The National Ignition Facility (NIF), the world’s most energetic laser system and the first to demonstrate fusion ignition, routinely operates at energies that can damage its final optics. To enable sustained operation, NIF recycles optics by ablating fractured material associated with damage, leaving behind a benign, cone-shaped void. These mitigation cones range in depth and diameter and typically are applied using the smallest effective cone for one damage site. However, when multiple damage sites are closely situated, various combinations of larger and smaller cones can be used to repair the region, and the number of options grows exponentially. Standard brute force approaches that iterate through each of these possibilities are thus computationally impractical beyond a relatively low threshold. To overcome these limitations, we introduce Combinatorial Optimization for OPtic Repair (COOPR), a novel combinatorial optimization framework to solve the problem of cone placement given any configuration of damage sites. Using tools from the seemingly unrelated literature on facility location problems in urban planning, we formulate and solve mixed-integer linear programs that identify optimal cone configurations for damage mitigation with respect to a multi-objective cost function. We show that even for optics with hundreds of clustered damage sites, COOPR finds more effective cone placements faster than existing approaches, thus enabling a more efficient optic mitigation cycle with a reduced need for human intervention.
Identifying laser induced damage on the surface of optical components for the purpose of tracking its growth over time and repairing it is an important part of the economical operation of the National Ignition Facility (NIF). Optics installed on NIF are monitored in situ for damage growth and can be removed as needed for repair and re-use. An ex-situ automated microscopy system is used to inspect full sized NIF optics allowing for the detection of damage sites <10 μm in diameter. Due to the various morphology of laser damage, several algorithms are used to analyze the microscopy data and identify damage regardless of size, while ignoring features not related to laser damage. This system has significantly increased the lifetime of NIF final optics (≈2.3x) thereby extending beyond the capabilities of the in-situ inspection by itself.
Overview of machine learning methods and the practical steps needed for implementation. We will use examples from NIF to demonstrate steps from data preparation to the pros and cons of different machine learning methods.
The National Ignition Facility in northern California routinely operates at twice the intensity (fluence) known to damage fused silica optics. With this in mind, the facility was designed and built with removable optic modules that allow for optic exchanges which in turn enable an optics "recycle loop" to extend the life of highly specialized optics.
The recycle loop includes automated optics inspection whereby every damage site is identified, measured, tracked through time, protected (once it approaches an optic-specific size limit), and then repaired in a laboratory so the optic can be reused.
Here we describe an overview of custom image analysis, machine learning, and deep learning methods used throughout the recycle loop for optics inspection on the NIF beamlines and off. Since 2007 we’ve used machine learning to improve accuracy and automate tedious processes to enable and inform an efficient optics recycle loop.
This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL-CONF-749953
The National Ignition Facility (NIF) uses an in-situ system called the Final Optics Damage Inspection (FODI) system to monitor the extent of damage on installed optical components. Among this system's uses is to alert operators when damage sites on a Grating Debris Shield (GDS) require repair (≈300 microns) and triggers the removal of the damaged optic. FODI, which can reliably detect damage sites larger than 50 microns, records the size and location of observed sub-critical damage observed on the optic, so each of these sites can be repaired before the optic is next installed. However, by only identifying, and hence repairing sites larger than ≈50 microns, optics are left with numerous smaller sites, some fraction of which resume growing when the host optic is reinstalled. This work presents a method of identifying and repairing damage sites below the FODI detection limit that have a significant probability of growth. High resolution images are collected of all likely damage candidates on each optic, and a machine learning based automated classification algorithm is used to determine if each candidate is a damage site or something benign (particle, previously repaired site, etc.). Any damage site greater than 20 microns is flagged for subsequent repair. By repairing these smaller sites, recycled optics had a 40% increased lifetime on the NIF.
The National Ignition Facility (NIF) regularly operates at fluences above the onset of laser-induced optics damage. To do so, it is necessary to routinely recycle the NIF final optics, which involves removing an optic from a beamline, inspecting and repairing the laser-induced damage sites, and re-installing the optic. The inspection and repair takes place in our Optics Mitigation Facility (OMF), consisting of four identical processing stations for performing the repair protocols. Until recently, OMF has been a labor-intensive facility, requiring 10 skilled operators over two shifts to meet the throughput requirements. Here we report on the implementation of an automated control system—informed by machine learning— that significantly improves the throughput capability for recycling of NIF optics while reducing staffing requirements. Performance metrics for mid-2018 show that approximately 85% of all damage sites can be automatically inspected and repaired without any required operator input. Computer keystrokes have been reduced from about 6000 per optic to under 300.
Two machine-learning methods were evaluated to help automate the quality control process for mitigating damage sites on laser optics. The mitigation is a cone-like structure etched into locations on large optics that have been chipped by the high fluence (energy per unit area) laser light. Sometimes the repair leaves a difficult to detect remnant of the damage that needs to be addressed before the optic can be placed back on the beam line. We would like to be able to automatically detect these remnants. We try Deep Learning (convolutional neural networks using features autogenerated from large stores of labeled data, like ImageNet) and find it outperforms ensembles of decision trees (using custom-built features) in finding these subtle, rare, incomplete repairs of damage. We also implemented an unsupervised method for helping operators visualize where the network has spotted problems. This is done by projecting the credit for the result backwards onto the input image. This shows regions in an image most responsible for the networks decision. This can also be used to help understand the black box decisions the network is making and potentially improve the training process.
A challenging aspect of preparing cryogenic targets for National Ignition Facility (NIF) ignition experiments is growing a single crystal layer (~ 70 m thick) of solid frozen deuterium-tritium (DT) fuel on the inner surface of a spherical hollow plastic capsule 2 mm in diameter. For the most critical fusion experiments, the layer must be smooth, having uniform thickness, and largely free of isolated defects (e.g. grooves). A single target layer typically takes up to 18 hours to form. X-ray images on 3 orthogonal axes are used to monitor the growth of the crystal and evaluate the quality of the layer. While these methods provide a good indicator of target layer condition, new metrics are currently being developed to take advantage of other properties in the x-ray image, which may give earlier indications of target quality. These properties include symmetry of texture, seed formation, and eigenimage analysis. We describe the approach and associated image processing to evaluate and classify these metrics, whose goal is to improve overall layer production and better quantify the quality of the layer during its growth.
KEYWORDS: Machine learning, Inspection, Optical inspection, Data modeling, Image analysis, Image processing, Reflection, Data mining, National Ignition Facility, Signal processing
The Final Optics Damage Inspection (FODI) system automatically acquires and utilizes the Optics
Inspection (OI) system to analyze images of the final optics at the National Ignition Facility (NIF). During each
inspection cycle up to 1000 images acquired by FODI are examined by OI to identify and track damage sites on the
optics. The process of tracking growing damage sites on the surface of an optic can be made more effective by
identifying and removing signals associated with debris or reflections. The manual process to filter these false sites
is daunting and time consuming. In this paper we discuss the use of machine learning tools and data mining
techniques to help with this task. We describe the process to prepare a data set that can be used for training and
identifying hardware reflections in the image data. In order to collect training data, the images are first
automatically acquired and analyzed with existing software and then relevant features such as spatial, physical and
luminosity measures are extracted for each site. A subset of these sites is "truthed" or manually assigned a class to
create training data. A supervised classification algorithm is used to test if the features can predict the class
membership of new sites. A suite of self-configuring machine learning tools called "Avatar Tools" is applied to
classify all sites. To verify, we used 10-fold cross correlation and found the accuracy was above 99%. This
substantially reduces the number of false alarms that would otherwise be sent for more extensive investigation.
The National Ignition Facility (NIF) at the Lawrence Livermore National Laboratory (LLNL)
routinely fires high energy shots (> 6 kJ per beamline) through the final optics, located on the
target chamber. After a high fluence shot, exceeding 4J/cm2 at 351 nm wavelength, the final
optics are inspected for laser-induced damage. The FODI (Final Optics Damage Inspection)
system has been developed for this purpose, with requirements to detect laser-induced damage
initiation and to track and size it's growth to the point at which the optic is removed and the site
mitigated. The FODI system is the "corner stone" of the NIF optic recycle strategy. We will
describe the FODI system and discuss the challenges to make optics inspection a routine part of
NIF operations.
Phase-defects on optics used in high-power lasers can cause light intensification leading to laser-induced damage of
downstream optics. We introduce Linescan Phase Differential Imaging (LPDI), a large-area dark-field imaging
technique able to identify phase-defects in the bulk or surface of large-aperture optics with a 67 second scan-time.
Potential phase-defects in the LPDI images are indentified by an image analysis code and measured with a Phase
Shifting Diffraction Interferometer (PSDI). The PSDI data is used to calculate the defects potential for downstream
damage using an empirical laser-damage model that incorporates a laser propagation code. A ray tracing model of LPDI
was developed to enhance our understanding of its phase-defect detection mechanism and reveal limitations.
Lawrence Livermore National Laboratory is a large, multidisciplinary institution that conducts fundamental
and applied research in the physical sciences. Research programs at the Laboratory run the
gamut from theoretical investigations, to modeling and simulation, to validation through experiment.
Over the years, the Laboratory has developed a substantial research component in the areas of signal
and image processing to support these activities. This paper surveys some of the current research in
signal and image processing at the Laboratory. Of necessity, the paper does not delve deeply into any
one research area, but an extensive citation list is provided for further study of the topics presented.
Laser-induced damage growth on the surface of fused silica optics has been extensively studied and has
been found to depend on a number of factors including fluence and the surface on which the damage site
resides. It has been demonstrated that damage sites as small as a few tens of microns can be detected and
tracked on optics installed a fusion-class laser, however, determining the surface of an optic on which a
damage site resides in situ can be a significant challenge. In this work demonstrate that a machine-learning
algorithm can successfully predict the surface location of the damage site using an expanded set of
characteristics for each damage site, some of which are not historically associated with growth rate.
In many high energy laser systems, optics with HMDS sol gel antireflective coatings are placed in close proximity to
each other making them particularly susceptible to certain types of strong optical interactions. During the coating
process, halo shaped coating flaws develop around surface digs and particles. Depending on the shape and size of the
flaw, the extent of laser light intensity modulation and consequent probability of damaging downstream optics may
increase significantly. To prevent these defects from causing damage, a coating flaw removal tool was developed that
deploys a spot of decane with a syringe and dissolves away the coating flaw. The residual liquid is evacuated leaving an
uncoated circular spot approximately 1mm in diameter. The resulting uncoated region causes little light intensity
modulation and thus has a low probability of causing damage in optics downstream from the mitigated flaw site.
The National Ignition Facility (NIF) at the Lawrence Livermore National Laboratory (LLNL) will routinely
fire high energy shots (approaching 10 kJ per beamline) through the final optics, located on the target
chamber. After a high fluence shot, exceeding 4J/cm2 at 351 nm wavelength, the final optics will be
inspected for laser-induced damage. The FODI (Final Optics Damage Inspection) system has been
developed for this purpose, with requirements to detect laser-induced damage initiation and to track and size
it's the growth to the point at which the optic is removed and the site mitigated. The FODI system is the
"corner stone" of the NIF optic recycle strategy. We will describe the FODI system and discuss the
challenges to make optics inspection a routine part of NIF operations.
Many automated image-based applications have need of finding small spots in a variably noisy image. For
humans, it is relatively easy to distinguish objects from local surroundings no matter what else may be in
the image. We attempt to capture this distinguishing capability computationally by calculating a
measurement that estimates the strength of signal within an object versus the noise in its local
neighborhood. First, we hypothesize various sizes for the object and corresponding background areas.
Then, we compute the Local Area Signal to Noise Ratio (LASNR) at every pixel in the image, resulting in
a new image with LASNR values for each pixel. All pixels exceeding a pre-selected LASNR value
become seed pixels, or initiation points, and are grown to include the full area extent of the object. Since
growing the seed is a separate operation from finding the seed, each object can be any size and shape. Thus,
the overall process is a 2-stage segmentation method that first finds object seeds and then grows them to
find the full extent of the object.
This algorithm was designed, optimized and is in daily use for the accurate and rapid inspection of optics
from a large laser system (National Ignition Facility (NIF), Lawrence Livermore National Laboratory,
Livermore, CA), which includes images with background noise, ghost reflections, different illumination
and other sources of variation.
That National Ignition Facility (NIF) at Lawrence Livermore National Laboratory (LLNL) will be the world's
largest and most energetic laser. It has thousands of optics and depends heavily on the quality and performance of these
optics. Over the past several years, we have developed the NIF Optics Inspection Analysis System that automatically
finds defects in a specific optic by analyzing images taken of that optic.
This paper describes a new and complementary approach for the automatic detection of defects based on
detecting the diffraction ring patterns in downstream optic images caused by defects in upstream optics. Our approach
applies a robust pattern matching algorithm for images called Gradient Direction Matching (GDM). GDM compares
the gradient directions (the direction of flow from dark to light) of pixels in a test image to those of a specified model
and identifies regions in the test image whose gradient directions are most in line with those of the specified model. For
finding rings, we use luminance disk models whose pixels have gradient directions all pointing toward the center of the
disk. After GDM identifies potential rings locations, we rank these rings by how well they fit the theoretical diffraction
ring pattern equation. We perform false alarm mitigation by throwing out rings of low fit. A byproduct of this fitting
procedure is an estimate of the size of the defect and its distance from the image plane. We demonstrate the potential
effectiveness of this approach by showing examples of rings detected in real images of NIF optics.
To more accurately measure fluorescent signals from microarrays, we calibrated our acquisition and analysis systems by using groundtruth samples comprised of known quantities of red and green gene-specific DNA probes hybridized to cDNA targets. We imaged the slides with a full-field, white light CCD imager and analyzed them with our custom analysis software. Here we compare, for multiple genes, results obtained with and without preprocessing (alignment, color crosstalk compensation, dark field subtraction, and integration time). We also evaluate the accuracy of various image processing and analysis techniques (background subtraction, segmentation, quantitation and normalization). This methodology calibrates and validates our system for accurate quantitative measurement of microarrays. Specifically, we show that preprocessing the images produces results substantially closer to the known groundtruth for these samples.
Recent scientific studies evaluating laser energy for tissue welding and thermokeratoplasty have demonstrated that the application of laser energy at non-ablative levels can alter collagen's structural and biochemical properties. A recent pilot study has demonstrated that the non-ablative application of holmium: yttrium-aluminum-garnet (Ho:YAG) laser energy to the joint capsule of patients with glenohumeral instability shrank the joint capsule, stabilizing the shoulder in the majority of the patients treated. Based on the collective findings of these studies, we hypothesized that thermal modification of dense collagenous tissues such as joint capsule, ligament, and tendon can be achieved by applying non-ablative laser energy. The purpose of this study was to evaluate the effect of laser energy at non-ablative levels on joint capsular mechanical, biochemical, histological, and ultrastructural properties in an in vitro rabbit model.
Digital mammography offers the promise of significant advances in early detection of breast cancer. Our overall goal is to design a digital system which improves upon every aspect of current mammography technology: the x-ray source, detector, visual presentation of the mammogram and computer-aided diagnosis capabilities. This paper will discuss one part of our whole-system approach--the development of a computer algorithm using gray-scale morphology to automatically analyze and flag microcalcifications in digital mammograms in hopes of reducing the current percentage of false-negative diagnoses, which is estimated at 20%. The mammograms used for developing this 'mammographers assistant' are film mammograms which we have digitized at either 70 micrometers or 35 micrometers per pixel resolution with 4096 (12 bits) of gray level per pixel. For each potential microcalcification detected in these images, we compute a number of features in order to distinguish between the different kinds of objects detected.
KEYWORDS: 3D image processing, 3D metrology, Image segmentation, Confocal microscopy, Signal detection, Image processing, Microscopes, Analytical research, Data acquisition
Three-dimensional (3D) data are obtained on biological cells by collecting optical slices at different depths in the cell using confocal laser scanning microscopy. A method has been developed which allows such three-dimensional data to be automatically segmented so that objects within the cell can be identified and labelled. Segmentation allows an object in a field to be recognized as distinct from other objects and from the background. This is done using an algorithm which performs a cell-dependent sequence of image processing functions on each slice of the image data (2D). Identification requires that a single object be recognized as one entity through all of the slices of data (3D). This is done with a 3D labelling routine. In this application, cell nuclei are first defined in 3D, and then structures within them (i.e. chromosome domains) can be identified. Measurements are made on these domains for the purpose of determining the relationship between chromosomal locations and the possibility that they will exchange genetic information.
Medical researchers are seeking a method for detecting chromosomal abnormalities in unborn children without requiring invasive procedures such as anmiocentesis. Software has been developed to utilize a light microscope to detect fetal cells that occur with very low frequency in a sample of maternal blood. This rare event detection involves dividing a microscope slide containing a maternal blood sample into as many as 40,000 fields, automatically focusing on each field-of-view, and searching for fetal cells. Size and shape information is obtained by calculating a figure of merit through various binary operations and is used to discriminate fetal cells from noise and artifacts. Once the rare fetal cells are located, the slide is automatically rescanned to count the total number of cells on the slide. Binary operations and image processing hardware are used as much as possible to reduce the total amount of time to analyze one slide. Current runtime for scoring one full slide is about four hours, with motorized stage movement and focusing being the speed-limiting factors. Fetal cells occurring with a frequency of less than 1 in 200,000 maternal cells have been consistently found with this system.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.