Professional Documents
Culture Documents
Engineering Innovations,
Research & Technology Report
April 2011
Lawrence Livermore National Laboratory
PO Box 808, L-151
Lawrence Livermore National Laboratory
Livermore, CA 94551-0808
http:www-eng.llnl.gov/ LLNL-TR-468271
Manuscript Date April 2011
Acknowledgments
Distribution Category UC-42
Scientific Editors
Don McNichols This report has been reproduced directly from the
Camille Minichino best copy available.
Available from
National Technical Information Service
Graphic Designers 5285 Port Royal Road
Jeffrey B. Bonivert
Springfield, VA 22161
Lucy C. Dobson
Debbie A. Ortega
Kathy J. Seibert
Or online at www-eng.llnl.gov/pubs.html
This work was performed under the auspices of the U.S. Department of Energy
by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.
ST-10-0070
FY10
Engineering Innovations,
Research & Technology Report
April 2011
Lawrence Livermore National Laboratory
LLNL-TR-468271
Introduction
A Message From
Monya A. Lane
Associate Director for Engineering
Introduction
A Message from Monya A. Lane ................................................................................................................................. ii
Engineering Innovations
Advanced Fuel Storage for Tomorrow's Clean Hydrogen Vehicles
Salvador M. Aceves .....................................................................................................................................................2
Computational Engineering
Direct Observation of Phase Transformations and Twinning Under Extreme Conditions:
In Situ Measurements at the Crystal Scale
Joel V. Bernier ............................................................................................................................................................32
Enabling Transparent Ceramic Optics with Nanostructured Materials Tailored in Three Dimensions
Joshua D. Kuntz .........................................................................................................................................................52
Extraction of White Blood Cells from Whole Blood Through Acoustic Focusing
Elizabeth K. Wheeler .................................................................................................................................................66
Measurement Technologies
Detection, Classification, and Estimation of Radioactive Contraband from
Uncertain, Low-Count Measurements
James V. Candy ..........................................................................................................................................................70
Low-Energy, Fast-Pulsed, Power-Driven Dense Plasma Focus for WCI and NIF Relevant Experiments
Vincent Tang ..............................................................................................................................................................74
Robust Ensemble Classifier Methods for Detection Problems with Unequal and Evolving Error Costs
Barry Y. Chen .............................................................................................................................................................86
Energy Manipulation
High Voltage Vacuum Insulator Flashover
Timothy L. Houck .......................................................................................................................................................94
Salvador M. Aceves
Insulated cryogenic fuel tanks could be used in mass-produced, (925) 422-0864
hydrogen-powered vehicles. aceves6@llnl.gov
Figure 3. The cryogenic pressure vessel installed onboard a Toyota Prius hydrogen hybrid vehicle, which set a world record for the
longest distance driven on one tank of hydrogen fuel.
Record-Setting Technology
The LLNL group put their cryogenic 40 gallons). The overall fuel economy versus about 3 cm in the previous
tank design to the test in 2007 and set was about 105 kilometers per kilogram design) while still maintaining adequate
a world record for the longest distance of hydrogen—equivalent to about dormancy (see Figure 4). The resulting
driven on one tank of hydrogen fuel, 65 miles per gallon of gasoline—when cryogenic pressure vessel is the most
albeit at reduced speeds and utiliz- driven at 40 to 56 kilometers (25 to compact automotive hydrogen storage
ing the trunk for storage space. The 35 miles) per hour (see Figure 3). vessel ever built, and is the only system
group drove a Toyota Prius hybrid A subsequent cryogenic pressure that meets DOE’s very challenging 2015
vehicle, modified to run on hydrogen, vessel design uses LLNL proprietary targets for weight and volume per-
1050 kilometers (653 miles) on one tank technology to shrink the insulation formance at higher storage capacities
of liquid hydrogen (150 liters, or almost thickness considerably (to about 1.5 cm (10 kg versus 5 kg).
Figure 4. The latest cryogenic pressure vessel prototype (right) is 23% more compact than the previous (left), a feat achieved
through reduced insulation thickness (1.5 cm versus 3 cm). This reduction in external volume suffices for meeting the very
challenging 2015 DOE weight and volume targets.
Evolution of iRadar
For more than a decade, Livermore
researchers have been working on
applying their patented ultrawideband
technology to the worldwide problem
of demining. In 1993, LLNL invented
the micropower impulse radar (MIR), a
technology that could be leveraged to
build low-cost, low-power, and compact
radar devices based on advances in
transient digitizing circuits needed for
LLNL laser programs. Two years later,
LLNL began two separate multiyear ef-
forts to develop ground-penetrating ra-
dar (GPR) systems using ultrawideband
radar technology: the High-performance
Electromagnetic Roadway Mapping and
Evaluation System (HERMES), designed
to identify structural deterioration in
bridge decks, and the Landmine Detec-
tion Advanced Radar Concept (LAND-
MARC) funded by the Department
of Defense. These efforts helped to
accelerate MIR development to where
it is today.
The LANDMARC system was devel-
oped as a small array system mounted
Figure 2. The Landmine Detection Advanced Radar Concept showed the feasibility of on a cart, as shown in Figure 2. This
discriminating buried landmines from innocuous clutter. portable system collected data at a
slower rate than the truck-mounted for post-processing, which could take reconstruction processing. This ability
HERMES array but was capable of col- several hours to produce a result and attracted several externally funded ef-
lecting data at much higher resolution. required human adjustment of model forts, which further advanced the GPR
Traditional GPR systems face difficulties parameters to match the dielectric con- effort.
detecting mines buried at or near the ditions and thicknesses of subsurface
surface of the ground. This is because soil layers. Today’s iRadar GPR Technology
the strong radar reflection from the Over the next several years, sub- In 2008, First Alliance Technologies,
surface, combined with stray reflections sequent developments in other re- LLC, entered into a cooperative research
from innocuous surface clutter, tends to lated technology projects significantly and development agreement (CRADA)
obscure the return signal from the hid- improved LLNL core capabilities in the with LLNL to develop an iRadar land-
den landmine. One of the key innova- area of ultrawideband radar imaging. mine detection system designed for
tions in using micropower impulse radar These developments included a super- deployment on an aerial platform. To
for this application was the ability to resolution imaging technique, advanced achieve rapid ground coverage, suf-
remove the effects of surface reflections time reversal reconstruction algorithms, ficient resolution, and superior imaging
through time-gating and better discrimi- model-based tomography schemes, a quality, a 16-element linear multi-
nation of rocks, roots, voids, and other self-contained digital synthetic aperture static array was developed, as shown
subsurface clutter from mines. radar imaging system that enabled elec- in Figure 3. This technology formed
While both efforts yielded promis- tronic beam-forming and beam-steering the initial technological basis of what
ing results, the prototype UWB GPR of the UWB signal, and advances in would later become the JIEDDO vehicle-
technology was not yet ready for opera- parallel computation technology to mounted iRadar IED detection platform.
tional use. One of the most significant create a dedicated real-time image The 2010 JIEDDO system, whose
hurdles was that collected data had to reconstruction circuit. This series of precursor is shown in Figure 4, consists
be processed offline, due to limitations technology projects enabled LLNL to ad- of a ruggedized multistatic iRadar array
on data acquisition speed, data stor- vance the ultrawideband radar imaging with military standard (MIL-SPEC) com-
age capacity, and processing speed. capability from a monostatic array with ponents and sophisticated positioning
HERMES and LANDMARC data were post-processing, to a digitally controlled, sensors. The JIEDDO project had similar
captured and stored on hard drives multistatic array with real-time image imaging requirements to previous
Figure 4. An early prototype of the JIEDDO vehicle-mounted IED detection platform scans for buried targets in a desert environment.
James V. Candy
SRaDS provides a fast and reliable radionuclide detection and (925) 422-8675
identification capability that can dramatically enhance the utility candy1@llnl.gov
of existing detection systems.
Figure 1. The statistical radiation detection system (SRaDS) is an innovative software solution that can be integrated easily into any
gamma-detection system to combat illicit trafficking of radioactive material through customs, border crossings, and limited-access areas.
SRaDS identifies radionuclides in low-count situations when measurement time is short and demand for reliability is high. The processed
data are displayed in intuitive plots showing results that a nontechnical user can interpret.
detection performance with high reli- • enhances the channel energy and the estimates of its energy, rate, and
ability and short decision times. rate (interarrival) parameters; emission probability and then used to
Each unique energy/arrival compo- • updates the corresponding decision update the decision function.
nent of the target radionuclide is pro- function; and Results of this photon-by-photon
cessed individually in a separate chan- • detects/identifies the target radio- processor with downscatter are shown
nel, resulting in the parallel/distributed nuclide by thresholding the decision in Figure 5. In this three-column figure,
processor structure. After the photon is function. the first column is the composite PHS
acquired, the distributed processor As diagrammed in Figure 4, the (which is not used). The second column
• discriminates the individual SRaDS processor consists of a discrimi- shows the measured photon energies
monoenergetic (single-energy) nator for both energy (amplitude) and (arrivals) as red dots. Circles (green)
arrival identifying one of the parallel rate (interarrival time). If the photon represent the discriminator output
channels; does not pass this test, it is sent to the photoelectrons, and squares (purple)
• discriminates the corresponding de- Compton (downscatter) processor or re- represent the discriminated downscat-
tection rate (interarrival) parameter jected (photoelectron only). If accepted, ter photons. Notice that these align
for that particular channel; it is processed further to improve with the PHS column’s energy “lines.”
The third column shows the decision
function for each of the targeted radio-
nuclides. As each photon is processed,
the decision function is sequentially up-
Sequential radiation
dated until one of the thresholds, target
detection processor (SRaDS) Initialize or nontarget, is crossed (indicated by
solid red boxes in the figure), declaring a
Photon
threat or nonthreat.
When a cargo container arrives at an
SRaDS detector, the decision function
in the software is refreshed, updated,
Target Target and refined based on the energies and
Yes photoelectric? No downscatter? No
(discriminate) (discriminate)
Reject arrival times of the accepted photons.
Detection is declared only when statisti-
cally justified according to the three
Yes factors—the Bayesian algorithms, the
Photoelectric Downscatter
updated decision function, and the
Next conditions defined by the specific ROC
Parameter Parameter
estimation estimation curve obtained during initial calibration.
In contrast, conventional techniques
require manually setting a specific
Previous
Table Decision calculation Decision SPT model counting time in advance with the hope
(target PE function function (target DS that the data acquired can justify the
parms) (photoelectron) (downscatter) parms)
decision. By encompassing the statisti-
cal nature of radiation transport physics
+
and sequential Bayesian processing
techniques, SRaDS provides highly de-
Composite
decision function veloped quantitative statistical analysis
of the data received in real time.
What’s more, basic and advanced
Detection processor options are available with
SRaDS. Both processor options pro-
vide complete statistical analysis of
No Yes
RN? Alarm radionuclide data obtained from any
type of gamma detector. The basic and
advanced processors gather information
Figure 4. The statistical Bayesian design is shown (simply) as a photon-by-photon
processor enhancing the raw detector measurement while rejecting instrumentation noise from unscattered photons that de-
and estimating the photoelectrons and downscatter photons through energy/rate posit full photon energy. The advanced
discrimination and parameter estimation. This information is input to a function used to processor also gathers information from
detect which of the target radionuclides are present. Compton-scattered photons that exhibit
1300 1300 20
Cobalt Photoelectric photons
60Co
10
1200 1200 Cobalt
0 detection
(5.76 s)
1100 1100 –10
Downscatter photons 0 1 2 3 4 5
1000 1000
Time (s)
900 900
20
800 Cesium 800
Volts (keV)
137Cs
700 700 Photoelectric photons Cesium detection
–20 (0.47 s)
600 600
–40
500 500 0 1 2 3 4 5
Barium Time (s)
Photoelectric photons
400 400
300 300 30
20
200 200 Downscatter photons 10
133Ba
Barium detection
0 (0.46 s)
100 100 –10
Photoelectric photons
–20
0 0
40 30 20 10 0 2 3 4 5 6 0 1 2 3 4 5
Counts Time (s) Time (s)
Figure 5. Results of sequential Bayesian detection and identification. (a) Pulse-height spectrum (after calibration). (b) Photon arrivals
(red) with targeted photoelectron discrimination (green circles) and downscatter photons (purple squares). (c) Decision functions for
60Co (detection time: 5.76 s), 137Cs (detection time: 0.47 s) and 133Ba (detection time: 0.46 s) with thresholds for radionuclide
detection/identification.
diminished energy—a major break- The technology can also be installed ing results that a nontechnical user can
through in time-domain, low-count in portable gamma detectors used by interpret. Alternatively, SRaDS can be
detection technology. first responders to determine radia- configured to simply provide audio and
tion risks associated with local nuclear visual alerts indicating the presence of
Integrates into Any emergencies. The algorithms are easily targeted radionuclides at user-selected
Gamma-Detector System embedded in programmable gate arrays confidence levels. Users can also select
The Livermore team took special that users in the field can adjust to a false-alarm probabilities to reduce
care to ensure that SRaDS can easily be location’s specifications and detection or eliminate the occurrence of false
integrated into any gamma-detection requirements. positives depending on the level of
system, including large stationary detec- Depending on the hardware setup, detection required for a given situation.
tors at transportation portals that help the processed data can be graphi- SRaDS is a comprehensive software sys-
search for radioactive contraband mate- cally displayed on a computer monitor tem that combines outstanding radionu-
rial in moving vehicles, cargo containers, or portable unit. While conventional clide-detection performance with high
and railroad cars. SRaDS works equally gamma-detection systems require a reliability and a short acquisition time.
well in pedestrian monitors used to highly trained practitioner to analyze The system can be implemented easily
combat illicit trafficking of radioac- the results, refine the data, and guide in existing infrastructure to protect the
tive material through customs, border the interpretation procedure, SRaDS nation from the insidious threat of illicit
crossings, and limited-access areas. displays data in intuitive plots, show- radioactive materials.
Capturing Waveforms in a
Quadrillionth of a Second
For more information contact:
Corey V. Bennett
The FemtoScope time microscope provides dramatic (925) 422-9394
improvments in instruments’ resolution and dynamic range. bennett27@llnl.gov
Figure 1. Livermore development team for the FemtoScope (from left): Bryan Moran, Vincent Hernandez, Alex Drobshoff, and Corey Bennett.
the available dynamic range at a given When combined with a 20-GHz recorded in a single shot with a conven-
speed. In Figure 3, a simulation shows real-time oscilloscope, the Femto- tional imaging spectrometer.
how three optical pulses separated by Scope produces an instrument capable
6 picoseconds (first to last) can be “time of recording 850-GHz waveforms in Emerging Needs
magnified” so that they occur over 100-picosecond frames at 155 million The FemtoScope represents a fun-
18 picoseconds at the output. frames per second until its memory is damental paradigm shift in high-speed
The FemtoScope uses a single-shot full. When combined with an optical imaging technology. As researchers
process in real time to capture each streak camera, the FemtoScope pro- improve their understanding of physical
window of time (or frame) of interest duces an instrument with a 20-times phenomena, they will need to exam-
and stretches out the waveform so that increase in temporal resolution and a ine processes on shorter and shorter
greater detail is revealed. Furthermore, 30-times increase in dynamic range, time scales. The FemtoScope will be
this process can be repeated at a rate resulting in an overall improvement of an invaluable tool for collecting de-
of more than 100 million frames per 600 times compared with the perfor- tailed dynamic data at faster temporal
second to record the real-time evolution mance of the streak camera alone. The resolution.
of a signal. With ultrafast resolution and same time-lens technology can also The Laboratory plans to use the
nearly endless recording length, this be configured to Fourier-transform FemtoScope on NIF experiments, which
instrument can uncover waveform data an input waveform. This produces an will need diagnostics with time resolu-
with peaks and valleys never before output spectrum that looks like the tions on the scale of 1 picosecond or
detectable. input temporal profile and which can be less to determine when high-energy
photons first appear and what happens
from their first appearance to their peak
production. The FemtoScope will also
Input Input Time Output be useful for detecting and recording a
signal fiber lens Fiber Bragg grating signal
broad range of signal strengths—from
very weak signal intensities to very
–15 strong.
The time-lens technology is funda-
–10 mentally an optical signal manipulation
tool. NIF has applications in which the
–5 fundamental measurement desired is
that of an x-ray signal. We have been
Time (ps)
Grating
Nonlinear
Result:
t mixing crystal
Focus control, fiber Mirror Time of
X-ray Optical probing optic input dispersion Detector rad. profile
Lens array in space
or
Chirped pump pulse
Record magnified time
(past approach)
Result: Time of
2009 R&D 100 Award, FemtoScope front end with Fiber Bragg rad. profile in
improved read-out for single-event and high-dynamic-range Grating (FBG) magnified time
output dispersion
“Time lens” based Fourier transform, operates single-shot
Figure 4. System conceptual diagram integrating an ultrafast x-ray sensor that produces a modulated optical response with a time-lens
based processor. Past work focused on time magnification using fiber Bragg gratings at the output to create a magnified waveform recorded
in time with streak cameras or oscilloscopes. Today, our system utilizes a conventional grating-based spectrometer to map the waveform to
space where it can be recorded on a conventional high-performance CCD camera.
the magnification. Today we are using a The FemtoScope will also be a Real-Time Recording Using Temporal Imag-
conventional grating-based spectrom- valuable tool for Livermore researchers ing,” OSA Conference on Lasers and Electro-
eter to map the waveform out in space, who are beginning development of a Optics CTuA6, San Jose, CA, May 6, 2008.
and record it with a charge coupled new energy concept known as the Laser 3. Lowry, M. E., C. V. Bennett, S. P. Vernon,
device (CCD) camera (Figure 4). Inertial Fusion Engine, or LIFE, which R. Stewart, R. Welty, J. Heebner, O. Landen,
The true potential of temporal im- is based on physics and technology and P. M. Bell, “X-ray Detection by Direct
aging is just beginning to be explored. developed for NIF. LIFE has the potential Modulation of an Optical Probe Beam–
The FemtoScope could also be applied to meet future worldwide energy needs RadSensor: Progress on Development for
to several other research facilities and in an inherently safe, sustainable man- Imaging Applications,” Rev. Scientific Instru-
experiments with diagnostic needs ner without carbon dioxide emissions, ments, 75, 10, pp. 3995–3997, 2004.
similar to those of the NIF. The Defense while dramatically shrinking the planet’s 4. Bennett, C. V., and B. H. Kolner, “Principles
Advanced Research Projects Agency co- stockpile of spent nuclear fuel. of Parametric Temporal Imaging—Part I:
funded Livermore to develop the tech- System Configurations,” IEEE J. Quantum
nology for LIDAR, which measures the Related References Electronics, 36, 6, pp. 430–437, 2000.
properties of scattered light to gather 1. Hernandez, V. J., C. V. Bennett, 5. Bennett, C. V., and B. H. Kolner, “Principles
information about a distant target. The B. D. Moran, A. D. Drobshoff, C. Langrock, of Parametric Temporal Imaging—Part II:
same high repetition rate system has D. Chang, M. M. Fejer, and M. Ibsen, System Performance,” IEEE J. Quantum Elec-
applications in particle counting and “745 fs Resolution Single-shot Recording tronics, 36, 6, pp. 649–655, 2000.
sub-picosecond resolution time-of-flight at 2.1 TSample/s and 104 Mframes/s Using 6. Bennett, C. V., and B. H. Kolner, “Upcon-
measurements at the Large Hadron Temporal Imaging,” OSA Nonlinear Optics version Time Microscope Demonstrating
Collider at CERN, where scientists are Conference, PDNFA2, Honolulu, HI, July 17, 103 x Magnification of Femtosecond Wave-
recreating in the lab conditions just af- 2009. forms,” Optics Letters, 24, 11, pp. 783–785,
ter the Big Bang and studying the basic 2. Bennett, C. V., B. D. Moran, C. Lan- 1999.
nature of subatomic particles. grock, M. M. Fejer, and M. Ibsen, “640 GHz
work-head spindles and turret are posi- glove-box design, which is essential in the glove-box industry, the NWC, and
tioned at an operational height deter- for the machine operation including the UK Atomic Weapons Establishment
mined by glove-box ergonomics. Mount- loading, unloading, part transfer, tool (AWE) also were consulted.
ing these units on horizontal slide-ways change, machining, chip management, Proposed designs were reviewed
maintained the required operational cleaning, and maintenance. All op- and optimized using Delmia VNC (a
height for all machining operations. erations are performed through glove software package that allowed combi-
Good operator access is also achieved ports, so positioning of these ports rela- nation of both the machine and glove-
by keeping the glove-box profile parallel tive to the machine was important, as box design models), which allowed
and close to the machine axis. was positioning of windows. Arm reach simulation of the machine’s kinematics
To maximize machine flexibility, a through the glove port has a critical and its controller, including part pro-
vertical servo axis was incorporated to effective length: if too short or long to gram simulation for part fabrication.
allow raising and lowering the turret. carry out an operation, operator strain The same software was used to model
This provided greater flexibility for the and injury could result. Awkward reach operator ergonomics. The designs were
live tooling (drilling and milling), use of is also the main cause of a glove breach. also evaluated by AWE using their own
a three-axis “tool-setting station,” tool- Reducing operator injury and glove specially developed 3-D simulation
setting after indexing (which eliminated port breaches was a major objective of package. Using computer-aided design
turret indexing errors), and a parking the system design, so the design team models of the machine and glove box, a
zone for better access for cleaning and comprised engineers, machine opera- full-size, three-dimensional image of the
work fixture change-out. tors, maintenance technicians, and an machine was generated. Figure 3 shows
An important aspect of the ma- occupational physiotherapist with a typical model of the machine and sup-
chine design is its integration with the knowledge of glove-box injuries. Experts port equipment.
Using only a head-mounted vision both unit cost and waste material and builder has been an integral part of the
system, the operators and designers also eliminated a thermal process that process, selecting and integrating their
were able to move 360° around the vir- removed a production bottleneck. The proven products into the design and
tual machine. This enabled glove ports near-net-shape casting enabled a 50% producing a cost-effective solution they
to be evaluated for position and reach, reduction in material to be removed can maintain and support in the future.
and ensured window placement was by the TS350 machine and a potential The TS350 provides NNSA with a robust,
optimal for visibility. Machine operabil- 50% reduction in machining time and modern manufacturing center that is
ity was also tested, providing essential collection of waste material, which in fully capable and sufficiently flexible to
feedback to the machine design without turn greatly reduced operator radiation resolve technical problems in the stock-
the need for building a full-scale model. exposure. The reduction in mass re- pile, and able to respond to adverse
This system provided instant operational duces the potential for operator injuries geopolitical change. AWE indicated they
feedback for design changes, greatly during part handling and transfer within intend to purchase the TS350 to meet
reducing the risk of discovering design the glove box. their future needs, and the machine tool
problems after manufacture and avoid- With the award of the manufactur- builder may offer a commercial version
ing cost and schedule overruns. ing contract to Hardinge, the TS350 of the machine. Keith Carlisle received
In parallel with the TS350 machine is now undergoing trials at their site an Award of Excellence for the planning,
design, LLNL worked within the Pluto- (Figure 4) where the machine is meeting managing, and execution of the modern
nium Sustainability program to develop all expectations. turning center evaluation and selection
a casting process for plutonium shells This project owes its success to the process and for significant contributions
that achieved near net-shape. This collaboration of many engineers across to the Stockpile Stewardship Program.
work resulted in major reductions in the NWC and AWE. The machine tool
Diana C. Chen
Enabling early diagnosis and treatment of blinding eye disease. (925) 423-5664
chen47@llnl.gov
4.5˚TR 4.5˚NR
8˚N8˚IR
4.5˚T4.5˚IR
corrector. The aberration-free signals beams travel along different paths until Inside the detector, a spectrometer and
are then integrated with the OCT for 3-D they ultimately reunite in a detector a charge-coupled-device camera record
image acquisition. Implementing AO in that measures their interference. In the sample and reference signatures.
the system increases lateral resolution MEMS-AO-OCT, an ultrabroadband light Custom computer software interprets
by approximately an order of magni- is generated using a superluminescent the recorded signals and produces
tude. Without aberration correction diode, and the sample beam propagates high-resolution, 3-D, digital images.
from the AO, the lateral resolution of a through a series of telescopes, mirrors, The device has a total footprint of ap-
clinical OCT system is not sufficient for and horizontal and vertical scanners proximately 0.5 cubic meters and can
imaging individual cellular structure. before reaching the patient’s eye. The be easily placed and moved within a
The wavefront corrector is a critical light beam is focused onto the patient’s physician’s office. In addition, its com-
component in the system. In traditional retina in a raster, or uniform, pattern, mercial components make the system a
AO systems, the wavefront corrector creating individual “snapshots” of each financially feasible option for practices,
can be both large and expensive. The layer. A wavefront sensor automatically and its cost is competitive with exist-
MEMS-AO-OCT was designed using a measures the patient’s optical aber- ing instruments that have much lower
MEMS-deformable mirror to reduce rations. A MEMS deformable mirror resolution.
the size and cost of the system without working in conjunction with a Badal op-
sacrificing speed or accuracy. Using this tometer and a pair of rotating cylinders Clinical Results
state-of-the-art AO technology enables then compensates for the distortions. To date, more than 100 individuals
the implementation of an instrument These components make the device have been clinically tested with AO-OCTs
suitable for clinical practice. The MEMS- effective even for patients who have built and operated at the University of
AO-OCT constructed at LLNL is the first large refractive errors, obviating the California, Davis and Indiana University.
instrument that has been optimized for need to fit patients with trial lenses. These instruments have been used
clinical use. This optimization involved The light reflected off the retina is then to image both healthy and diseased
a number of critical features including relayed back through the system to the eyes with different amounts of ocular
compact design and automating several detector. The reference beam, whose aberration.
components to enable the instrument path length matches that of the sample The AO-OCT system can acquire
to be operated efficiently by a clinician. beam, reflects off a pair of mirrors into data at different retinal eccentricities
OCT systems are based on inter- the detector. to allow high-resolution sampling of
ferometry, where light from a single Compact afocal telescopes align the retinal features of interest that are not
source is split into a sample and a system components with the patient’s resolvable by other methods. Figure 2
reference beam. These two separate pupil to achieve precise measurements. shows results of an AO-OCT imaging
15N 2SR
16N 6SR
12NR
15N 2SR
17N 1SR
14N 2IR
12NR
14N 2IR
Figure 3. Example of AO-OCT imaging of the optic nerve head. (Left) Fundus photo with superimposed OCT-fundus reconstructed
from the AO-OCT volumes. (Right) Representative B-scans and C-scans.
session during which several retinal ec- makes the imaging of eccentric loca- and monitoring the progression of ONH
centricities were imaged with 1 × 1 mm tions such as the ONH more difficult. diseases. Figure 3 shows results of ONH
sampling volumes at the fovea (dark We find that retinal structures are re- imaging with AO-OCT.
region) and optical nerve head (ONH, solved and revealed using AO-OCT that Figure 4 shows micro-traction in the
bright region). were not visible using OCT without AO center of the fovea. Because of its small
Chromatic aberration effects in- correction. Imaging these microscopic size, it was misdiagnosed using other
crease with retinal eccentricity, which ONH structures is critical for diagnosing image modalities including commercial
Figure 4. Micro-traction was not detected with other imaging Figure 5. Microscotoma cannot be detected with other imaging
modalities. modalities.
Daniel A. White
Implementation of magnetohydrodynamics in ALE3D provides (925) 422-9870
new and exciting capabilities for pulse power, microfluidics, and white37@llnl.gov
other applications.
Figure 2. Simulation of an explosive magnetic flux compression generator. The graphic on the left is a snapshot just as the high
explosive (HE) is detonated; the graphic on the right is at a later time when the magnetic field has been compressed. The vectors
represent the magnetic field; the contours represent pressure.
Computer codes for simulating this class Opening New Horizons in Pulse As a specific example, the device shown
of magnetohydrodynamics problems Power Research in Figure 2 is designed to be axially sym-
have existed for some time. The impor- The primary motivation for develop- metric, but the effect of deviations from
tant feature of our magnetohydrody- ing an ALE3D magnetohydrodynamics perfect symmetry (due to manufactur-
namic implementation in ALE3D is that capability was to simulate explosive ing tolerances, material imperfections,
it excels for problems that involve solid magnetic flux compression generators. etc.) can be investigated via full three-
materials that undergo gross deforma- These generators are energy conver- dimensional simulation by incorporating
tion, melting, and transition to plasma. sion devices, in that they convert the imperfections into the simulation. The
The mutiphysics nature of such prob- chemical energy in a high explosive to ability of ALE3D to simulate the deto-
lems is illustrated in Figure 1, where mechanical energy, and the mechanical nation of a high explosive, combined
electromagnetics is represented by a energy is in turn converted to electri- with material motion, electromagnetic
white module, heat transfer is repre- cal energy. The purpose of the device fields and currents, and heating and
sented by yellow, and hydrodynamics is is to generate a large pulse of electric melting of conductors, is a very power-
represented by green. The arrows show current with a faster rise time than ful and unique capability that is having
the physical quantities that are the can be achieved using other means a tremendous impact on the design of
input and the output of each module. (such as capacitor banks). To function, explosive magnetic flux compression
The coupled equations are solved using the device is seeded with an electric generators.
an operator-splitting algorithm, and current, generating a magnetic field A railgun is another energy conver-
within each module the equations are in a cavity; the high explosive is then sion device, in some sense the opposite
solved using the finite element method. detonated, compressing the cavity, of a magnetic flux compression gen-
The illustration in Figure 1 is deceptive; and as the cavity is compressed, the erator. In a railgun, electric current is
it looks simple, but the development of initial seed current is amplified in order input into the device, and the output
this magnetohydrodynamics capability to maintain the magnetic flux in the is mechanical energy in the form of a
required hundreds of thousands of lines cavity (in accordance with Lenz’s law). high-velocity projectile. A railgun is es-
of software. Simulations using magne- An example simulation of this process sentially a linear electric motor, and the
tohydrodynamics often require large- is shown in Figure 2. Such simulations projectile is the armature of the motor.
scale computing—a typical problem will have enabled engineers to visualize Railguns have application in defense as
require a week of processing time using fields and currents that are difficult to an alternative to traditional explosive-
256 processors, while the largest prob- measure in experiments, thus giving based guns. One advantage of a railgun
lems require thousands of processors. insight into the operation of the device. is the very high velocity that can be
Magnetic field
Stress
Figure 3. Simulation of a generic railgun. The armature is between two rails; current flows into one rail, through the armature,
and out the other rail. This graphic shows only one-half of the geometry so that the stress within the rails can be seen.
achieved by the projectile compared hence accurate estimation of stress Naval Research in their goal to design
to that of a traditional gun. A second via full three-dimensional simulation is a naval railgun. ALE3D simulations are
advantage is that the velocity can be essential. A novel capability of ALE3D is being used to investigate alternative
“dialed” by the electrical power sup- its ability, via an algorithm, to correctly rail geometries, alternative rail materi-
ply. Figure 3 is a simulation of a generic simulate a sliding electromagnetic con- als, and the effect of Joule heating and
railgun, with the graphic showing a tact. This algorithm allows two materials viscous heating on armature erosion.
cross-section of the gun. The purpose to slide past each other while allow-
of this particular simulation was to ing electric current to flow across the Simulation Enables Virtual
investigate stress in the rails, which is material interface. The sliding electrical Prototyping
caused by the enormous magnetic pres- contact algorithm is quite sophisticated, The two previous examples of
sure in the gun. A high magnetic field involving advanced mortar finite ele- magnetohydrodynamic simulation were
is required to achieve high acceleration ment methods with implicit constraints pulse power applications involving
of the armature, but a high magnetic to satisfy conservation of current and possibly millions of amperes of current.
field also causes stress in the rails. If continuity of the magnetic field across A third example, simulating a micro-
the gun is not properly designed (i.e., the interface. Currently, LLNL is using fluidic pump (Figure 4), is quite differ-
if the stress is too high) it can explode, the ALE3D code to assist the Office of ent and illustrates the broad utility of
Magnetic field
Inflow
Outflow
Magnetic membrane
Figure 4. Simulation of a magnetoelastic-based microfluidic pump. The applied magnetic field (magenta vectors) deforms a magnetoelastic
membrane (solid black), which pulls fluid into one side of the pump and expels fluid out the other side of the pump. The pseudocolor scalar
field is the speed of the fluid, blue being slow and red being fast.
Joel V. Bernier
Measurements at the Crystal Scale (925) 423-3708
bernier2@llnl.gov
Related References
1. Lee, J. H., C. C. Aydiner, J. Almer,
J. V. Bernier, K. W. Chapman, P. J. Chupas,
D. Haeffner, K. Kump, P. L. Lee, U. Lienert,
A. Miceli, and G. Vera, “Synchrotron Ap-
plications of an Amorphous Silicon Flat-
Panel Detector,” J. Synchrotron Rad., 15, 5,
pp. 477–488, 2008.
2. Lienert, U., J. Almer, B. Jakobsen,
W. Pantleon, H. F. Poulsen, D. Hennessy,
C. Xiao, and R. M. Suter, “3-Dimensional
ε {102}
Characterization of Polycrystalline Bulk
Materials Using High-Energy Synchrotron
Radiation,” Mater. Sci. Forum, 539–543,
pp. 2353–2358, 2007.
3. Aydiner, C. C., J. V. Bernier, B. Clausen,
U. Lienert, C. N. Tomé, and D. W. Brown,
“Evolution of Stress in Individual Grains and
Figure 2. Results from experiment in which the α↔ε phase transformation was observed
Twins in a Magnesium Alloy Aggregate,”
in situ. The location of Bragg reflections for the two indicated types of ε-phase lattice
planes are plotted as pole figures in an equal area projection. The different glyphs denote Phys. Rev. B, 80, 2, 024113, 6, 2009.
the predicted locations of reflections from each variant under the Burgers path. The red 4. Merkel, S., H. R. Wenk, P. Gillet, H. K. Mao,
and green colors correspond to the extrema of the orientation spread of the parent α phase and R. J. Hemly, “Deformation of Polycrystal-
(~15° misorientation). The correspondence of the strongest variants is quite good, even in line Iron Up To 30 GPa and 1000 K,” Phys.
this cursory analysis, and all variants are observed. Earth Planet. Inter., 145, pp. 239–251, 2004.
(a) (b)
(a) (b) Relevance to LLNL Mission (Fig. 2a) is compared to the basic ALE
The technologies being developed approach (Fig. 2b). Figure 2 shows
can support a variety of multiphys- the coarsest mesh in the study; two
ics simulations for the Laboratory. additional mesh refinements were made
The initial focus on blast effects upon to establish convergence. The embed-
structures is consistent with many LLNL ded mesh exploits an Eulerian mesh of
national security missions. Its generality the fluid and a Lagrangian mesh of the
will be demonstrated by extension to plate. The ALE model uses the default
include fluid-structural-electromagnetic ALE3D relaxation with advection in the
coupling for moving solid conductors fluid and a Lagrangian mesh of the plate.
such as flux compression generation The blast bends the plate and forces
and rail guns. the plate to flow through the fluid a
significant distance (Figs. 2 and 3). While
FY2010 Accomplishments the Eulerian fluid background mesh is
and Results stationary, the ALE fluid mesh advects
We completed the first version with the moving plate. Figure 4a com-
of the FEusion software library that pares the pressure at the plate surface
identifies the overlap of foreground on the centerline, and Fig. 4b compares
and background meshes and generates the velocity at the center and ends of
constraints between the solid and fluid the plate for embedded mesh and ALE
regions. The library was ported to three models. Both pressure and displace-
codes: DYNA3D, ALE3D and NIKE3D. In ments match well for long times (10 ms).
this process we identified and imple- Figure 4c shows the difference in plate
mented modifications to the ALE3D center displacement for different ele-
advection routines for transporting ment sizes h and demonstrates that the
material response quantities between results appear to converge.
elements. Initial computational results
with multiple model problems across all
three codes are highly encouraging. FY2011 Proposed Work
The ALE3D implementation is dem- In FY2011, we plan to
onstrated in Fig. 2 where its standard 1) improve the advection scheme;
discretization is used as a benchmark 2) incorporate a new method for
Figure 3. Pressure field in fluid and
for the embedded mesh method. A handling potential locking; 3) add
effective plastic strain in plate for the two-inch-thick steel plate is exposed a shell element capability; and
(a)embedded mesh and (b) ALE to an underwater blast and an analysis 4) extend our work to electromag-
simulations at successive, identical times. using the new embedded mesh method netic analysis.
Pressure (embedded)
Center (conforming) 7
0.020 0.020
End (conforming 6
0.015 0.015
5
0.010 0.010
4
0.005 0.005
0 0 3
–0.005 –0.005
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000 2.0 1.0 0.8 0.6 0.5
Time (µs) Time (µs) h
Figure 4. (a) Comparison of centerline pressure in plates. (b) Comparison of displacement at center and ends. (c) Difference in center of
mass displacement for three mesh sizes. Results (a–b) are in good agreement and (c) demonstrates some convergence.
0.8
Project Goals 0.6
Our ultimate goal is to perform
0.4
a series of large-scale simulations of
dilute polymer solutions in turbulent 0.2
flows with a detailed model of the 0
0 0.2 0.4 0.6 0.8 1.0 1.2
polymer chains and the hydrodynamic
z/Rg
interactions. To resolve the relevant
scales we are incorporating the follow-
ing improvements to our existing LB Figure 1. Polymer center of mass distri-
bution in a confined channel with height,
polymer code: 1) enhanced numerical
h = 2.5Rg, as a function of distance (z) from
stability; 2) accurate hydrodynamic the lower wall. Rg is the polymer radius of
fluctuations; and 3) integration of the gyration, which is a measure of the average
solver with an adaptive mesh refine- size of the chain. Note the curves are sym-
ment (AMR) framework. metric so only half the channel is shown.
validation. One of the main drawbacks finding the polymer at a certain height, 4. Ansumali, S., I. V. Karlin, and
of the LB method is that it becomes z, above the lower wall for the case H. C. Ottinger, “Minimal Entropic
unstable for high Reynolds number flows without any flow in the channel. With Kinetic Models for Hydrodynamics,”
relevant to drag reduction. Fortunately, a flow present, the polymer can stretch Europhys. Lett., 63, 6, pp. 798–804,
a new generation of methods, known as and sample regions closer to the wall, 2003.
entropic LB has overcome this stabil- as is indicated by the purple line. These 5. Rohde, M., D. Kandhai, J. J.
ity issue and we have developed an results provided additional validation for Derksen, and H. E. A. van den Akker,
algorithm for fluctuating hydrodynamics our algorithm. “A Generic, Mass Conservative
that incorporates these entropic solvers. One of the key challenges was Local Grid Refinement Technique
We have successfully run simulations incorporating mesh refinement for the for Lattice-Boltzmann Schemes,”
of turbulent channel flow for Reynolds LB method, which required developing Int. J. Numer. Methods Fluids, 51, 4,
numbers up to 12,000 using our new the methodology to transfer informa- pp. 439–468, 2006.
solver. These calculations are not realiz- tion between coarse and fine grids, as
able with traditional LB methods. shown in Fig. 2. We chose the Chombo
In addition, we performed bench- library developed at LBNL for the AMR x 10-4
marking laminar-flow simulations of a framework. Since LB evolves a discrete 1.5
single polymer migration in a microchan- velocity distribution function and not 1.0
Method from
nel using the new entropic code. We the macroscopic hydrodynamic variables ref. 5
LB Chombo
Velocity error
want to understand how the polymer (i.e., mass and momentum) the stan- 0.5
will migrate relative to the wall in differ- dard finite difference flux conservation 0
ent flow conditions. In a narrow chan- approach at grid interfaces was not ap-
–0.5
nel the chain will migrate toward the propriate. Instead we developed a novel
wall if the flow rate is high enough. Our algorithm that minimizes the numerical –1.0
simulations reproduced this behavior as error associated with refinement. The –1.5
the shift in the distribution curves of the plot in Fig. 3 compares the errors from –20 –10 0 10 20
polymer center of mass in Fig. 1 demon- both our method and a popular state- z
strates. The blue line is the probability of of-the-art refinement scheme for a uni-
formly accelerating flow parallel to the Figure 3. Mesh refinement error near
coarse–fine interface (z=0) for uniformly
coarse–fine interface. Our error is orders
accelerating flow parallel to the interface.
of magnitude lower. The fine mesh is located at negative z
values and the coarse at positive z.
Related References
1. Toms, B. A., “Some Observations on the
Flow of Linear Polymer Solutions Through
Straight Tubes at Large Reynolds Numbers,” FY2011 Proposed Work
Proc. 1st Int. Cong. Rheol. N. Holland, Am- In FY2011 we plan to: 1) com-
sterdam, 2, pp. 135–141, 1948. plete the adaptive mesh refinement
2. Zhang, Y., A. Donev, T. Weisgraber, development; 2) begin Newtonian
B. J. Alder, M. D. Graham, and J. J. de Pablo, turbulent channel simulations to
“Tethered DNA Dynamics in Shear Flow,” validate the AMR code; 3) continue
J. Chem. Phys., 130, 23, pp. 234902–234913, the laminar flow migration
y
2009. simulations with wall roughness;
z x 3. Usta, O. B., A. J. C. Ladd, and J. E. Butler, and 4) investigate how many-
“Lattice-Boltzmann Simulations of the core graphics processors could
Figure 2. Example mesh for flow in a Dynamics of Polymer Solutions in Periodic accelerate our hybrid LB/MD
channel with grid refinement near the and Confined Geometries,” J. Chem. Phys., computations.
walls. 122, 2005.
Z Project Goals
Y
The project provides ongoing user
X
support and software maintenance for
t = 3.5 × 10–3 LLNL’s visualization and postprocessing
Figure 1. “Mesh-less” particle representation of soil being ejected tools and adds new capabilities to these
by a buried explosive charge, as simulated by ParaDyn. Griz is being tools to support evolving, multipro-
extended to support additional types of discretization technologies. grammatic requirements.
FY2010 Accomplishments
and Results
Our baseline commitment to LLNL do vis
users, the annual release of updated ()
production software, was achieved with
invis
the issue of Version 10.1 of Griz, Mili,
and XmiliCS. This is a necessary comple-
ment to annual releases of the DYNA3D
and ParaDyn simulation tools. The grow- Pipeline of commands
ing “mesh-less” capability in those codes to VisIt
for representing highly deforming mate- Figure 2. GrizIt Command Line Interface. The common “factory design pattern” software
rial creates new needs for support from construct provides engineering users with familiar, highly productive ways of accessing and
the visualization and data management investigating their simulation results while leveraging the high-performance visualization
tools. Figure 1 provides an example of capabilities of the VisIt tool and the flexibility of its Python command interface.
how Griz provides color contour plots
of basic solution quantities over the
ensemble of material points used in a of the particle sets arising from the new permit the set of commands supported
“mesh-less” scheme. Efforts are now “mesh-less” simulations. to smoothly expand. Proof of concept
underway to robustly extend Griz’s The Laboratory’s VisIt data analysis has been demonstrated and a number
capabilities for standard continuum ele- tool is a world-class tool for rendering of specific commands have been imple-
ments, e.g., cutting planes to expose the extremely large scientific data sets. This mented.
interior of a body, to this added simula- capability is of interest to engineering
tion option. analysts as our own models grow, but
We are halfway through an effort to adoption has been slow because the
re-architect and re-implement the Xmil- operational flow of VisIt is so different FY2011 Proposed Work
iCS utility for consolidating the very large from that of Griz. We are now creating a Targets for next year include
results databases created with ParaDyn. capability so users can type the concise 1) delivering the new XmiliCS com-
This is being used as an opportunity to commands familiar to them from Griz biner tool; 2) having a user-testable
redefine the boundary between Mili and while the computer automatically issues (“beta”) version of GrizIt exercised
XmiliCS so that common operations are the more complex stream of multiple sufficiently to support a decision
consolidated in the Mili interface and instructions required to have VisIt on delivery with Version 11.1 of
better leveraged (re-used) by XmiliCS. generate the specific image the user the ParaDyn Suite; and 3) initial op-
This major evolution of the products desires. We have named this lightweight erational capability for end-to-end
also provides an opportunity to incor- add-on GrizIt. A modern object-oriented regression testing of the ParaDyn
porate a dedicated “particle” data type design (Fig. 2) allowing for flexible class Suite workflow.
to more specifically support definition manipulation and reusable code will
Jerry I. Lin
(925) 423-0907
lin5@llnl.gov
Figure 1. First principal material axes automatically generated Figure 2. Second principal material axes at points of a generic
at points of a generic conical/cylindrical structure. The dots conical/cylindrical structure, locally orthogonal to the axes
represent mesh points and the line segment the orientation of shown in Fig. 1.
the local material axis.
Relevance to LLNL Mission great hindrances is assigning local mate- by tracking a set of discrete points and
Engineering analysts supporting rial directions at the various locations assessing the local stress state in the
a variety of programs require new throughout the material. For parts of body via their relative motions.
structural modeling functionalities and special geometric shapes that align with Figure 3 shows a simple example of
technical support to complete their the material directions, computational two bars striking end-on-end. We are
missions. Some of these programs algorithms can be created to automati- leveraging an SPH capability originally
and projects involve the Laboratory’s cally calculate the material orientations. created as a standalone code at LLNL.
collaboration with other institutions Two such geometric configurations, During its prove-in phase, the module
and federal agencies, including the Los conical and ellipsoidal, were added to remained nearly a separate code as it
Alamos National Laboratory (LANL), the existing choices for all anisotro- used an independent (small) input file
the Missile Defense Agency, the Naval pic continuum materials in DYNA3D. and output a separate visualization data-
Surface Warfare Center, the U.S. Army Figures 1 and 2 depict the automatically base. The capability is now more seam-
Corps of Engineers, and the Department generated two principal material axes, lessly integrated from a user perspective
of Homeland Security. the first one along the structure longi- as the SPH input parameters are defined
tudinal direction and the second along within the DYNA3D input file. Similarly,
FY2010 Accomplishments the transverse direction, at points of a the response quantities computed
and Results merged conical/cylindrical part. within the SPH region are now written
The use of composite materials such The drive to model more extreme as part of the standard DYNA3D output
as graphite-epoxy has risen signifi- structural deformations requires sus- database. This both simplifies the col-
cantly in structures and protective gear. tained investment in additional nu- lection of files the analyst must manage
Fiber-reinforced materials are typi- merical representations in DYNA3D. One and facilitates visualization and interpre-
cally anisotropic, having non-uniform option mechanical analysts have found tation of the simulation results.
mechanical and thermal behaviors in useful for high-speed impacts is Smooth
different directions. For model prepara- Particle Hydrodynamics (SPH). This tech-
tion involving these materials, one of the nique approximates material behavior
FY2011 Proposed Work
General technical support for
DYNA3D users, the addition of
user-requested capabilities, and
SQA-compliance work will continue
under FY2011 funding. For users
modeling more complex shapes
constructed of anisotropic materi-
als we are introducing a general
auxiliary file that can be read for
the local material orientation
specifications. This will free them
to write or adapt small, problem-
specific utilities.
Code modifications leading to a
completely keyword driven input
file will continue as our goal is to
provide users with a more conve-
niently read and modified problem
specification. We will use a regres-
sion test suite traceability matrix
completed in FY2010 to guide and
prioritize further test problems to
assure performance and stability of
Figure 3. Hopkinson bar high-speed impact modeled with the SPH option now DYNA3D’s features.
integrated with DYNA3D.
Michael A. Puso
(925) 422-8198
puso1@llnl.gov
x 1 thread factorization
5000 to four subdomains per node are used
for very large analyses (i.e., greater than
4000
1 subdomain x 2 thread
3000
1 subdomain x 4 thread
2000
1000
4 subdomains η1 η2 ηk
x 4 thread
0 σ σ
0 5 10 15 20
# of processors per node E1 E2 Ek
–0.020
–0.019
Displacement
–0.018
–0.017
–0.016
New logarithmic creep
–0.015 Standard linear creep
–0.014
0 1x108 2x108 3x108
Time (s)
several EM problems of interest to is essentially a “change of basis” result- user, and new samples were chosen to
Global Security projects. The outline of ing in a significantly more compact minimize the error. The process and an
the MOR algorithm is as follows. representation than the full model. example are shown in Figs. 1 and 2.
Let A(s) x = b be the boundary ele- The reduced system is evaluated at M As an example, consider the moti-
ment discretization of Maxwell’s equa- sample points (M different values of the vating problem of computing electrical
tions, where A(s) is the N × N bound- parameter vector s); between sample currents on cables where a brute force
ary element matrix, x is the vector of points the reduced system is defined by approach would result in one million
unknowns (surface current and charge), linear interpolation using Radial Basis EIGER calculations. Using the automatic
b is the source vector, and s = {s1, s2, Functions (RBF). The basis function MOR algorithm outlined above, we
..., sk} is the vector of parameters. expansion is were able to completely explore this
The parameters could be, for example, parameter space to within an error
frequency, angle of incidence, material a(s) = Σ a φ ( s–s )+θ(s)
i
i i of 1% by performing only about 800
properties, or positions of cables or EIGER calculations, a reduction of over
apertures. The number of unknowns, where the RBFs are of the form 1000 ×. It is also important to note that
N , is determined by the computational f(r) = r2p, and q(r) is a polynomial of the MOR software implementation is
mesh and can be quite large. N ≈ 104 is order p. The coefficients of the basis parallel. The EIGER calculations were
common and N ≈ 106 is possible when function expansion are determined by executed using 8 nodes (32 processors)
a fast multiple multipole algorithm is solving a Vandermonde system of equa- and completed in just one day.
used for the boundary element repre- tions, this system is only M × M.
sentation. The process of choosing the sample Related References
The reduced system is given by points is iterative. The process is like 1. Sharpe, R., et al., “EIGER: Electromagnetic
a(s) = VT A(s)V, where V is a N × M structured adaptive mesh refinement, Interactions Generalized,” IEEE Ant. Prop. Int.
matrix with M « N. The columns of V but in k-dimensional sampling space Symp., pp. 2366–2369, July 1997.
are the basis vectors of the reduced rather than the physical spatial domain. 2. Buhmann, M. D., “Radial Basis Functions,”
order model, the reduced order model An error tolerance is specified by the Acta Numerica, 9, pp. 1–38, 2000.
(a) (b)
Parameter space sampling and refinement
1.0
0.6130
0.4507 0.9
0.3066
0.1572 0.8
0
0.7
0.6
0.5
S2
0.4
0.3
0.2
0.1
X Y Z 0
0 0.2 0.4 0.6 0.8 1.0
S1
Figure 2. Example calculation using our automatic Model Order Reduction algorithm. (a) Current density computed on a circuit for a single
combination of two input parameters, s1 and s2. (b) Graphic showing the limited number of cases sampled in the plane needed to capture
the behavior over the entire parameter range.
Figure 1. Detail image of a microarray Figure 2. R&D system, showing user load-
scan after autonomously hybridizing E. coli, ing sample vial onto system prior to begin-
E. faecalis, and S. aureus on the Virulence ning integrated hybridization experiments.
Array. The microarray flow cell shown on right
hand side is connected to fluidics lines.
Cadmium–Zinc–Telluride Sandwich
For more information contact:
Detectors for Gamma Radiation Adam M. Conway
(925) 422-2412
conway8@llnl.gov
e
h
Anode Cathode
γ
Figure 1. Schematic diagram of a-Se/CZT/a-Si detector
layer structure.
Energy (eV)
FY2010 Accomplishments 0
and Results –0.5
Over the course of this project we
have 1) developed finite element mod- –1.0 a-Si
a-Se
eling capabilities for amorphous-CZT –1.5
heterojunctions to understand electronic Figure 2. Simulated energy
conduction mechanisms; 2) studied –2.0 band diagram of
–0.2 0.2 0.6 1.0 1.4 a-Se/CdZnTe/a-Si:H
amorphous-CZT heterojunctions using Position (µm) layered structure.
current vs. voltage vs. temperature
measurements for characterization of
Schottky barrier height; and 3) fabri-
cated amorphous Se-CZT-amorphous
Si heterojunction detectors (Fig. 2) that 10–6
have reduced the leakage current by No amorphous layer
100 × resulting in an effective resistiv- w/ a-Si
ity of greater than 1012 ohm-cm in w/ a-Se
Current density (A/cm2)
Related References
1. Voss, L. F., P. R. Beck, A. M. Conway, Figure 3. Comparison of
10–9 current versus voltage
R. T. Graff, R. J. Nikolic, A. J. Nelson, and –100 –75 –50 –25 0 characteristics with and
S. A. Payne, “Surface Current Reduction in Reverse bias voltage (V)
without amorphous contacts.
(211) Oriented CdZnTe Crystals by Ar Bom-
bardment,” J. Appl. Phys., 108, 014510, 2010.
2. Conway, A. M., B. W. Sturm, L. F. Voss,
P. R. Beck, R. T. Graff, R. J. Nikolic, A. J. Nelson,
and S. A. Payne, “Amorphous Semiconductor
600
Blocking Contacts on CdZnTe Gamma Detec- w/ a-Se
tors,” International Semiconductor Device w/ a-Si
500 No amorphous layer
Research Symposium, December 2009.
3. Voss, L. F., A. M. Conway, B. W. Sturm, 400
R. T. Graff, R. J. Nikolic, A. J. Nelson, and
Counts
300
S. A. Payne, “Amorphous Semiconductor
Blocking Contacts on CdTe Gamma Detec- 200
tors,” IEEE Nuclear Science Symposium, FWHM = 11.8 keV
October 2009. 100 FWHM = 15.5 keV
4. Nelson, A. J., A. M. Conway, C. E. FWHM = 21.8 keV
0
Figure 4. Comparison of
Reinhardt, J. J. Ferreira, R. J. Nikolic, and S. Am-241 gamma spectra with
A. Payne, “X-Ray Photoemission Analysis 0 40 80 120 160
and without amorphous
Energy (keV)
of CdZnTe Surfaces for Improved Radiation contacts.
Detectors,” Materials Lett., 63, 180, 2009.
Patterned deposition Au
RF
1 cm
70 nm gold & RF
Figure 2. Deposition of 70-nm gold particles onto a fixed electrode Figure 3. Example of a transparent ceramic fabricated using EPD.
pattern that is gelled in situ using resorcinol formaldehyde (RF). The The ceramic material is Nd:GYSAG and was deposited to a thickness
white areas are gold and the red areas are the gelled RF. Minimum of 1.4 mm in 10 min using a 10-V/mm electric field. The part shown
feature size at the center of the pattern is approximately 10 µm. in (a) was vacuum sintered at 1675 °C and then hot isostatically
pressed to create the final transparent part (b).
Project Goals process and demonstrate materials and 3. Developed a new gelation process
The goals of this project are: 1) structures of relevance to LLNL missions that enables deposition in an aque-
demonstrate the fabrication of function- and programs. The main demonstra- ous solution and then “locks in” the
ally graded materials with composition tions for this project align with current pattern by gelling the solution at
profiles tailored in three dimensions and future needs in NIF as well as the elevated temperature.
while maintaining desired bulk prop- LIFE and ALOSA thrust areas. These are: 4. Synthesized near-monodisperse
erties; 2) demonstrate the use of the 1) to create transparent ceramic optics fluorapatite nanorods and demon-
EPD deposition field to simultaneously with doping profiles tailored in three strated alignment of the rods in a
align nanorod particles of precursor dimensions to enable new high-pow- 300-V/cm electric field.
material as they are deposited; and 3) ered laser designs (NIF/LIFE); and 2) to 5. Demonstrated transparent ceramic
demonstrate the fabrication of compos- deposit aligned nanoparticles of noncu- structures fabricated using the EPD
ite structures with controlled material bic ceramics to create a new family of process (Fig 3).
composition and create smooth or sharp transparent ceramics (NIF/LIFE/ALOSA).
material transitions along the z-axis of a
composite structure. FY2010 Accomplishments
and Results FY2011 Proposed Work
Relevance to LLNL Mission Accomplishments and results in the In FY2011 we will 1) demon-
The project is intended to establish second year include the following: strate combined orientation and
LLNL leadership in bottom-up nanofabri- 1. Demonstrated the ability to change deposition control and fabricate a
cation of functionally graded materials. material composition and thickness transparent optic from a noncubic
Our dynamic EPD system will position us of deposition layers, and created material; 2) implement fixed-mask
to deliver the next generation of nano- both sharp and gradual material x-y control and demonstrate a
manufacturing capabilities for projects transitions during deposition (Fig. 1). transparent sintered part with a
throughout the Laboratory. Using these 2. Successfully deposited a 2-D extrud- planar composition gradient; and
capabilities, we are working to pro- ed pattern with 10-µm resolution 3) use in-situ AFM to monitor the
duce a number of novel materials and (Fig. 2). The material is deposited EPD process and validate our pro-
structures. These structures will both onto a photolithographically pat- cess model.
illustrate the capabilities of the new terned metal electrode.
High-Resolution Projection
For more information contact:
Micro-Stereolithography (PµSL) Christopher M. Spadaccini
for Advanced Target Fabrication (925) 423-3185
spadaccini2@llnl.gov
Project Goals
Overall project goals included:
1. Establishing a high functioning PµSL
system at LLNL that can produce 3-D
components on demand.
2. Improving the resolution of this
system by integrating a plasmonic
superlens.
Figure 2. SEM images of (a) cylinders, (b) lattice structures, and (c) 3-D components with 3. Broadening the range of materi-
overhanging features. als that can be used with PµSL to
Three-Dimensional Polymer
For more information contact:
Fabrication Techniques Christopher M. Spadaccini
(925) 423-3185
spadaccini2@llnl.gov
Inner fluid
Figure 1. Schematic of flow-focusing microfluidic geometry for microencapsulation. Figure 2. Optical micrograph of micro-
fluidic geometry.
We targeted a microfluidic approach middle fluid (the shell material) and the shows that the encapsulated material
for microcapsule fabrication based on a outer fluid, which focuses and pinches does not diffuse out of the shell.
flow-focusing geometry. We identified off the microcapsule droplets in the col- Dimpled structures were fabricated
this approach because it offers exquisite lection tube (Fig. 2). The dimensions of out of hexanediol diacrylate (HDDA) us-
control of shell diameter and wall thick- the capsules are controlled by the fluid ing our PµSL system. This technique uses
ness, which are important parameters viscosities and flow rate as well as the UV light and polymerizes 3-D shapes in
for mass transfer consideration and channel size. a layer-by-layer fashion. An example of
mechanical robustness. In addition, this We have assembled an apparatus a dimpled structure fabricated using this
approach conserves expensive materials for microcapsule fabrication (Fig. 3). Our technique is shown in Fig. 6.
since it encapsulates nearly 100% of the setup consists of an inverted microscope
desired solution, which is a significant and high-speed camera for visualization, Related References
improvement over other batch process- three syringe pumps for independent 1. Thorsen, T., et al., “Dynamic Pattern For-
es. In this approach, a tapered capillary flow control of the three fluids, a UV LED mation in a Vesicle-Generating Microfluidic
nozzle with circular cross-section is for curing of the photosensitive shell Device,” Physical Review Letters, 86, 4163,
aligned within a square capillary (Fig. 1). material, and a laptop computer for 2001.
To do this, the outer diameter of the control. The microfluidic device in Fig. 4 2. Utada, A., et al., “Monodisperse Double
circular capillary must be approximately shows the three inlets for the fluids and Emulsions Generated from a Microcapillary
equal to the inner diameter of the the outlet for collection of the microcap- Device,” Science, 308, 537, 2005.
square capillary. A second circular capil- sules. We used a photopolymerizable 3. Dendukuri, D., et al., “Controlled Synthesis
lary with a larger opening is inserted into material for the shell. The shells were of Nonspherical Microparticles Using Micro-
the square capillary from the opposite exposed to UV light after collecting in a fluidics,” Langmuir, 21, 2113, 2005.
end, which forces alignment. This con- vial (Fig. 5). We encapsulate a fluores- 4. Shepherd, R. F., et al., “Microfluidic
figuration allows the flow of three fluids: cent dye within the shells in order to Assembly of Homogeneous and Janus Colloi-
the inner fluid (to be encapsulated), the confirm that they are capsules. This also dal Granules,” Langmuir, 22, 8616, 2006.
500 µm
Figure 1. A close-up view of PDMS micro- Figure 2. PDMS microfluidic chip containing four separate devices.
channels and valves used for cell cultures.
The culture loop holds approximately 5 nL
of fluid and thousands of cells.
Figure 4. Microscope and environmental control system Figure 5. System control software.
used to monitor and maintain growth conditions on the
microfluidic system.
Laser excitation
Gas sensing tip
To spectrometer 1 µm
Nanopillars
Intensity
Element 1
10 1000
1
10–1 100
J (A/cm2)
Counts
10–2
10–3
10
10–4
D1 D1
10–5
10–6 1
–2.0 –1.6 –1.2 –0.8 –0.4 0 0.4 0 20 40 60 80 100 120 140
V (V) Channel
Element 4
10 1000
1
10–1 100
J (A/cm2)
Counts
10–2
10–3
10
10–4
D4 D4
10–5
10 –6 1
–2.0 –1.6 –1.2 –0.8 –0.4 0 0.4 0 20 40 60 80 100 120 140
V (V) Channel
Element 7
10 1000
1
10–1 100
J (A/cm2)
Counts
10–2
10–3
10–4 10
10–5 D7 D7
10–6 1
–2.0 –1.6 –1.2 –0.8 –0.4 0 0.4 0 20 40 60 80 100 120 140
V (V) Channel
Figure 2. Nine-element detector. Each detector has an active area of 2 mm × 2 mm. Current vs. voltage and thermal neutron response
are shown for the first column of the 3 × 3 set.
Figure 2 shows a picture of a 3-×-3 Transport in Pillar Structured Solid State and C. L. Cheung, “Pillar Structured Thermal
array of detectors integrated into the Thermal Neutron Detector,” International Neutron Detector With 6:1 Aspect Ratio,”
9-channel read-out system, showing the Semiconductor Device Research Conference, Appl. Phys. Lett., 93, p. 133502, 2008.
electrical and radiation response. Figure College Park, Maryland, December 12–14,
3 shows the final integrated system that 2007.
was fully characterized under this proj- 3. Nikolic, R. J., A. M. Conway,
ect. The next step is the participation C. E. Reinhardt, R. T. Graff, T. F. Wang, N. Deo,
of our detector in the “DNDO Neutron and C. L. Cheung, “Pillar Structured Thermal
Cookoff” tentatively scheduled for fall Neutron Detectors,” International Conference
2010. on Solid State and Integrated Circuit Technol-
ogy, Beijing, China, October 20–23, 2008.
Related References 4. Nikolic, R. J., C. L. Cheung, C. E. Reinhardt
1. Fabris, L., N. W. Madden, and H. Yaver, “A and T. F. Wang, “Roadmap for High Efficiency
Fast, Compact Solution for Low Noise Charge Solid-State Neutron Detectors,” SPIE – Inter-
Preamplifiers,” NIM-A 4242, pp. 545–551, national Symposium on Integrated Optoelec-
1999. tronic Devices, 6013, 1, pp. 36–44, 2005. Figure 3. Nine-channel COTS read-out
2. Conway, A. M., R. J. Nikolic, and 5. Nikolic, R. J., A. M. Conway, integrated with nine-element pillar
T. F. Wang, “Numerical Simulations of Carrier C. E. Reinhardt, R. T. Graff, T. F. Wang, N. Deo detector array.
Isotachophoretic Separation of
For more information contact:
Actinides Raymond P. Mariella, Jr.
(925) 422-8905
mariella1@llnl.gov
Isotachophoresis I.
t0 T** A* B* L
t1 T** T* B* A* B* AB A L
t2 T** T* B* B AB A L
t3 T** T* B AB A L
tdet
t4 T** T* T B AB A L
t5 T** T* T B A L tres
t6 T** T* T B A L
xdet xres
Sampling
compartment Separation compartment
µi(µL + µLC)
Ci = CL (2)
µL(µi + µLC)
Figure 1. Image of the front (top) and back (bottom) of microfluidic acoustic chip. The gray
region in the lower Image is the piezoelectric transducer that generates the acoustic waves
within the etched microchannels.
blood for device testing; 2) demonstrate white blood cells, is a critical step to end-to-end detection platform.
that the acoustic forces in the exist- maximize the selectivity and sensitiv-
ing devices are sufficient to focus and ity of downstream biological assays to FY2010 Accomplishments
separate white blood cells in a purified detect important markers for infec- and Results
sample (1X phosphate buffered saline tion and/or disease. To date, the best Accomplishments and results for this
(PBS) solution); and 3) demonstrate selective preparation results have been year include the following:
separation of white blood cells from achieved in laboratories where effec- 1. Implemented protocols for cultur-
whole blood. tive sample preparation using bench- ing and staining Raji cells that are B
top techniques, such as membrane lymphocytes from a leukemia cell
Relevance to LLNL Mission filtration, centrifugation, and chemical line with an average size of 7–8 µm
This work directly impacts ongoing methods, have been applied to the (Fig. 2).
and future efforts within LLNL’s Global samples. Our goal is to demonstrate a 2. Demonstrated focusing and separa-
Security Principal Directorate for new robust, automated sample preparation tion of white blood cells out of a
platforms for biosecurity applications. that reduces preparation time, improves phosphate buffered saline stream
Preparation of complex samples, such performance of downstream detection using the microfluidic acoustic
as the removal and extraction of target assays, and can be integrated into an device shown in Fig. 3.
3. Identified the need for a coating
(e.g., heparin) in the microfluidic
(a) (b) separator and associated fluidics to
reduce clogging due to adsorption
of material from the blood onto the
exposed surfaces.
Figure 2. (a) Stained white blood cells (Raji cells) used in acoustic separation experiments.
The cells are generally 7–8 µm in diameter and are shown next to (b) 5.78-µm fluorescent
particles for comparison.
le
mp
Image Sa
region
Input
Ce
lls
Figure 3. Acoustic focusing and extraction of white blood cells from an input sample stream (cells suspended in PBS). (a) Top view of
microfluidic chip near the bifurcating exit. The sample flows in from the left side in the upper half of the channel and exits the top if the
acoustic field is turned off and the bottom when the field is on. (b) Raji cells exiting through the upper outlet with the acoustic field turned
off. (c) Raji cells focused and exiting the lower stream when the acoustic field is turned on.
(detection) whether or not the targeted experimental data and comparing it to 3. Candy, J. V., D. H. Chambers, E. F.
RN is present. the GAMANAL software solution, where Breitfeller, B. L. Guidry, J. M. Verbeke, M. A.
Results of this photon-by-photon its detection rate of 98% easily exceeded Axelrod, K. E. Sale, and A. M. Meyer, “Model-
processor with downscatter are shown that of 47% both at essentially 0% false Based Detection of Radioactive Contraband
in Fig. 2. In the figure, three columns are alarm rate. These results demonstrate for Harbor Defense Incorporating Compton
shown. The first column is the compos- the potential capability of the sequen- Scattering Physics,” Proc. OCEANS09, IEEE
ite (not used) pulse-height spectrum tial Bayesian model-based approach to OES Soc., 2010.
(PHS), with the second the measured solving a variety of radiation-detection 4. Candy, J. V., D. H. Chambers, E. F.
photon energies (arrivals) in red circles problems. Breitfeller, B. L. Guidry, J. M. Verbeke, M.
with the green circles representing the A. Axelrod, K. E. Sale, and A. M. Meyer,
discriminator output photoelectrons and Related References “Radioactive Threat Detection with Scat-
the purple squares the discriminated 1. Candy J. V., E. F. Breitfeller, B. L. Guidry, tering Physics: a Model-Based Application,”
downscatter photons. Notice that they D. Manatt, K. E. Sale, D. H. Chambers, M. A. Proc. CIP, IEEE Comp. Soc., 2010.
align with the PHS energy “lines.” The Axelrod, and A. M. Meyer, “Physics-Based
final column is the decision function for Detection of Radioactive Contraband: A
each of the targeted RNs. As each pho- Sequential Bayesian Approach,” IEEE Trans.
ton is processed, the decision function Nuclr. Sci., 56, 6, pp. 3694–3711, 2009. FY2011 Proposed Work
is sequentially updated until one of the 2. Candy, J. V., D. H. Chambers, E. F. Teaming with ICx, a well-known,
thresholds (target/non-target) is crossed Breitfeller, B. L. Guidry, J. M. Verbeke, portable radiation detection system
(solid red box in figure) declaring a tar- M. A. Axelrod, K. E. Sale, and A. M. Meyer, manufacturer, an R&D 100 award
geted threat or non-threat. “Threat Detection of Radioactive Contraband was received for this project with a
The performance of the proces- Incorporating Compton Scattering Physics: high anticipation of “bringing it to
sor was substantiated by extracting a A Model-Based Processing Approach,” IEEE market” in the future.
100-member ensemble of controlled Trans. Nuclr. Sci., 57, 6, 2010.
1300 1300 20
Cobalt Photoelectric photons 60Co
10
1200 1200 Cobalt
0 detection
(5.76 s)
1100 1100 –10
Downscatter photons 0 1 2 3 4 5
1000 1000
Time (s)
900 900
20
800 Cesium 800
Volts (keV)
0
137Cs
500 500 0 1 2 3 4 5
Barium Time (s)
Photoelectric photons
400 400
300 300 30
20
200 200 Downscatter photons 10
133Ba
Barium detection
0 (0.46 s)
100 100 –10
Photoelectric photons
–20
0 0
40 30 20 10 0 2 3 4 5 6 0 1 2 3 4 5
Counts Time (s) Time (s)
Figure 2. Sequential Bayesian detection and identification. (a) Pulse-height spectrum (after calibration). (b) Photon arrivals (red circles)
with photoelectron discrimination (green circles) and downscatter photons (purple squares). (c) Decision functions for 60Co.
FY2010 Accomplishments
and Results
The acquisition of x-ray array source
data was obtained through a collabora-
tion with Triple Ring Technologies, a
Figure 1. Photograph of the x-ray array source and array detector system designed and research, development, and array source
developed by industrial collaborators NovaRay and Triple Ring. and detector system manufacturing
Attenuation
15000 Analytic
Window max at mean steel was analyzed in various multiplex-
10000
AMCG
5000 ing methods as well as the available
RSEM
0 single-source-at-a-time (single-series)
0 100
Pixel along lineout
illumination method. Single-series col-
lection methods were found to contain
Lineout – bar and jelly less noise for the data collection rates
89%* 63%*
Window max at max jelly
8000
95%* currently achievable by array source
Attenuation Analytic systems.
4000 AMCG
RSEM
0
0 50 100
Related Reference
Pixel along lineout De Man, B., S. Basu, D. Bequé, B. Claus,
* error reduction using RSEM P. Edic, M. Iatrou, J. LeBlanc, B. Senzig,
R. Thompson, M. Vermilye, C. Wilson, Z. Yin,
Figure 2. Comparison of algorithms for traditional 2-D cross-sectional data. and N. Pelc, “Multi-Source Inverse Geometry
CT: A New System Concept for X-Ray Comput-
ed Tomography,” Medical Imaging : Physics of
company. The partnership was formal- a cylindrically symmetric “As-Built” Medical Imaging Proc. of SPIE, 6510, 2007.
ized in FY2010. A Triple Ring Technolo- phantom, and a Contrast and Resolu-
gies system is presented in Fig. 1. tion Interleaved Stacked Plate (CRISP)
The project identified and tested phantom. The CRISP phantom was FY2011 Proposed Work
three reconstruction algorithms on 2-D designed and manufactured within this Instead of acquiring data by
array source data for algorithm testing project and metrology was performed to rotation and only in a single plane,
and comparison. The three algorithms determine the ground truth feature sizes project goals for FY2011 will
were: a traditional analytic method (Fil- and locations for system comparison. investigate a portable, field-ready
tered Back-Projection (FBP)); the Adjoint The data collected using the Triple Ring system by acquiring select views
Method Conjugate Gradient (AMCG) array source for both the “As-Built” and of objects from any vantage (i.e.,
algorithm developed at LLNL; and the CRISP phantoms were reconstructed region-of-interest (ROI) acquisi-
Ordered Subset Expectation Maximiza- with OSEM (Fig. 3). tion). Simulations will provide
tion (OSEM) method, which is used direction for object construction as
widely in the medical imaging field for well as choosing which views to use
emission tomographic reconstruction. (a) (b) for ROI imaging. To collect these
OSEM showed a significant impact data, new equipment and acquisi-
on reducing artifacts. Particularly evi- tion gantries will be necessary,
dent was a reduction in streak artifacts which will be developed and used
(>65% reduction in streak artifacts in through a collaboration with Triple
regions, as shown in Fig. 2). Although Ring and Stanford University.
OSEM was specifically included in this We will also expand on our
study because no modification was nec- work accomplished in FY2010 by
essary to reconstruct array source data, acquiring data on the previously
it has shown it can also improve the investigated phantoms using a 1-D
reconstruction of traditionally acquired array source system developed by
data as well (reduction in streak arti- Figure 3. OSEM reconstructed images XinRay. All systems (cone-beam,
facts, also shown in Fig. 2). for array source-acquired cross-sectional Triple Ring (2-D) array and XinRay
Two test objects were identified data on both the (a) as-built and (b) CRISP array) will then be evaluated.
and scanned with the array source: phantoms.
Number of arms
radius, powered by 10’s of modular new low-inductance, high-voltage
100
arms arranged in a circle. Lower energy capacitors and transmission lines, along
100 kV
with fast high-current switches. 200 kV
50
Figure 1 illustrates the device and con- 300 kV
400 kV
cept, along with some of the enabling
0
Z-pinch forms here technologies.
IV 6000
Peak I(kA)
DPF Gun
Relevance to LLNL Mission
4000
III The ability to reach plasma con-
ditions useful for LLNL missions in a 2000
1 to ~10 cm
Cathode
Anode
II
220
Peak I time (ns)
L0 200
C
Energy (kJ)
To DPF gun
Figure 2. Key parameters of the modular
Figure 1. Left: DPF gun driven by a switched capacitor bank. Right: New modular Marx driver concept in Fig. 1 as a function of
driver concept using new pulsed-power technologies. By shortening the pulse-width total pulser radius. The different curves
through low inductance and resistance, higher peak currents can be delivered at lower on each plot are for drivers with different
total energy and driver size. Marx setups.
Te[eV]
a lumped RLC circuit. 102 on the ~cm scale. The black dotted line is
Figure 2 provides short-circuit per- the electronic coupling parameter; below
formance of several drivers as a function the line the plasma is strongly coupled.
101 The red lines are the electron degeneracy
of size and voltage using modern com- Θ=10 Θ=0.1
Γe=1 parameter; below Θ = 0.1 degeneracy
ponents. These models were coupled to effects dominate, and above Θ = 10
a version of our three-phase DPF z-pinch 100
1018 1020 1022 1024 1026 1028
degeneracy effects are negligible. The green
model, which allowed us to perform dashed box indicates conditions of interest
ne [cm–3]
scans of engineering parameters to de- to NIF LPI studies.
termine achievable plasma conditions.
Figure 3 shows relevant plasma
regimes that can be achieved with a
2-m radius, 69-arm, 200-kV driver with
(a) (b)
various gun geometries. Data for plasma
regimes using loads for only 2 and 8 of
the 69 arms are also shown. Overall,
the modeling shows that it should be
possible to achieve relevant plasma
conditions with a low-energy machine
at tabletop dimensions on the order
of ~3–4 m due to newer pulsed-power ~9 in.
technology. This is an improvement over ~1 in.
previous leading machines with 8-m
diameter footprints.
We constructed, from existing
components, a small, up to 200-J DPF
device to enable experiments to validate
our model. Figure 4 shows the 1-arm
100-kV pulsed-power driver equipped Figure 4. (a) Small, 100-kV, 200-J pulsed-power driver and (b) DPF gun. The experiment
with a simple self-break oil switch and used existing equipment with the objective of validating our code at low energies.
a starting DPF gun load. Figure 5 shows
sample short circuit data from the driver
with a simple RLC fit, along with dI/dt
data indicating pinch formation and par- (a) (b)
x 104
ticle emission from initial experiments.
2.5 dl/dt
2.0 0.6
Related References Beam x5
1.5 0.4 Dip indicates
1. Tang, V., M. L. Adams, and B. Rusnak,
1.0 0.2 pinch
“Dense Plasma Focus Z-pinches for High
I(A)
AU
0.5
Gradient Particle Acceleration,” IEEE Transac- 0
0
tions on Plasma Science, 38, 4, pp. 719–727, –0.2
–0.5
2010. Data, Mret = 4 x 10–10
–1.0 –0.4
2. Decker, G., W. Kies, M. Malzig, et al., “High Fit
–1.5 –0.6
Performance 300 kV Driver Speed 2 for MA
0 100 200 300 –100 0 100 200 300 400 500
Pinch Discharges,” Nuclear Instruments and
Time (ns) Time (ns)
Methods, A249, 1986.
3. Soto, L., “New Trends and Future Perspec- Figure 5. (a) Short circuit current profile of the driver in Fig. 4 at 64 kV and 82 J. A peak
tives on Plasma Focus Research,” Plasma current of ~25 kA was reached with a rise time of ~100 ns. (b) Raw dI/dt data from the DPF
Phys. Cont. Fusion, 47, 5A, p. A361, 2005. gun in Fig. 4.
Threshold 1
Photons
No decision
Decision function
(take another
sample)
Threshold 0
No target
Time
Target
our partners in industry to find solutions absent
for the problems involved in monitoring 0 10 20 30 40 50 60 70 80 90 100
for illicit proliferation activities as well Trial number (n)
as monitoring for treaty verification. To
Figure 3. SRaDS detection results using “Target Present” data.
that end, the SRaDS system was cre-
ated to help address these challenging
issues.
Timing 95-GHz
Transmitter
board and IMPATT
control
processor diode Tx RF signal
Diagnostic
Circulator Antenna
computer
Rx RF signal
Video Sampling Envelope
board receiver detector
Toward Understanding
For more information contact:
Higher-Adaptive Systems Brenda M. Ng
(925) 422-4553
ng30@llnl.gov
a0
Agent i only Agents i and j
Update
a1 belief
a0
Does i model Does i/j model
a0 Update j as learning? j/i as learning?
a1 belief
Update
a0 S2* Y N S4 Y N
a1
belief Update ai,optimal
a1 belief Does i model j as Does i/j model j/i as
a0
bt–1
i,l Update using the correct (static) using the correct (static)
a1 belief model for itself? model for itself?
bt–1
i,l bti,l
Update Y N Y N
zi S5*
belief S3 S6
Does j model
i as learning?
Resample Propagate
Weight bti,l
S7 Y N
bt–1
j,l–1 bt–1
j,l–1 btj,l–1 Does j assume i is using
[s, ] [s, ] [s, ] the correct (static) model
for itself? Notes
Y N * Agent j assumes that Agent i
solve Update
is using the correct (static)
Policy aj,optimal belief
S1 S8 model for itself.
Figure 2. Representation of beliefs, approximated by samples Figure 3. Nine scenarios are considered, differing according to
over the physical states and “counts” that parametrize the model. which agent is learning and what it assumes of its opponent’s
Beliefs are updated via interactive particle filtering, which entails capabilities.
a recursive procedure.
state, action, and observation spaces In FY2009, we introduced approxi- average rewards and model conver-
are finite and known, but the model mations such as the interactive particle gence depend on how accurately the
parameters (namely, the state transi- filter (I-PF) to address the curse of agent models its opponent’s learning
tion probabilities and the observation dimensionality (belief complexity that process; and 3) the parameter learn-
probabilities) are not fully known. We increases with the number of states), ing introduced by our BA-IPOMDP
include these parameters as part of and reachability tree sampling (RTS) significantly improved the agent’s initial
the state space and maintain beliefs, in to address the curse of history (policy estimate of the model parameters while
the form of probability distributions, to complexity that increases with number maintaining reasonable runtime. The
represent our uncertainty about these of time steps or horizons in the decision ability to incorporate learning into these
parameters and the physical state of the process). models brings us one step closer to
environment. In FY2010, we optimized I-PF to their applicability in realistic adversarial
At each time step, an agent searches compute an approximation to the BA- modeling.
for and executes the optimal action belief, and we used RTS to construct
given its current belief. When it receives a pruned version of the finite-horizon Related References
an observation, it uses this observation reachability tree and applied dynamic 1. Ng, B., C. Meyers, K. Boakye, and
to iteratively refine its estimate of the programming to search this tree for the J. Nitao, “Towards Applying Interactive
state transition and observation models, optimal policy (see Fig. 2). POMDPs to Real-World Adversary Model-
and applies these learned models to To evaluate the performance of ing,” Proceedings of the 22nd Innovative
update its belief about the physical these methods, we tested our algo- Applications of Artificial Intelligence Confer-
state of the environment. During this rithms against a popular academic ence, pp. 1814–1820, 2010.
process, the agent must anticipate the problem, known as the multiagent Tiger 2. Doshi, P., and P. Gmytrasiewicz, “Monte
action, observation, and belief update problem, over multiple scenarios (see Carlo Sampling Methods for Approximating
of its opponent. Thus, belief updates are Fig. 3). Runtime was recorded for vary- Interactive POMDPs,” Journal of Artificial In-
recursive. ing resolutions of the approximation telligence Research, 34, pp. 297–337, 2009.
To solve an IPOMDP is to determine algorithms; average rewards and model 3. Ross, S., B. Chaib-draa, and J. Pineau,
the optimal policy, which for every pos- parameter convergence was compared “Bayes-Adaptive POMDPs,” Proceedings
sible observation, produces an optimal for different learning scenarios. of the 21st Neural Information Processing
action that maximizes the agent’s Baselining against the standard Systems, pp. 1225–1232, 2008.
expected reward. Previous work has IPOMDP, we found that 1) the agent
shown that value iteration can be used quickly converges on a model which
to solve IPOMDPs. is slightly off the true model; 2) the
Extr. 2
tr.
Existing Data
(truth known)
gua
ge
Nonproliferation, Counterterrorism,
a
Learner Plugin ngu and other national security missions
La
Hyp. Ranker Plugin rely on the acquisition of knowledge
that is buried in unstructured text docu-
Likelihood
Pattern
ce
BMA
Pattern
ce
(probabilistically
BMA
ranked)
ment by LLNL and its customers that
Seq
Calibration Aggregation
must automatically extract critical
information from these sources. To
Figure 1. Meta-extraction system. enable effective knowledge discovery,
F measure
Engineering Systems for Knowledge and 0.85 0.84
Inference (ESKI) Text to Inference area 0.82
and the Cyber, Space, and Intelligence 0.80
0.80
strategic mission thrust in the LLNL
Five-year Strategic Roadmap. Successful 0.75 0.78
completion of this research will pro- 0.76
vide highly valued and unprecedented 0.70
BAILE GATE LP SNER X-Man BAILE GATE LP SNER X-Man
inference and decision-making capabili-
ties to internal programs, such as IOAP
and CAPS, and to external customers Figure 2. Box plots showing bootstrapped samples of the weighted mean of F measure.
Extractors and X-Man were trained on MUC6 (Message Understanding Conferences).
such as DHS, DoD, and the intelligence
Results are shown for testing on (a) MUC6 and (b) CoNLL-2003.
community.
FY2010 Accomplishments
and Results X-Man incorporates machine learn- assigns a probability to each ground
Insights gained in event extraction ing and probabilistic methods, ranging truth hypothesis. The resulting relative
error analyses performed in FY2008 from classical probability to Bayesian ranking of hypotheses for each piece of
motivated a graduated approach to Model Averaging, into several novel al- text provides not only an ordered list
triple/event aggregation that is founded gorithms, each consisting of a calibration of the most probable ground truths,
fundamentally on the aggregation of component coupled with an aggregation but a mechanism for determining those
extracted entities. component (Fig.1). hypotheses that are significantly more
To this end, we have developed For each of these algorithms, calibra- likely in a statistical sense. Thus, the
a novel text extraction management tion hinges on the estimation of prob- ranking informs downstream decision-
methodology (“X-Man”) focused on ability distributions over a joint hierar- making and analysis by enabling confi-
entity extraction that can be generalized chical error space arising from the suite dence assessments of extracted data.
to multiscale triples (i.e., simple events) of underlying extractors. The generated Figure 2 shows sample test results.
and more complex event aggregation error distributions characterize each X-Man has been shown to 1) pro-
solutions. X-Man is a flexible, general- extractor’s performance relative to dis- duce statistically significant improve-
ized framework for the aggregation of joint regions of contiguous text. Perfor- ments in extraction relative to standard
named entity extraction technologies mance takes into account joint extractor performance metrics (up to 120%
that uses the joint characteristics of characteristics, as well as the statistical improvement under certain operating
its constituent extractors’ output to behaviors (both individual and joint) of conditions); 2) be able to reconstruct
aggregate extracted text. Hence, exist- the errors occupying the defined error truth when all of its constituent extrac-
ing extraction tools (e.g., commercial, space. These algorithms are particu- tors fail; and 3) provide a framework
academic) can be readily incorporated larly distinguished by their reliance on for quantifying uncertainty in extracted
to enhance the quality of extracted data. a range of underlying models and/or output. Moreover, mechanisms have
Moreover, the methodology has been assumptions. Accordingly, X-Man’s final been developed to help X-Man adapt to
designed to enable the incorporation of stage of calibration involves using state- sparse data conditions.
new extractor evaluation and aggrega- of-the-art machine learning methods
tion methodologies, as well as language to determine an optimal deployment Related Reference
modules that can leverage language- strategy for incoming data. Lemmond, T., N. Perry, J. Guensche, J. Nitao,
specific resources such as gazetteers, When newly extracted output data R. Glaser, P. Kidwell, P., and W. Hanley,
stop word lists, and parsers. This unprec- are encountered, the X-Man System “Enhanced Named Entity Extraction via Error-
edented level of flexibility makes X-Man constructs a space of hypotheses over Driven Aggregation,” International Confer-
highly customizable and adaptable to a ground truth for each piece of contigu- ence on Data Mining, Las Vegas, Nevada, July
wide range of applications and problem ous text and then deploys the text to its 2010.
domains. optimal aggregation algorithm, which
Score distribution for Class 0 static classifier over the low false alarm
Threshold rate regions of the receiver operating
µ0 characteristic (ROC) curve.
False
σ02 alarms Increase class 1 Related References
classifier Cost-sensitive ensemble
Detection rate
agreement using error bound results 1. Chen, B. Y., T. D. Lemmond, and
0 1 W. G. Hanley, “Building Ultra-Low False Alarm
III II I
µ1 Rate Support Vector Machine Ensembles
Detections Using Random Subspaces,” Proc. IEEE Sympo-
Conventional
σ12 sium on Computational Intelligence and Data
Cost-sensitive
ensemble Mining, 2009.
0 1 False alarm rate 2. Lemmond, T. D., W. G. Hanley, L. J. Hiller,
Score distribution for Class 1 D. A. Knapp, M. J. Mugge, and B. Y. Chen,
“Discriminant Random Forest,” U.S. Provi-
Figure 1. Class-specific error bounds governing ensemble classification performance. sional Patent Filed, May 2008.
Increasing classifier agreement in the ensemble can lead to higher detection rates at 3. Lemmond, T. D., A. O. Hatch, B. Y. Chen,
extremely low false alarm rates. D. A. Knapp, L. J. Hiller, M. J. Mugge, and
W. G. Hanley, “Discriminant Random
Forests,” Proceedings of 2008 International
Conference on Data Mining, 2008.
fruit on a very difficult hidden signal a 32.5 × speedup using two orders of 4. Lemmond, T. D., B. Y. Chen, A. O. Hatch,
detection application, where our new magnitude fewer bytes while maintain- and W. G. Hanley, “An Extended Study of the
classifiers were able to achieve a previ- ing high classification accuracy on an Discriminant Random Forest,” Data Mining
ously unattainable level of performance: important cyberthreat detection applica- Special, Annals of Information Systems, 8,
non-zero detection rates at 0% false tion (see Table). 2010.
alarms. To address the challenge of changing 5. Prenger, R. J., T. D. Lemmond,
In applications where computational data distributions, we have developed K. R. Varshney, B. Y. Chen, and W. G. Hanley,
resources are scarce, it is extremely im- several new approaches for dynamic “Class-Specific Error Bounds for Ensemble
portant to have ensemble classifiers that density estimation. The Forest Based Classifiers,” Proc. ACM SIGKDD Conference on
are compact and fast. We developed Density Estimator (FBDE) is an approach Knowledge Discovery and Data Mining, July
a Bayesian Random Forest to provide inspired by Random Forests whose trees 2010.
high-throughput classification using randomly partition feature space into
orders of magnitude fewer bytes. Un- appropriately sized hyper-rectangles.
Cyberthreat detection ROC curves
like traditional Random Forests whose An ensemble of these trees allows for 1.0
trees are grown until homogeneity in a robust estimate of the probability
data class is reached, Bayesian Random density of high dimensional data. As 0.9
Detection rate
Forests stop growing when the data no data distributions change, the trees in
longer justifies the increased complexity the forest evolve to track changes in the 0.8
of adding another layer of split nodes. data. Using dynamic density estimators,
0.7
The Bayesian Random Forest achieved we developed a classification system
that is able to maintain high detection 0.6 Static 2-class
classifier
rates at low false alarm rates even as the
Bayesian Random Forest compared to a Dynamic 2-class
underlying detection pattern changes. 0.5 classifier
standard Random Forest for cyberthreat
detection.
In Fig. 2, we compare a standard
static classifier to our dynamic classifi- 0.4
10–3 10–2 10–1 100
Standard RF Bayesian RF cation system on a cyberthreat detec- False alarm rate
Memory 16 MB 0.021 MB tion application where the threat class
Classification undergoes six different changes during Figure 2. Our dynamic classifier, which out-
195 6
speed (s) the experiment. Our dynamic classifier performs the traditional static classifier in
Avg. class
99.1% 98.7%
successfully tracks the changing threat cyberthreat detection when the threat class
accuracy class and significantly outperforms the undergoes six distinct changes.
Tracy D. Lemmond
(925) 422-0219
lemmond1@llnl.gov
Search
Relevance to LLNL Mission
[New Experiment] [Query Experiments] [Root Folder] [Team] [Manage]
Nonproliferation, counterterrorism,
and other national security missions
rely on the acquisition of knowledge
Main Dispatch Alg 0 Alg 1 Alg 2 Alg 3
that is buried within unstructured
NOTE: Alg 0 must be a Pattern Algorithm.
Algorithm Dispatcher
text documents too numerous to be
Push Condition Alg to manually processed. Systems are being
Run
Pattern not found in Pattern Dictionary Alg 1 constructed by LLNL and its customers
Push when Pattern Algorithm encounters punctuation errors
... and the number of extractors used
is greater than
1 Alg 2 that must automatically extract entities
Meta-Entity length greater than (set to -1 to disable) 2 Meta-Entity length type SegmentBased Alg 3
from these sources, and methodologies
Relative Difference Between top two hypotheses less than
or equal to (-1 to disable)
1 Alg 2
have been produced that significantly
Total # votes (all hypotheses) less than (set to -1 to disable) Alg 1
WASHINGTON -- The flier whose Navy F-14A fighter plunged into a Nashville suburb on Monday, killing himself and four other people, crashed another jet into the sea last April.
But Navy investigators and senior admirals forgave him, saying he made a mistake in pursuit of the combative flying that the Navy wants and encourages in its pilots.
The flier, Lt. Comdr.
John Stacy Bates, flew aggresively, a Navy official said on Tuesday, but he added: “We want them to fly aggresively.
Bates was highly motivated and that accident was a one-time glitch on his record.
He was a great aviator.”
The Navy invests years and more than $1 million to train each of its fighter pilots, and is reluctant to dismiss them if senior officers believe an erring pilot can learn from mistakes.
But as military investigators sifted through the wreckage on Tuesday for clues to what caused the crash that killed the fighter’s two-man crew and three people on the ground, Navy officials said they did not know what caused Bates’
second crash, or why his squadron had lost so many F-14 Tomcats.
The crash was the fourth in 16 months for Fighter Squadron 213, a 14-plane unit known as the Fighting Blacklions and one of six F-14 squadrons assigned to Miramar Naval Air Station near San Diego.
The unit’s safety record is by far the worst among the Navy’s 13 F-14 squadrons.
Bates was blamed for losing control of his F-14 last April while conducting training maneuvers off Hawaii.
Last September, and F-14A from the squadron exploded in flight off the Philippines, but both crew members ejected safely.
The cause of that accident is still under investigation.
In October 1994, one the Navy’s first female fighter pilots, Lt. Kara S. Hultgreen, died in a training accident off Southern California, rekindling tensions within the military over the decision to expand some combat roles for women.
The Navy concluded that that accident resulted from a combination of pilot error and mechanical failure.
“You go back 10 or 15 years and they are snake bit,” said a retired admiral who once commanded the squadron.
“We’ve tried to put top-notch pilots and maintenance people there.
You can’t believe in luck or superstition, but they’re behind the eight ball and have stayed there.”
The Navy ordered the squadron to suspend its operations for three days for safety reasons after the second of the squadron’s four crashes.
Vice Adm. Brent Bennitt, the commander of naval air forces in the Pacific, immediately ordered the squadron to stand down again after the crash on Monday to review its safety record and prodedures.
The crash underscores the fact that even in peacetime, operating complex weapons of war is a hazardous business.
Twelve F-14 fliers have died in training accidents since 1992.
But the accident also raises questions about the F-14’s safety record.
Since 1991, the fighter has a major crash rate of 5.93 per 100,000 flight hours, compared with 4.82 major crashes per 100,000 hours for all Navy tactical aircraft.
Navy officials note that since 1981, the F-14’s major accident rate is slightly lower than the overall tactical aircraft rate.
Many naval aviators have complained that the engines on the older A-model F-14’s are not powerful enough to perform the demanding aerial maneuvers they fly.
The Navy is replacing them with a more powerful engine that is now on about 30 percent of the fleet’s F-14’s. Fighter Squadron 213 flies all A-model F-14’s. In the latest accident, the twin-engine, two-seat Tomcat crashed shortly
after takeoff from Berry Field, an Air National Guard airfield adjacent to Nashville International Airport.
The jet left Miramar Air Station in San Diego for Nashville on Friday on a routine training mission.
Bennit said on Tuesday that Navy officials approved Bates’ request to use a maximum-performance takeoff, in which a pilot turns on the jet’s after-burner and soars straight up moments after the aircraft leaves the ground
This effort directly supports collectively allow 1) joint character- these folds in parallel for increased
Engineering Systems for Knowledge ization of extracted data relative to efficiency. When an experiment has
and Inference (ESKI) Text to Inference various features of interest (e.g., entity been completed, the user is provided
R&D area and the Cyber, Space, and length, data sparseness, entity type); with an array of statistics associated
Intelligence strategic mission thrust in 2) performance evaluation (in terms of with the execution. These include
the LLNL Five-year Strategic Roadmap. F-measure, exact match, miss and/or 1) error counts and probability esti-
The completed system will provide false alarm) of aggregation algorithms mates for the base and meta-extractor
highly valued and unprecedented entity and extractors relative to these data algorithms; 2) detailed output of the
extraction capabilities to internal pro- characteristics; and 3) either manual or entities extracted by the base extrac-
grams, such as IOAP and CAPS, and to automatically optimized deployment of tors, the space of hypothesized ground
external customers such as DHS, DoD, data to the corresponding aggregation truths proposed by the meta-extractor,
and the intelligence community. algorithms. and the corresponding meta-extractor
Figure 1 shows an example of result; 3) the rate that events of inter-
FY2010 Accomplishments manual dispatching of extracted entity est occur (e.g., the frequency that the
and Results data to four different algorithms that meta-extractor recreates the truth
The entity extractor aggregation may be defined by the user. These when all base extractors fail); and 4) the
tool was originally constructed to serve algorithms may include any of several original text with extracted and ground
as both a prototype of first genera- aggregation algorithms (e.g., Pattern- truth entities highlighted (Fig. 2).
tion aggregation methodologies (i.e., based, Bayesian Model Averaging), This information collectively
meta-extraction algorithms) and as an or the base extractors themselves, if provides substantial insight into the
environment for the incorporation of desired. Automatic deployment, when behaviors and performance of the base
these advancements. selected, is optimized using state-of-the- extractors, as well as of X-Man itself,
In FY2010, the final prototype, art machine learning techniques (e.g., enabling the potential for algorithm
called the Extraction Manager (X-Man), ensemble predictors) to determine an optimization and enhancement.
contains not only more advanced optimal mapping of extracted data fea-
aggregation algorithms that improved tures to appropriate aggregation and/or Related Reference
the effectiveness of entity extractor base algorithms. Lemmond, T., N. Perry, J. Guensche, J. Nitao,
aggregation, but also additional features Performance estimation takes place R. Glaser, P. Kidwell, P., and W. Hanley, “En-
that enhanced the flexibility provided through cross-validation, in which the hanced Named Entity Extraction via Error-
to its users. Specifically, the new meta- data are partitioned into multiple folds Driven Aggregation,” International Confer-
extraction system consists of various with associated performance estimates ence on Data Mining, Las Vegas, Nevada,
modules, linked through a central that are typically averaged to obtain an July 2010.
component called the dispatcher, that overall estimate. The X-Man tool runs
2100 generators Load, storage, reserves, Grid dispatch mixed integer programming model with
across Western U.S. transmission requirements 225,000 variables (34,000 integer) and 400,000 constraints
Figure 1. Composition of grid dispatch model currently used for planning in California.
Surface view
Insulator
Insulator
Vacuum
Vacuum
Vacuum
+ angle
y (mm)
0 0 0
25 5 50
z (mm) x (mm) z (mm)
Surface view
Insulator
Insulator
Vacuum
Vacuum
Vacuum
– angle
Figure 2. Simulation results for positive 55° (top) and negative 30° (bottom) angle insulators. The red dots represent ionized gas and the
blue are electrons. The two views are looking across (left) and directly at (right) the insulator surface. The graph on right is a color contour
of the electric field magnitude (red is low and blue is high). The anode is at the top and the cathode is the bottom for each plot.
Author Index
Available from
National Technical Information Service
Graphic Designers 5285 Port Royal Road
Jeffrey B. Bonivert
Springfield, VA 22161
Lucy C. Dobson
Debbie A. Ortega
Kathy J. Seibert
Or online at www-eng.llnl.gov/pubs.html
This work was performed under the auspices of the U.S. Department of Energy
by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.
ST-10-0070
FY10
Engineering Innovations,
Research & Technology Report
April 2011
Lawrence Livermore National Laboratory
PO Box 808, L-151
Lawrence Livermore National Laboratory
Livermore, CA 94551-0808
http:www-eng.llnl.gov/ LLNL-TR-468271