Papers<br />
Borenstein, J.; Fiering, J.; Mescher, M.; Kujawa, S.; Sewell,<br />
W.; McKenna, M.<br />
Microsystems Technology for Long-Term Drug<br />
Delivery to the Inner Ear<br />
5 th International Symposium Meniere’s Disease and<br />
Inner Ear Homeostasis Disorders, Los Angeles, CA,<br />
April 2-5, 2005<br />
Brazzel, J.; Clark, F.; Spehar, P.<br />
RPOP Enhancements to Support the Space Shuttle R-<br />
Bar Pitch Maneuver for Tile Inspection<br />
Proceedings of the AIAA Guidance, Navigation, and<br />
Control Conference, August 2005<br />
Brown, R.<br />
Automating Space Operations Using Timeliner and<br />
Adept<br />
Enhancing Space Operations Workshop, Houston, TX,<br />
May 4-6, 2005. Sponsored by: AIAA<br />
Campbell, D.; Fill, T.; Hattis, P.; Tavan, S.<br />
An On-Board Mission Planning System to Facilitate<br />
Precision Airdrop<br />
Infotech@Aerospoace, Arlington, VA, AIAA-2005-7071,<br />
September 2005, pp. 26-29<br />
Carlen, E.; Heng, K-H; Bakshi, S.; Pareek, A.;<br />
Mastrangelo, C.<br />
High-Aspect Ratio Vertical Comb-Drive Actuator<br />
with Small Self-Aligned Finger Gaps<br />
Journal of Microelectromechanical Systems, Vol. 14, No.<br />
5, October 2005, pp. 1144-1155<br />
Carr, F.; Kolitz, S.; Lepanto, J.; Scheidler, P.; Wilde, J.;<br />
Smith, J.<br />
An Air Traffic System Simulation of the NAS with<br />
CNS and CD&R Models<br />
Modeling and Simulation Technologies, Conference and<br />
Exhibit, San Francisco, CA, August 15-18, 2005.<br />
Sponsored by: AIAA<br />
Carter, D.; George, S.; Hattis, P.; Singh, L.; Tavan, S.<br />
Autonomous Guidance, Navigation, and Control of<br />
Large Parafoils<br />
Aerodynamic Decelerator Systems Conference (ADSC),<br />
Munich, Germany, May 23-26, 2005<br />
71
Chen, Z.; Kujawa, S.; McKenna, M.; Fiering, J.; Mescher,<br />
M.; Borenstein, J.; Leary Swan, E.; Sewell, W.,<br />
Inner Ear Drug Delivery Via a Reciprocating<br />
Perfusion System in the Guinea Pig,<br />
J. Controlled Release, Vol. 110, No. 1, 2005, pp. 1-19.<br />
Cleary, M.; Dai, L.<br />
Situation Awareness Improvements for UUVs<br />
Unmanned Systems, North America, Baltimore, MD,<br />
June 28-30, 2005. Sponsored by: Association for<br />
Unmanned Vehicle Systems International (AUVSI)<br />
Colwell, W.<br />
Implementing Collocation Groups<br />
Facing the Future, Tivioli Storage Manager (TSM)<br />
Symposium, September 26-29, 2005. Sponsored by:<br />
TSM<br />
Coskren, D.; Easterly, T.; Polutchko, R.<br />
More Bang, Less Buck. Low-Cost GPS/INS Guidance<br />
for Navy Munitions Launches<br />
GPS World, September 2005, pp. 22-28<br />
Davis, C.<br />
Microfabricated Sensors for Clinical Diagnostic<br />
Applications<br />
Gordon Research Conference on Chemical Sensors and<br />
Interfacial Design, Oxford, England, August 28-<br />
September 2, 2005<br />
Diel, D.; Debitetto, P.<br />
Epipolar Constraints for Vision-Aided Inertial<br />
Navigation<br />
Computer Society Workshop on Motion and Video<br />
Computing, Breckenridge, CO, January 5-6, 2005.<br />
Sponsored by: IEEE<br />
Fidkowski, C.; Kaazempur-Mofrad, M.; Borenstein, J.;<br />
Vacanti, J.; Langer, R.; Wang, Y.<br />
Endothelialized Microvasculature Based on a<br />
Biodegradable Elastomer<br />
Tissue Engineering, Vol. 11, No. 1-2, January 2005, pp.<br />
302-309<br />
72<br />
2005 Published Papers<br />
Flueckiger, K.; Gustafson, D.; Dowdle, J.<br />
INS/GPS Deep Integration Navigation Hardware<br />
Testbed<br />
61 st Institute of Navigation (ION) Annual Meeting,<br />
Cambridge, MA, June 27-29, 2005. Sponsored by: ION<br />
Friend, S.; Monopoli, D.; Natoli, L.; Haas, D.<br />
SH-60B HUMS Experience Using a Satellite Data<br />
Link<br />
Aerospace Conference, Big Sky, MT, March 5-12, 2005.<br />
Sponsored by: IEEE<br />
Fuhrman, L.<br />
The NASA Human Exploration Program<br />
Aerospace Control and Guidance Systems Committee<br />
Meeting, Salt Lake City, UT March 2-4, 2005. Sponsored<br />
by: Society of Automotive Engineers (SAE).<br />
Fuhrman, L.; Fill, T.; Forest, L.; Norris, L.; Paschall II, S.;<br />
Tao,Y.C.<br />
A Reusable Design for Precision Lunar Landing<br />
Systems<br />
Proceeding of the 2005 International Lunar Conference,<br />
Toronto, Canada, September 2005<br />
Furis, M.; Cartwright, A.; Waldron, E.; Schubert, E.<br />
Spectral and Temporal Resolution of Recombination<br />
from Multiple Excitation States in Modulation-<br />
Doped AlGaN/GaN Multiple Quantum Well<br />
Heterostructures<br />
Applied Physics Letters, Vol. 86, No. 16, April 18, 2005<br />
George, S.; Carter, D.; Berland, J-C.; Dunker, S.; Tavan, S.;<br />
Barber, J.<br />
The Dragonfly 4,500-kg Class Guided Airdrop<br />
System<br />
Infotech@Aerospace, Arlington, VA, September 26-29,<br />
2005. Sponsored by: AIAA<br />
Guerra, C.; Page, L.<br />
Autonomous Planning for Spacecraft Rendezvous<br />
and Proximity Operations<br />
Infotech@Aerospace, Arlington, VA, September 26-29,<br />
2005. Sponsored by: AIAA
Gustafson, D.; Dowdle, J.; Monopoli, D.; Anszperger, J.<br />
King Air Demonstration of Deep Integration<br />
(KADDI)<br />
2005 Joint Navigation Conference, Orlando, FL, April<br />
2005<br />
Hall, R.; Barrington, R.; Kirchwey, K.; Alaniz, A.;<br />
Grigoriadis, K.<br />
Shuttle Stability and Control During the Orbiter<br />
Repair Maneuver<br />
3 rd International Energy Conversion, Engineering<br />
Conference, San Francisco, CA, August 15-18, 2005.<br />
Sponsored by: AIAA<br />
Hopkins, R.<br />
Inertial Navigation System Technology: A Short<br />
Course<br />
30 th Joint Navigation Conference, Orlando, FL, April<br />
11, 2005. Sponsored by: Joint Services Data Exchange<br />
(JSDE)<br />
Hopkins, R.; Miola, J.; Sawyer, W.; Setterlund, R.; Dow, B.<br />
The Silicon Oscillating Accelerometer: A High-<br />
Performance MEMS Accelerometer for Precision<br />
Navigation and Strategic Guidance Applications<br />
10 th ION National Technical Meeting, San Diego, CA,<br />
January 24-26, 2005. Sponsored by: ION<br />
Hopkins, R.; Miola, J.; Setterlund, R.; Dow, B.; Sawyer, W.<br />
The Silicon Oscillating Accelerometer: A High-<br />
Performance MEMS Accelerometer for Precision<br />
Navigation and Strategic Guidance Applications<br />
61 th ION Annual Meeting, Cambridge, MA, June 27-29,<br />
2005. Sponsored by: ION<br />
Huntington, G.; Rao, A.<br />
Optimal Reconfiguration of a Tetrahedral Formation<br />
via a Gauss Pseudospectral Method<br />
Proceedings of the 2005 Astrodynamics Specialist<br />
Conference, AAS Paper 05-338, Lake Tahoe, California,<br />
August 7-10, 2005<br />
Huntington, G.; Rao, A.<br />
Optimal Spacecraft Formation Configuration Using<br />
a Gauss Pseudospectral Method<br />
15 th Space Flight Mechanics Meeting, Copper<br />
Mountain, CO, January 23-27, 2005. Sponsored by:<br />
American Astronautic Society (AAS)/AIAA<br />
2005 Published Papers<br />
Huntington, G.; Benson, D.; Rao, A.<br />
Post-Optimality Evaluation and Analysis of a<br />
Formation Flying Problem via a Gauss Pseudospectral<br />
Method<br />
Proceedings of the 2005 Astrodynamics Specialist<br />
Conference, AAS Paper 05-339, Lake Tahoe, California,<br />
August 7-10, 2005<br />
Jackson, M.; D’Souza, C.; Lane, H.<br />
Autonomous Mission Management for Spacecraft<br />
Rendezvous Using an Agent Hierarchy<br />
Infotech@Aerospace. Arlington, VA, September 26-29,<br />
2005. Sponsored by: AIAA<br />
Kaya, T.; Lin, P.; Noubir, G.; Qian, W.<br />
Efficient Multicast Trees with Local Knowledge on<br />
Wireless ad hoc Networks<br />
Wireless Communications and Networking Conference<br />
(WCNC), New Orleans, LA, March 13-17, 2005.<br />
Sponsored by: IEEE<br />
Kaya, T.; Lin, P.; Noubir, G.; Qian, W.<br />
Efficient Multicast Trees with Local Knowledge on<br />
Wireless Ad Hoc Networks<br />
3 rd Wired/Wireless Internet Communications, International<br />
Conference, Xanthi, Greece, May 11-13, 2005<br />
Key, R.<br />
Routing in Stochastic Networks<br />
International Conference on Networking, Sensing, and<br />
Control, Tucson, AZ, March 19-22, 2005. Sponsored by:<br />
IEEE<br />
Khademhosseini, A.; Yeh, J.; Eng, G.; Karp, J.; Kaji, H.;<br />
Borenstein, J.; Farokhzad, O.; Langer, R.<br />
Cell Docking Inside Microwells Within Reversibly<br />
Sealed Microfluidic Channels for Fabricating<br />
Multiphenotype Cell Arrays<br />
Lab on a Chip, Vol. 5, No. 12, 2005, Royal Soc.<br />
Chemistry, Cambridge, England, pp. 1380-1386<br />
Kostas, T.; Lin, P.<br />
Flexible Satellite Communication Design<br />
Intelligence Community Chief Information Officer (IC<br />
CIO), Digital Convergence Conference, Chantilly, VA,<br />
May 16-17, 2005. Sponsored by: IC CIO<br />
73
Krebs M.; Tingley R.; Zeskind J.; Kang J.; Holmboe M.;<br />
Davis C.<br />
Autoregressive Modeling of Analytical Sensor Data<br />
Can Yield Classifiers in the Predictor Coefficient<br />
Parameter Space<br />
Bioinformatics, Vol. 21, No. 8, April 15, 2005, pp.<br />
1325-1331<br />
Krebs, M.; Zapata, A.; Nazarov, E.; Miller, R.; Costa, I.;<br />
Stonenshein, A.; Davis, C.<br />
Detection of Biological and Chemical Agents Using<br />
Differential Mobility Spectrometry (DMS)<br />
Technology<br />
IEEE Sensors Journal, Vol. 5, No. 4, August 2005, pp.<br />
696-703<br />
Kwok, P.; Weinberg, M.; Breuer, K.<br />
Fluid Effects in Vibrating Micromachined Structures<br />
Journal of Microelectromechanical Systems, Vol. 14, No.<br />
4, August 2005, pp. 770-781<br />
Malasky, J.; Forest, L; Kahn, A; Key, J.<br />
Experimental Evaluation of Human-Machine<br />
Collaboration in Resource Allocation and Planning<br />
for Multiple UAVs<br />
International Conference on Systems, Man, and<br />
Cybernetics, Hawaii, October 10-12, 2005. Sponsored<br />
by: IEEE<br />
Marinis, T.; Soucy, J.; Lawrence, J.; Owens, M.<br />
Wafer-Level Vacuum Packaging of MEMS Sensors<br />
55 th Electronic Components and Technology<br />
Conference, Lake Buena Vista, FL, May 31-June 3,<br />
2005. Sponsored by: IEEE/Components, Packaging, and<br />
Manufacturing Technology (CPMT) Society<br />
Mescher, M.; Lutwak, R.; Varghese, M.<br />
An Ultra-Low-Power Physics Package for a Chip-<br />
Scale Atomic Clock<br />
13 th International Conference on Solid-State Sensors,<br />
Actuators, and Microsystems (Transducers), Seoul,<br />
Korea, June 5-9, 2005. Sponsored by: IEEE<br />
Monopoli, D.; Natoli, L.; Haas, D.J.; Baker, T.<br />
JAHUMS ACTD Operational Experience<br />
Annual Forum 61, New Frontiers in Vertical Flight,<br />
Grapevine, TX, June1-3, 2005. Sponsored by: American<br />
Helicopter Society (AHS)<br />
74<br />
2005 Published Papers<br />
Parry, J.; Ricard, M.<br />
Vectors of Autonomy<br />
14 th Unmanned Untethered Submersible Technology,<br />
Durham, NH, August 21-24, 2005. Sponsored by:<br />
Autonomous Undersea Systems Institute (AUSI)<br />
Plump, J.; Ricard, M.; Keegan, M.; Jarriel, M.<br />
ONR’s Maritime Reconnaissance Demonstration –<br />
Simulation-Based Test and Integration<br />
Unmanned Systems North America, Baltimore, MD,<br />
June 28-30, 2005. Sponsored by: AUVSI<br />
Postma, B.; Jang, J.; Bedrossian; N.; Spanos, P.<br />
Robust Constrained Optimization Approach for<br />
International Space Station Centrifuge Rotor Auto-<br />
Balancing Controller<br />
3 rd International Energy Conversion, Engineering<br />
Conference, San Francisco, CA, August 15-18, 2005.<br />
Sponsored by: AIAA<br />
Proulx, R.; Carter, D.; Cefola, P.; Draim, J.; Inciari, R.;<br />
The Orbital Perturbation Environment for the<br />
COBRA and COBRA Teardrop Elliptical<br />
Constellations<br />
Journal of the Astronautical Sciences, Vol. 53, No. 2,<br />
April-June 2005, pp. 111-129<br />
Ricard, M.; Nervegna, M.; Keegan, M.; Jarriel, M.<br />
Autonomy and Control Architectures<br />
Unmanned Systems North America, Baltimore, MD,<br />
June 28-30, 2005. Sponsored by: AUVSI<br />
Rosch, G.; Hall, R.A.<br />
Stability of Angular Rate Damping for the NASA<br />
Space Shuttle<br />
3 rd International Energy Conversion, Engineering<br />
Conference, San Francisco, CA, August 15-18, 2005.<br />
Sponsored by: AIAA<br />
Ross, I.; D’Souza, C.<br />
Hybrid Optimal Control Framework for Mission<br />
Planning<br />
AIAA Journal of Guidance Control and Dynamics, Vol.<br />
28, No. 4, July-August 2005, pp. 686-697
Sawyer, W.; Prince, M.; Brown, G.<br />
SOI-Bonded Wafer Process for High-Precision MEMS<br />
Inertial Sensors<br />
Journal of Micromechanics and Microengineering, Vol.<br />
15, No. 8, pp. 1588-93<br />
Sherman, P.; Kourepenis, A.; Holmes, S.; Girolamo, H.;<br />
Zimmer, G.; Sokolowski, S.<br />
Personal Navigation for the Warfighter<br />
Joint Navigation Conference, Orlando, FL, April 11-15,<br />
2005. Sponsored by: DoD<br />
Shnayderman, M.; Mansfield, B.; Yip, P.; Clark, H.; Krebs,<br />
M.; Cohen, S.; Zeskind, J.; Ryan, E.; Dorkin, H.;<br />
Callahan, M.; Stair, T.; Gelfand, J.; Gill, C.; Hitt, B.;<br />
Davis, C.<br />
Species-Specific Bacteria Identification Using<br />
Differential Mobility Spectrometry and<br />
Bioinformatics Pattern Recognition<br />
Analytical Chemistry, Vol. 77, No. 18, August 12, 2005,<br />
pp. 5930-5937<br />
Singh, L.; Yang, L.; McConley, M.; Appleby, B.<br />
Autonomous Guidance and Control for Agile UAV<br />
Maneuvering<br />
2005 American Helicopter Society Forum 61,<br />
Grapevine, Texas June 1-3, 2005<br />
Singh, L.; Lapp, T.<br />
Model Predictive Control in Nap-of-Earth Flight<br />
Using Polynomial Control Basis Functions<br />
Proceedings of the 2005 American Control Conference,<br />
Vol. 2, pp. 840-5<br />
Stocker, D.; Waldron, E.; Boyle, J.; Kisler, Y.<br />
Neutron-Induced Degradation of CCD and CMOS<br />
Imager Optical Responsivity<br />
Hardened Electronics and Radiation Technology<br />
(HEART) Conference, Monterey, CA, March 1-5, 2005.<br />
Sponsored by: DoD/Department of Energy (DoE)<br />
Stolfi, M.; Negro, L.; Michel, J.; Duan, X.; LeBlanc, J.;<br />
Haavisto, J.; Kimerling, L.C.<br />
CMOS Compatible Erbium Coupled Si Nanocrystal<br />
Thin Films for Microphotonics<br />
Materials Research Society Symposium Proceedings, Vol.<br />
832, 2005, pp. F11.8<br />
2005 Published Papers<br />
Sullivan, M.; Bedrossian, N.; Nagarajaiah, S.<br />
State Estimation for International Space Station<br />
Centrifuge Rotor<br />
Guidance, Navigation, and Control Conference, San<br />
Francisco, CA, August 15-18, 2005. Sponsored by:<br />
AIAA<br />
Tetewsky, A.; Anszperger, J.<br />
Exactly What Does the GPS Satellite Transmit?<br />
61 st Institute of Navigation Annual Meeting, Cambridge,<br />
MA, June 27-29, 2005. Sponsored by: ION<br />
Tudryn, C.; Schweizer, S.; Hopkins, R.; Hobbs, L.; Garratt-<br />
Reed, A.<br />
Characterization of Si and CVD SiC to Glass Anodic<br />
Bonding Using TEM and STEM Analysis<br />
Journal of the Electrochemical Society, Vol. 152, No. 4,<br />
March 7, 2005, pp. E131-E134<br />
Underwood, J.; Poppe, D.; Weck de, O.<br />
Distributed Satellite Communications Systems:<br />
First-Order Interactions Between System and<br />
Network Architectures<br />
International Communications Satellite Systems<br />
Conference (ICSSC), Rome, Italy, September 25-28,<br />
2005. Sponsored by: ICSSC<br />
Van Beusekom, C.; Miotto, P.; Shepperd, S.<br />
Guidance and Control for Highly Constrained<br />
Rendezvous<br />
15 th Flight Mechanics Conference, Copper Mountain,<br />
CO, January 23-27, 2005. Sponsored by: AAS/AIAA<br />
Weinberg, E.J.; Kaazempur-Mofrad, M.R.,<br />
A Large-Strain Finite-Element Formulation for<br />
Biological Tissues with Application to Mitral Valve<br />
Leaflet Tissue Mechanics<br />
Journal of Biomechanics, July 20, 2005<br />
Weinberg, E.J.; Kaazempur-Mofrad, M.R.<br />
On the Constitutive Models for Heart Valve Leaflet<br />
Mechanics<br />
Cardiovascular Engineering, Vol. 5, No. 1, 2005, pp. 37-<br />
43<br />
75
Weinberg, M.; Wall, C., III<br />
Vestibular Prostheses for the Balance Impaired<br />
5 th International Symposium Meniere’s Disease and<br />
Inner Ear Homeostasis Disorders, Los Angeles, CA,<br />
April 2-5, 2005<br />
Weiss, L.; White, T.<br />
Autonomy Standards for Unmanned Undersea<br />
Vehicles<br />
ASTM Standardization News, November 2005, pp. 30-<br />
33.<br />
Wholey, L.; Singh, L.<br />
Automatic Low-Visibility Trajectory Optimization<br />
for Visually Identifying a Suspected Aircraft<br />
Guidance, Navigation, and Control, Conference and<br />
Exhibit, San Francisco, CA, August 15-18, 2005.<br />
Sponsored by: AIAA<br />
Willig, R.<br />
Solid-Core Laminar Photonic-Crystal Waveguide<br />
Optics East. Sensors and Applications, Boston, MA<br />
October 23-26, 2005, Sponsored by: SPIE<br />
76<br />
2005 Published Papers<br />
Zhu, M.; Perreault, D.; Caliskan, V.; Neugebauer, T.;<br />
Guttowski, S.; Kassakian, J.<br />
Design and Evaluation of Feedforward Active Ripple<br />
Filters<br />
IEEE Transactions on Power Electronics, Vol. 20, No. 2,<br />
March 2005, pp. 276-285<br />
Zimpfer, D.; Kachmar, P.; Tuohy, S.<br />
Autonomous Rendezvous, Capture, and In-Space<br />
Assembly: Past, Present, and Future<br />
1 st Space Exploration Conference: Continuing the<br />
Voyage of Discovery, Orlando, FL, January 30-February<br />
1, 2005. Sponsored by: AIAA<br />
Zimpfer, D.; Spehar, P.; Clark, F.; D’Souza, C.; Jackson, M.<br />
Autonomous Rendezvous and Capture Guidance,<br />
Navigation, and Control<br />
Flight Mechanics Symposium, Greenbelt, MD, October<br />
18-20, 2005. Sponsored by: NASA
Low-Cost, Low-Power Geolocation System:<br />
Patent Number 6,934,626<br />
<strong>Draper</strong> <strong>Laboratory</strong>’s enduring trademark is its ability to<br />
integrate widely diverse technical capabilities and technologies<br />
into innovative and creative solutions for<br />
problems of national importance. <strong>Draper</strong> scientists and<br />
engineers are encouraged in their quest to advance the<br />
application of science and technology, expand the functions<br />
of existing technologies, and to create new ones.<br />
The disclosure of inventions is an important step in documenting<br />
these creative efforts and is required under<br />
<strong>Laboratory</strong> contracts (and by an agreement with <strong>Draper</strong><br />
that all employees sign). <strong>Draper</strong> has an established patent<br />
policy and understands the value of patents in directing<br />
attention to individual accomplishments. Pursuing patent<br />
protection enables the <strong>Laboratory</strong> to pursue its strategic<br />
mission and to recognize its employees’ valuable contributions<br />
to advancing the state-of-the-art in their technical<br />
areas. An issued patent is also recognition by a critical<br />
third party (the U.S. Patent Office) of novel and creative<br />
work for which the inventor should be justly proud.<br />
Achromatic Fiber-Optic Power Splitter and Related Methods:<br />
Patent Number 6,959,131 B2<br />
Patents Introduction<br />
On average, <strong>Draper</strong>’s Patent Committee typically recommends<br />
seeking patent protection for 50 percent of the<br />
disclosures received. Millions of U.S. patents have been<br />
issued since the first patent in 1836. Through December<br />
31, 2005, 1252 <strong>Draper</strong> patent disclosures have been submitted<br />
to the Patent Committee since 1973; 638 of which<br />
were approved by <strong>Draper</strong>'s Patent Committee for further<br />
patent action. As of December 31, a total of 479 patents<br />
have been granted for inventions made by <strong>Draper</strong> personnel.<br />
Sixteen patents were issued for calendar year 2005.<br />
This year’s featured patent is:<br />
Optically Rebalanced Accelerometer<br />
Tuning-Fork Gyro:<br />
Patent Number 6,862,934 B2<br />
The following pages contain an overview of the technology<br />
covered in the patent, followed by the official patent<br />
abstract issued by the U.S. Patent Office.<br />
77
78<br />
Optically Rebalanced Accelerometer<br />
Patent No. 6,867,411 B2 Issued: March 15, 2005<br />
William Kelleher, Stephen Smith, Richard Stoner<br />
This invention is an all-optical accelerometer that uses radiation pressure to stabilize the position<br />
of a proof mass and a rebalance mechanism to measure acceleration. Instruments that can sense departures<br />
of their own reference frame from an inertial reference frame are used in many areas, including<br />
inertial navigation and guidance. Such departures include accelerations, which are commonly sensed<br />
by measuring either the displacement of a proof mass in response to an inertial force or by the restoring<br />
force necessary to restore the displacement of a proof mass.<br />
In the past, such sensors have been constructed from relatively large and expensive electromagnetic<br />
components. More recently, MEMS sensors have been fabricated from silicon wafers using techniques<br />
such as photolithography. Advantages of microfabricated sensors include small size and weight and<br />
the possibility of lower-cost large-scale production. There is now a growing manufacturing base and<br />
body of skilled workers in the fiber-optic communications industry as well as a large and growing<br />
infrastructure. An inertial sensor that uses only electro-optical components would thus share many<br />
subsystems and components with the fiber-optics industry and can be built economically.<br />
This innovative all-optical accelerometer could provide many advantages over prior art electromechanical<br />
inertial sensors. It has no moving wear surfaces: its projected lifetime is much greater than<br />
that of electromechanical accelerometers, since the lifetime of an all-optical accelerometer is limited<br />
only by the optical source lifetime. It may also be built as a flexureless and very linear instrument,<br />
eliminating the need for building flexural support structures into the device. Further, unlike prior-art<br />
MEMS sensors, it would be possible to recalibrate an all-optical inertial sensor during the operation<br />
of the device. An all-optical inertial sensor can be built as a closed-loop instrument with a high<br />
dynamic range, and using integrated optics and fiber-optic components, the space and energy requirements<br />
of the accelerometer can be minimized.<br />
Optically Rebalanced Accelerometer:<br />
Patent Number 6,867,411<br />
340<br />
310<br />
320<br />
330<br />
325
The Optically Rebalanced Accelerometer: U.S. Patent No. 6,867,411 B2<br />
79
80<br />
The Optically Rebalanced Accelerometer: U.S. Patent No. 6,867,411 B2<br />
(l-r) Stephen P. Smith, William P.<br />
Kelleher, and Richard E. Stoner<br />
William P. Kelleher joined the Electromagnetics Group at <strong>Draper</strong> <strong>Laboratory</strong> after working at Los Alamos<br />
National <strong>Laboratory</strong>. He has since served as group leader in electro-optics, SP23 IFOG Task Leader, and<br />
SP23 Sensor Development Technical Director. He is currently an Associate Division Leader in Guidance<br />
Hardware. Dr. Kelleher received a PhD in Nuclear Engineering from Rensselaer Polytechnic Institute.<br />
Stephen P. Smith developed a novel fiber-optic-based quench diagnostic for massive superconducting<br />
magnet systems to be used in Tokamak plasma fusion. He then worked as a post-doctoral researcher in<br />
the laboratory of Prof. Mara Prentiss at Harvard, where he was involved in a diverse array of research projects<br />
entailing biomedical applications of optical tweezers, laser cooling and trapping, and others. Except<br />
for a brief interlude, he has been employed at <strong>Draper</strong> since 1999, where he serves as Task Leader for the<br />
SP23 IFOG program as well as GBE-1 Group Leader. Mr. Smith received BS and ScD degrees in Electrical<br />
Engineering from MIT.<br />
Richard E. Stoner worked as a post-doctoral researcher and then a staff Scientist at the Harvard-<br />
Smithsonian Center for Astrophysics until June 2000. His work there comprised precision tests of fundamental<br />
symmetries using noble gas spin masers that he developed and constructed. He then joined<br />
<strong>Draper</strong>’s technical staff, where he has focused on the development of radiation-hard interferometric fiberoptical<br />
gyro (IFOG) technology, primarily in support of the SP23 program. He received the <strong>Draper</strong><br />
Distinguished Performance Award in 2002 for his work on precision radiation-hard scale-factor control for<br />
IFOGs. Dr. Stoner received a PhD in Physics from MIT.
Anderson, R.; Connelly, J.; Hanson, D.; Soucy, J.; Marinis, T.<br />
Integrated Sensor and Electronics Package<br />
Patent Number 6,891,239 B2, May 10, 2005<br />
Anderson, R.; Hanson, D.; Kasparian, F.; Marinis, T.;<br />
Soucy, J.<br />
Stress Isolation System<br />
Patent Number 6,936,479 B2, August 30, 2005<br />
Ash, M.; Martorana, R.<br />
Borehole Navigation System<br />
Patent Number 6,895,678 B2, May 24, 2005<br />
Ash, M.; DeBitetto, P.; Kourepenis, A.; Thorvaldsen, T.<br />
Compact Navigation System and Method<br />
Patent Number 6,918,186, July 19, 2005<br />
Bickford, J.; Petrovich, A.; Weinberg, M.<br />
Dual Microwave Cavity Accelerometer<br />
Patent Number 6,928,875, August 16, 2005<br />
Borenstein, J.; Sawyer, W.<br />
Method for Microfabricating Structures Using<br />
Silicon-on-Insulator Material<br />
Patent Number 6,946,314 B2, September 20, 2005<br />
Borenstein, J.; Connelly, J.; Cousens, J.; Duwel, A.;<br />
Kourepenis, A.; Lento, C.; Sawyer, W.; Weinberg, M.<br />
Tuning-Fork Gyro<br />
Patent Number 6,862,934 B2, March 8, 2005<br />
Bousquet, R.; Haley, H.; Hildebrant, E.; Martinez, S.;<br />
Ward, P.<br />
Charge Amplifier Device Having Fully Integrated<br />
DC Stabilization<br />
Patent Number 6,873,206, March 29, 2005<br />
Dual Microwave Cavity Accelerometer:<br />
Patent Number 6,928,875<br />
2005 Patents Issued<br />
Cunningham, B.; Williams, J.<br />
Flexural Plate Wave Sensor and Array<br />
Patent Number 6,837,097 B2, January 4, 2005<br />
Cunningham, B.; Williams, J.<br />
Flexural Plate Wave Sensor and Array<br />
Patent Number 6,851,297, February 8, 2005<br />
Dubé, C.; Petrovich, A.; Williams, J.<br />
Sensor Readout Circuit<br />
Patent Number 6,972,553, December 6, 2005<br />
Kelleher, W.; Smith, S.; Stoner, R.<br />
Optically Rebalanced Accelerometer<br />
Patent Number 6,867,411 B2, March 15, 2005<br />
Lane, P.; Tapalian, C.<br />
Thermo-Optical Switch Using Coated Microsphere<br />
Resonators<br />
Patent Number 6,934,436 B2, August 23, 2005<br />
Miller, R.; Nazarov, E.; Eiceman, G.; Krylov, E.<br />
Method and Apparatus for Electrospray Augmented<br />
High Field Asymmetric Ion Mobility Spectrometry<br />
Patent Number 6,972,407, December 6, 2005<br />
Tingley, R.<br />
Low-Cost, Low-Power Geolocation System<br />
Patent Number 6,934,626, August 23, 2005<br />
Willig, R.<br />
Achromatic Fiber-Optic Power Splitter and Related<br />
Methods<br />
Patent Number 6,959,131 B2, March 15, 2005<br />
Stress Isolation System (green section):<br />
Patent Number 6,936,479 B2<br />
81
Photo credit: National Academy of Engineering<br />
The NAE presented the 2006 Charles Stark <strong>Draper</strong> Prize to the inventors<br />
of charge-coupled devices, Willard S. Boyle and George E. Smith, on<br />
February 21 in Washington, D.C. The NAE’s citation reads, “for the invention<br />
of the Charge-Coupled Device (CCD), a light-sensitive component at<br />
the heart of digital cameras and other widely used imaging technologies.”<br />
Boyle and Smith share the $500,000 honorarium for their invention.<br />
CCDs, the first practical solid-state imaging devices, were invented in 1969<br />
by Boyle and Smith at Bell Laboratories. Since CCDs are silicon-based<br />
devices, they are relatively inexpensive to produce and are compact and<br />
fairly rugged, thus eminently suitable for commercial use. High sensitivity,<br />
excellent stability, and lack of distortion make them attractive for use in<br />
scientific research imaging systems. CCDs can image a variety of sources,<br />
such as optical, x-ray, ultraviolet, and infrared emissions. Currently, CCDs<br />
are used extensively in consumer products, including camcorders and<br />
cell phone cameras, and in such advanced electronic imaging tools as<br />
telescopes and imaging satellites.<br />
82<br />
Willard S. Boyle<br />
The 2006 Charles Stark <strong>Draper</strong> Prize<br />
The Charles Stark <strong>Draper</strong> Prize was established in 1988 to honor the<br />
memory of Dr. Charles Stark <strong>Draper</strong>, “the father of inertial navigation.”<br />
Awarded annually, the Prize was instituted by the National<br />
Academy of Engineering (NAE) and endowed by <strong>Draper</strong> <strong>Laboratory</strong>.<br />
It is recognized as one of the world’s preeminent awards for<br />
engineering achievement, honoring individuals who, like Dr. <strong>Draper</strong>,<br />
developed a unique concept that has contributed significantly to the<br />
advancement of science and technology and the welfare and freedom<br />
of society.<br />
For information on the prize, contact the Public Affairs Office<br />
at the National Academy of Engineering at (202) 334-1237 or<br />
http://www.nae.edu.<br />
Willard S. Boyle received a PhD in Physics at McGill University and worked for a year as a postdoctoral<br />
fellow in the Radiation <strong>Laboratory</strong>. After 2 years as an Assistant Professor at the Royal<br />
Military College, he joined the research staff of Bell Laboratories in 1953, and held positions in<br />
New Jersey, Washington, and Pennsylvania. He became Executive Director of Device<br />
Development and then Executive Director of the Communication Science Division until his<br />
retirement in 1979. His early work was on the electronic properties of semiconductors. He was<br />
the first to use infrared spectroscopy to measure donor energy levels and cyclotron energy levels.<br />
With Gene Kunzler, he discovered large, low-temperature magnetothermal oscillations, a powerful<br />
new tool for mapping out Fermi surfaces. With Donald Nelson, he developed the first<br />
continuously pumped ruby laser. He co-authored a major review paper with George E. Smith<br />
summarizing the unique properties of bismuth. Dr. Boyle is best known as the co-inventor, with<br />
George E. Smith, of the CCD. The first paper on the CCD was published in the Bell System<br />
Technical Journal on January 29, 1970. Dr. Boyle is a fellow of the American Physical Society and<br />
IEEE and a member of the National Academy of Engineering. He was awarded an Honorary<br />
Doctor of Laws from Dalhousie University. Since his retirement, he has served on the Research<br />
Council of the Canadian Institute of Advanced Research and the Science Council of the Province<br />
of Nova Scotia.<br />
Photo credit: National Academy of Engineering
The 2006 Charles Stark <strong>Draper</strong> Prize (cont.)<br />
George E. Smith received a BA in Physics from the University of Pennsylvania and MS and<br />
PhD degrees in Physics from the University of Chicago. He joined Bell Laboratories in<br />
1959,where he studied the electrical properties and band structures of semimetals, mostly<br />
bismuth and bismuth-antimony alloys, conducted microwave-resonance experiments<br />
and investigated magnetothermoelectric and galvanomagnetic effects. Heading the Device<br />
Concepts Department, he conducted investigations on junction lasers, semiconducting<br />
ferroelectrics, electroluminescence, transition-metal oxides, the silicon-diode-array<br />
camera tube, and CCDs. His major technical accomplishment was the development of the<br />
CCD with Willard S. Boyle, for which they have received numerous prestigious awards.<br />
They hold the basic patent (U.S. 3,858,232) and published the first paper disclosing the<br />
device concept, accompanied by a paper on its experimental verification in 1970. In April<br />
1986, he retired from Bell Laboratories as head of the VLSI Device Department, where he<br />
oversaw work on the physics of devices made with submicron lithography and their use<br />
in high-performance digital and analog circuits. Dr. Smith is a member of Pi Mu Epsilon,<br />
Phi Beta Kappa, Sigma Xi, and the National Academy of Engineering and is a fellow of<br />
the IEEE and American Physical Society. He holds 31 U.S. patents and has authored more<br />
than 40 papers. He received the IEEE Electron Devices Society Distinguished Service<br />
Award in 1997.<br />
Recipients of the Charles Stark <strong>Draper</strong> Prize<br />
George E. Smith<br />
2005: Minoru S. Araki, Francis J. Madden, Don H. Schoessler, Edward A. Miller, James W.<br />
Plummer for their invention of the Corona reconnaissance satellite technology.<br />
2004: Alan C. Kay, Butler W. Lampson, Robert W. Taylor, and Charles P. Thacker for the development<br />
of the Alto computer at Xerox’s Palo Alto Research Center (PARC).<br />
2003: Ivan A. Getting and Bradford W. Parkinson for their technological achievements in the<br />
development of the Global Positioning System.<br />
2002: Robert Langer for bioengineering revolutionary medical drug delivery systems.<br />
2001: Vinton Cerf, Robert Kahn, Leonard Kleinrock, and Lawrence Roberts for their individual<br />
contributions to the development of the Internet.<br />
1999: Charles Kao, Robert Maurer, and John MacChesney for spearheading advances in fiberoptic<br />
technology.<br />
1997: Vladimir Haensel for the development of the chemical engineering process of<br />
“Platforming” (short for Platinum Reforming), which was a platinum-based catalyst to<br />
efficiently convert petroleum into high-performance, cleaner-burning fuel.<br />
1995: John R. Pierce and Harold A. Rosen for their development of communication satellite<br />
technology.<br />
1993: John Backus for his development of FORTRAN, the first widely used, general-purpose,<br />
high-level computer language.<br />
1991: Sir Frank Whittle and Hans J.P. von Ohain for their independent development of the<br />
turbojet engine.<br />
1989: Jack S. Kilby and Robert N. Noyce for their independent development of the monolithic<br />
integrated circuit.<br />
83<br />
Photo credit: National Academy of Engineering
The CSAC program team developed the critical component<br />
for the world’s smallest and lowest power atomic clock –<br />
the physics package. This package, which uses MEMS<br />
technology, contains a VCSEL/photodiode, a mirror, a<br />
heater/sensor, and a cesium cell. The CSAC physics package<br />
achieves temperature control with ultra-low power by<br />
means of a novel thermal isolation mechanism (suspension<br />
tether system). The team demonstrated that the package<br />
consumes an order of magnitude lower power than the<br />
closest competing approach and two orders of magnitude<br />
lower power than commercially available atomic clock<br />
physics packages. This innovative design allowed <strong>Draper</strong><br />
to meet DARPA Phase III goals during Phase II.<br />
The development of signal processing algorithms for<br />
GPS in a weak signal environment addresses the unique<br />
challenges in evaluating the time and Doppler search<br />
space. Determining location using GPS depends on the<br />
ability to receive signals from GPS satellites and to measure<br />
the amount of time between the signal’s broadcast and its<br />
reception. Poor signal reception can compromise the accuracy<br />
of position determination. The team considered signal<br />
integration methods to extract the highest signal-to-noise<br />
ratio within the constraints imposed by the GPS signal<br />
structure. The algorithms they recently developed, implemented,<br />
and refined clearly enhance the effectiveness<br />
of the processing required for many real-world signal<br />
environments.<br />
84<br />
The <strong>Draper</strong> Distinguished Performance Awards<br />
Chairman of the Board Dr. John Kreick and President Vince Vitto presented the 2005 <strong>Draper</strong> Distinguished Performance<br />
Awards (DPAs) to two teams at the Annual Dinner of the Corporation on October 5. These awards recognize the development<br />
of a MEMS Physics Package for a Chip-Scale Atomic Clock (CSAC), performed by team members John Le Blanc, Mark<br />
Mescher, Gary Tepolt, and Mathew Varghese; and Signal Processing Algorithms for GPS in a Weak Signal Environment,<br />
performed by James Donna and James Scholten.<br />
(l-r) Mark Mescher, Mathew Varghese, John Le Blanc,<br />
and Gary Tepolt<br />
(l-r) James Scholten and James Donna<br />
Established in 1989, the Annual DPA is the most prestigious award that <strong>Draper</strong> can bestow to recognize extraordinary<br />
achievements by individuals or teams. These achievements must represent a high standard of excellence,<br />
provide significant benefit to the <strong>Laboratory</strong>, and be considered a major advance by the outside community. This<br />
year’s DPA Screening Committee was chaired by Heidi Perry and included James Bickford, Christopher Gibson,<br />
Roger Medeiros, David Owen, Donald Schwartz, Andrew Staugler, Heather Tahan, and Scott Uhland.
The Howard Musoff Student Mentoring Award<br />
The 2005 Howard Musoff Student Mentor Award was presented<br />
to Dr. Michael J. Ricard. In addition to his many<br />
responsibilities at <strong>Draper</strong>, he has served as thesis supervisor<br />
and mentor for nine <strong>Draper</strong> Fellows at both the<br />
master’s and PhD level. He has been a member of <strong>Draper</strong>’s<br />
technical staff since 1995. His work has focused on the<br />
design and development of software for unmanned,<br />
autonomous vehicles, and his specific research interests<br />
include path planning, mission planning, and other topics<br />
in applied combinatorial optimization. Dr. Ricard has<br />
broad experience in undersea programs. He is currently<br />
the Technical Director of the Maritime Reconnaissance<br />
Demonstration (MRD) and the Risk-Aware Mixed-<br />
Initiative Dynamic Replanning (RMDR) programs funded<br />
by the Office of Naval Research. He is also the Technical<br />
Director of <strong>Draper</strong>’s involvement with the US-UK<br />
Partnership demonstrating antisubmarine warfare (ASW)<br />
concepts with UUVs.<br />
Dr. Ricard has also been the Task Leader for the mission<br />
planner used on the Autonomous Minehunting and<br />
Mapping Technologies (AMMT) UUV sponsored by<br />
DARPA. He designed and implemented the real-time software<br />
that served as the high-level mission controller for<br />
the vehicle and was involved in the testing of that software<br />
Michael J. Ricard<br />
in both the simulation lab and in the field. He has also<br />
been the Principal Investigator on many joint research<br />
efforts with the Naval Undersea Warfare Center, Division<br />
Newport that extended the autonomous capabilities of<br />
UUVs. He holds a BS in Computer Science from Boston<br />
College (1988) and SM and PhD degrees in Operations<br />
Research from MIT (1992 and 1995, respectively).<br />
The Howard Musoff Mentor Award was established in<br />
2005. Musoff, who died in 2004, was a <strong>Draper</strong> employee<br />
for more than 40 years and advised and mentored<br />
many <strong>Draper</strong> Fellows. This award, given each February<br />
during National Engineers week, recognizes staff members<br />
who, like Musoff, share their expertise and<br />
supervise the professional development and research<br />
activities of <strong>Draper</strong> Fellows. The award is endowed by<br />
the Howard Musoff Charitable Foundation and includes<br />
a $1,000 honorarium and a plaque. Each Engineering<br />
Division Leader may submit one nomination of a staff<br />
person from his Division. The Education Office assists in<br />
the process by soliciting comments from students who<br />
were residents during that time period. The Selection<br />
Committee consists of the Vice President of Engineering,<br />
the Principal Director of Engineering, and the Director<br />
of Education.<br />
85
Agrawal, P.; Supervisors: Davis, C.; Belcher, A.<br />
Characterization of AP-MALDI and ESI for a<br />
Differential Mobility Spectrometer<br />
Master of Science Thesis, MIT, May 2005<br />
Akin, J.; Supervisors: Davis, C.; Irvine, D.<br />
Characterization of Human Skin Emanations by<br />
Solid Phase Microextraction (SPME) Extraction of<br />
Volatiles and Subsequent Analysis by Gas<br />
Chromatography-Mass Spectrometry (GC-MS)<br />
Master of Science Thesis, MIT, May 2005<br />
Becker, T.; Supervisors: Bottkol, M.; Schmidt, G.<br />
Approaches to Optimal Inertial Instrument<br />
Calibration Using Slewing<br />
Master of Science Thesis, MIT, June 2005<br />
Calhoun, P.; Supervisors: Hall, S.; White, D.<br />
Frequency Synthesis Using MEMS Piezoelectric<br />
Resonators<br />
Master of Science Thesis, MIT, February 2005<br />
Carr, K.; Supervisors: Keshava, N.; Greenberg, J.<br />
Detection of Contaminants Using a MEMS FAIMS<br />
Sensor<br />
Master of Science Thesis, MIT, May 2005<br />
Chen, D.; Supervisors: Lin, P.; Modiano, E.<br />
Minimum Energy Path Planning for Ad Hoc<br />
Networks<br />
Master of Science Thesis, MIT, September 2005<br />
Cooper, A.; Supervisors: Roy, N.; Gustafson, D.;<br />
McConley, M.<br />
A Comparison of Data Association Techniques for<br />
Simultaneous Localization and Mapping<br />
Master of Science Thesis, MIT, June 2005<br />
Diel, D.; Supervisors: DeBitetto, P.; Rowell, D.<br />
Stochastic Constraints for Vision-Aided Inertial<br />
Navigation<br />
Master of Science Thesis, MIT, January 2005<br />
Dobuzhskaya, M.; Supervisors: Brown, R.; Miller, R.<br />
Timeliner Integrated Development Environment<br />
Master of Engineering Thesis, MIT, June 2005<br />
86<br />
2005 Graduate Research Theses<br />
During 2005, the <strong>Draper</strong> Fellow Program consisted of approximately 60 students from MIT and several other universities.<br />
Abstracts of theses completed this year are available on the <strong>Laboratory</strong>’s web site at www.draper.com. The list of completed<br />
theses follows.<br />
Earnest, C.; Supervisors: Page, L.; Ricard, M.<br />
Dynamic Action Spaces for Autonomous Search<br />
Operations<br />
Master of Science Thesis, MIT, May 2005<br />
Gandhi, R.; Supervisors: Yang, L.; Roy, N.<br />
Examination of Planning Under Uncertainty<br />
Algorithms for Cooperative Unmanned Aerial<br />
Vehicles<br />
Master of Science Thesis, MIT, January 2005<br />
Hawkins, A.M.; Supervisors: Proulx, R.; Fill, T.; Feron, E.<br />
Constrained Trajectory Optimization of a Soft Lunar<br />
Landing from a Parking Orbit<br />
Master of Science Thesis, MIT, June 2005<br />
Hickie, M.; Supervisors: Homer, M.; Barnhart, C.; Appleby, B.<br />
Behavioral Representation of Military Tactics for<br />
Single-Vehicle Autonomous Rotorcraft via<br />
Statecharts<br />
Master of Science Thesis, MIT, June 2005<br />
Hickman, R.; Supervisors: Nervegna, M.; Tsitsiklis, J.<br />
Interception Algorithm for Autonomous Vehicles<br />
with Imperfect Information<br />
Master of Science Thesis, MIT, June 2005<br />
Hohreiter, L.; Supervisors: Duwel, A.; Chen, G.<br />
The Effects of Mechanical Coupling on the Electrical<br />
Impedance of MEMS Resonators for UHF Filter<br />
Applications<br />
Master of Science Thesis, MIT, January 2005<br />
Hyler, J.; Supervisors: Yu, C.; Madden, S.<br />
An Augmentation Algorithm for Improving<br />
Longevity in Ad Hoc Wireless Networks<br />
Master of Science Thesis, MIT, September 2005<br />
Liang, J.; Supervisor: Thordarson, P.<br />
The Application of the Value-Added Activity Model<br />
for the Mark-6 LE Integration Project<br />
Master of Engineering Thesis, MIT, June 2005<br />
Lommel, P.; Supervisors: Roy, N.; McConley, M.; Peraire, J.<br />
An Extended Kalman Filter Extension of the<br />
Augmented Markov Decision Process<br />
Master of Science Thesis, MIT, May 2005
MacLellan, S.; Supervisor: Staugler, A.<br />
Orbital Rendezvous Using an Augmented Lambert<br />
Guidance Scheme<br />
Master of Science Thesis, MIT, June 2005<br />
Malasky, J.; Supervisors: Barnhart, C.; Adams, M.<br />
Human Machine Collaborative Decision-Making in a<br />
Complex Optimization System<br />
Master of Science Thesis, MIT, June 2005<br />
McNurlen, A.; Supervisors: Magee, R.; Liskov, B.<br />
Petscope: A Standardized System for Ballistic<br />
Missile Guidance Data Analysis<br />
Master of Science Thesis, MIT, June 2005<br />
Merrick, W.; Supervisors: Davis, C.; Greenberg, J.<br />
Characterization of Human Expired Breath by Solid<br />
Phase Microextraction and Analysis Using Gas<br />
Chromatography-Mass Spectrometry and Differential<br />
Mobility Spectrometry<br />
Master of Engineering Thesis, MIT, September 2005<br />
Naresh, P.; Supervisors: Weinstein, W.; Shrobe, H.<br />
Distributing Network Identity to Mitigate Denial-of-<br />
Service Attacks<br />
Master of Engineering Thesis, MIT, August 2005<br />
Noonan, J.; Supervisors: White, D.; Dawson, J.<br />
Design of a High-Efficiency RF Power Amplifier for<br />
an MCM Process<br />
Master of Science Thesis, MIT, May 2005<br />
Oliver, M.; Supervisors: Harris, W.; Barrows, T.<br />
A Parametric Analysis of the Start-Up Procedure and<br />
Flight Characteristics of a Gliding Autogyro<br />
Master of Science Thesis, MIT, January 2005<br />
Postma, B.; Supervisor: Jang, J.W.<br />
Robust Constrained Optimization Approach to<br />
Control Design for International Space Station<br />
Centrifuge Rotor Auto Balancing Control System<br />
Master of Science Thesis, Rice University, April 2005<br />
Sakai, M.; Supervisors: Carter, D.; Tuller, H.<br />
Fabrication Process Changes for Performance<br />
Improvement of an RF MEMS Resonator:<br />
Conformable Contact Lithography, Moire Alignment,<br />
and Chlorine Dry Etching<br />
Master of Science Thesis, MIT, June 2005<br />
2005 Graduate Research Theses<br />
Stern, S.; Supervisors: Brown, R.; Berwick, R.<br />
An Extensible Object-Oriented Executor for the<br />
Timeliner User Interface Language<br />
Master of Science Thesis, MIT, May 2005<br />
Sullivan, M.; Supervisor: Bedrossian, N.<br />
State Estimation of International Space Station<br />
Centrifuge Rotor with Incomplete Knowledge of<br />
Disturbance Inputs<br />
Master of Science Thesis, Rice University, May 2005<br />
Thessin, R.N.; Supervisors: Bogner, A.; Herring, T.;<br />
Peraire, J.<br />
Atmospheric Signal Delay Affecting GPS<br />
Measurements Made by Space Vehicles During<br />
Launch, Orbit, and Reentry<br />
Master of Science Thesis, MIT, June 2005<br />
Thompson, G.; Supervisors: Hall, S.; Soltz, J.<br />
Inertial Measurement Unit Calibration Using Full<br />
Information Maximum Likelihood Optimal Filtering<br />
Master of Science Thesis, MIT, September 2005<br />
Torgerson, J.; Supervisors: Deyst, J.; George, S.<br />
Simulation and Control Design of a Gliding<br />
Autogyro for Precision Airdrop<br />
Master of Science Thesis, MIT, May 2005<br />
Underwood, J.; Supervisors: Poppe, D.; de Weck, O.<br />
Distributed Satellite Communication System<br />
Design: First-Order Interactions Between System<br />
and Network Architectures<br />
Master of Science Thesis, MIT, June 2005<br />
Van Beusekom, C.; Supervisors: Miotto, P.; Battin, R.<br />
A New Guidance Method for a DeltaV and Reentry<br />
Constrained Orbit Transfer Problem<br />
Master of Science Thesis, MIT, June 2005<br />
Warren, J., III; Supervisors: Perreault, D.; Neugebauer, T.<br />
Cell Modulated dc/dc Converter<br />
Master of Engineering Thesis, MIT, September 2005<br />
Weinberg, E.; Supervisors: Borenstein, J.; Mofrad, M.<br />
Dynamic Simulation of Heart Mitral Valve with<br />
Transversely Isotropic Material Model<br />
Master of Science Thesis, MIT, June 2005<br />
Wholey, L.; Supervisors: Singh, L.; Appleby, B.<br />
Trajectory Optimization with Detection Avoidance<br />
for Visually Identifying an Aircraft<br />
Master of Science Thesis, MIT, June 2005<br />
87
Each year, <strong>Draper</strong> <strong>Laboratory</strong> hosts a Technology<br />
Exposition (Tech Expo) to showcase recent projects and<br />
highlight <strong>Draper</strong>’s core competencies. The 2005 Tech<br />
Expo was held on October 5 and 6, which coincided with<br />
the fall meeting of <strong>Draper</strong>’s Board of Directors and the<br />
Annual Meeting of the Corporation. In addition to<br />
employees and Corporation members, guests included<br />
students from local universities and Cambridge public<br />
schools, as well as journalists and sponsors.<br />
The exhibits featured technologies under development in<br />
the <strong>Laboratory</strong>’s program areas: strategic, tactical, space<br />
systems, special operations, biomedical engineering, and<br />
independent research and development. The exhibits also<br />
reflected the <strong>Laboratory</strong>’s core competencies: guidance,<br />
navigation, and control; embedded, real-time software;<br />
microelectronics and packaging; autonomous systems;<br />
distributed systems; microelectromechanical systems; biomedical<br />
engineering; and prototyping system solutions. In<br />
coordination with <strong>Draper</strong>’s Education Office, a number of<br />
the projects also included graduate or undergraduate students<br />
on their teams.<br />
Tech Expo also featured <strong>Draper</strong>’s subsidiary venture capital<br />
firm, Navigator Technology Ventures, LLC (NTV). NTV<br />
displayed information about a number of its portfolio<br />
companies, which currently includes Actuality Systems,<br />
Aircuity, Assertive Design, Food Quality Sensor (FQS)<br />
International, HistoRx, Polnox Corp., Polychromix,<br />
Renalworks Medical Corp., Sionex Corp., and Tizor<br />
Systems.<br />
Dr. Angela Zapata describes technology under development in the<br />
Biomedical Engineering Group to local students.<br />
88<br />
2005 Technology Exposition<br />
<strong>Draper</strong>’s developing technology and systems for low-intensity conflict,<br />
surveillance, counterterrorism, and homeland defense are outlined by<br />
Dr. Paul Rosenstrach, Director of Special Operations (left).<br />
The Space Systems exhibit highlights <strong>Draper</strong>’s current focus on<br />
autonomous and highly reliable flight systems to meet the nation’s<br />
need for advanced space-based systems.<br />
David J. Carter describes advances in nanolithography and nanotechnology<br />
to <strong>Draper</strong> employee Andrea Prudente.
Inside Back Cover
Headquarters:<br />
The Charles Stark <strong>Draper</strong> <strong>Laboratory</strong>, Inc.<br />
555 Technology Square<br />
Cambridge, MA 02139-3563<br />
Phone: (617) 258-1000<br />
Fax: (617) 258-1131<br />
E-mail: techdigest@draper.com<br />
Business Development<br />
Phone: (617) 258-2124<br />
E-mail: busdev@draper.com<br />
Washington Area Offices:<br />
Suite 501<br />
1555 Wilson Boulevard<br />
Arlington, VA 22209<br />
Phone: (703) 243-2600<br />
Fax: (703) 528-5918<br />
www.draper.com<br />
#*