Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Training Time”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
NASA NTRS2017

Evaluation of the Next-Gen Exercise Software Interface in the NEEMO Analog

NSBRI (National Space Biomedical Research Institute) funded research grant to develop the 'NextGen' exercise software for the NEEMO (NASA Extreme Environment Mission Operations) analog. Develop a software architecture to integrate instructional, motivational and socialization techniques into a common portal to enhance exercise countermeasures in remote environments. Increase user efficiency and satisfaction, and institute commonality across multiple exercise systems. Utilized GUI (Graphical User Interface) design principals focused on intuitive ease of use to minimize training time and realize early user efficiency. Project requirement to test the software in an analog environment. Top Level Project Aims: 1) Improve the usability of crew interface software to exercise CMS (Crew Management System) through common app-like interfaces. 2) Introduce virtual instructional motion training. 3) Use virtual environment to provide remote socialization with family and friends, improve exercise technique, adherence, motivation and ultimately performance outcomes.

Hanson, Andrea↗
NASA NTRS2018

Evaluation of Crew-Centric Onboard Mission Operations Planning and Execution Tool: Year 2

Currently, mission planning for the International Space Station (ISS) is largely affected by ground operators in mission control. The task of creating a week-long mission plan for ISS crew takes dozens of people multiple days to complete, and is often created far in advance of its execution. As such, re-planning or adapting to changing real-time constraints or emergent issues is similarly taxing. As we design for future mission operations concepts to other planets or areas with limited connectivity to Earth, more of these ground-based tasks will need to be handled autonomously by the crew onboard.There is a need for a highly usable (including low training time) tool that enables efficient self-scheduling and execution within a single package. The ISS Program has identified Playbook as a potential option. It already has high crew acceptance as a plan viewer from previous analogs and can now support a crew self-scheduling assessment on ISS or on another mission. The goals of this work, a collaboration between the Human Research Program and the ISS Program, are to inform the design of systems for more autonomous crew operations and provide a platform for research on crew autonomy for future deep space missions. Our second year of the research effort have included new insights on the crew self-scheduling sessions performed by the crew through use on the HERA (Human Exploration Research Analog) and NEEMO (NASA Extreme Environment Mission Operations) analogs. Use on the NEEMO analog involved two self-scheduling strategies where the crew planned and executed two days of EVAs (Extra-Vehicular Activities). On HERA year two represented the first HERA campaign where we were able to perform research tasks. This involved selected flexible activities that the crew could schedule, mock timelines where the crew completed more complex planning exercises, usability evaluation of the crew self-scheduling features, and more insights into the limit of plan complexity that the crew could effectively self-schedule. In parallel we have added in new features and functionality in the Playbook tool based off of our insights from crew self-scheduling in the NASA analogs. In particular this year we have added in the ability for the crew to add, edit, and remove their own activities in the Playbook tool, expanding the type of planning and re-planning possible in the tool and opening up the ability for more free form plan creation. The ability to group and manipulate groups of activities from the plan task list was also added, allowing crew members to add predefined sets of activities onto their mission timeline. In addition we also added a way for crew members to roll back changes in their plan, in order to allow an undo like capability. These features expand and complement the initial self-scheduling features added in year one with the goal of making crew autonomous planning more efficient. As part of this work we have also finished developing the first version of our Playbook Data Analysis Tool, a research tool built to interpret and analyze the unobtrusively collected data obtained during the NASA analog missions through Playbook. This data which includes user click interaction as well as plan change information, through the Playbook Data Analysis Tool, allows us to playback this information as if a video camera was mounted over the crewmember's tablet. While the primary purpose of this tool is to allow usability analysis of crew self-scheduling sessions used on the NASA analog, since the data collected is structured, the tool can automatically derive metrics that would be traditionally tedious to achieve without manual analysis of video playback. We will demonstrate and discuss the ability for future derived metrics to be added to the tool. In addition to the current data and results gathered in year two we will also discuss the preparation and goals of our International Space Station (ISS) onboard technology demonstration with Playbook. This technology demonstration will be preformed as part of the CAST payload starting in late 2016.

onboard planning↗
DOE OSTI2023

Fast and accurate reduced-order modeling of a MOOSE-based additive manufacturing model with operator learning

One predominant challenge in additive manufacturing (AM) is to achieve specific material properties by manipulating manufacturing process parameters during the runtime. Such manipulation tends to increase the computational load imposed on existing simulation tools employed in AM. The goal of the present work is to construct a fast and accurate reduced-order model (ROM) for an AM model developed within the Multiphysics Object-Oriented Simulation Environment (MOOSE) framework, ultimately reducing the time/cost of AM control and optimization processes. Our adoption of the operator learning (OL) approach enabled us to learn a family of differential equations produced by altering process variables in the laser’s Gaussian point heat source. More specifically, we used the Fourier neural operator (FNO) and deep operator network (DeepONet) to develop ROMs for time-dependent responses. Furthermore, we benchmarked the performance of these OL methods against a conventional deep neural network (DNN)-based ROM. Ultimately, we found that OL methods offer comparable performance and, in terms of accuracy and generalizability, even outperform DNN at predicting scalar model responses. The DNN-based ROM afforded the fastest training time. Furthermore, all the ROMs were faster than the original MOOSE model yet still provided accurate predictions. FNO had a smaller mean prediction error than DeepONet, with a larger variance for time-dependent responses. Unlike DNN, both FNO and DeepONet were able to simulate time series data without the need for dimensionality reduction techniques. Finally, the present work can help facilitate the AM optimization process by enabling faster execution of simulation tools while still preserving evaluation accuracy.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Multi-Objective Adaptive Mesh Refinement Using Reinforcement Learning

Finite element methods approximate the solution to a partial differential equation (PDE) on a mesh consisting of many elements. In general, using more, smaller elements results in a lower error in the approximation. However, it is often possible to lower the error substantially by only refining, or decreasing the size of, the few elements in the mesh that have the highest error. Adaptive mesh refinement (AMR) is a process that selectively refines regions of a mesh with high error to achieve a desired accuracy in as few degrees of freedom (DOFs) as possible. AMR is favorable compared to uniform refinement, which refines all elements of the mesh equally, because it can often achieve the same accuracy without wasting extra computation time on refinement of elements that already have low error. However, it is difficult to know which elements to refine. In this report we explore ways to choose which elements to refine such that we minimize both the resulting error and the cumulative DOFs used in computation. In particular, we introduce a Pareto-front learning algorithm that trains a policy to give the optimal refinement actions to minimize the cumulative DOFs used to achieve a given target error. Such a policy is useful because it can be deployed on many different problem types where different accuracy levels are desired. Furthermore, training a single policy for a range of target errors allows us to use transfer learning to reduce the required training time.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Machine Learning-Enabled Image Classification for Automated Electron Microscopy

Abstract Traditionally, materials discovery has been driven more by evidence and intuition than by systematic design. However, the advent of “big data” and an exponential increase in computational power have reshaped the landscape. Today, we use simulations, artificial intelligence (AI), and machine learning (ML) to predict materials characteristics, which dramatically accelerates the discovery of novel materials. For instance, combinatorial megalibraries, where millions of distinct nanoparticles are created on a single chip, have spurred the need for automated characterization tools. This paper presents an ML model specifically developed to perform real-time binary classification of grayscale high-angle annular dark-field images of nanoparticles sourced from these megalibraries. Given the high costs associated with downstream processing errors, a primary requirement for our model was to minimize false positives while maintaining efficacy on unseen images. We elaborate on the computational challenges and our solutions, including managing memory constraints, optimizing training time, and utilizing Neural Architecture Search tools. The final model outperformed our expectations, achieving over 95% precision and a weighted F-score of more than 90% on our test data set. This paper discusses the development, challenges, and successful outcomes of this significant advancement in the application of AI and ML to materials discovery.

Materials Science↗
DOE OSTI2025

Boosting Noise2Inverse via enhanced model selection for denoising computed tomography data

Synchrotron-based x-ray tomographic imaging enables the examination of the internal structure of materials at high spatial and temporal resolution. Experimental constraints can impose dose and time limits on the measurements, introducing a higher level of noise and artifacts in the reconstructed images. Deep learning has emerged as a powerful tool to remove noise from reconstructed images. Recently, the Noise2Inverse method was designed specifically for denoising reconstructed images without requiring paired noisy and clean images. This method creates multiple statistically independent reconstructions used to pair the data in which training involves transforming one reconstruction into the other, and vice versa. Originally designed to be used after a fixed number of epochs, we see in practice that this approach may not produce the optimal model and may unnecessarily waste computational resources. Therefore, we propose an alternative method of identifying the best model during training that aligns with the Noise2Inverse method. During validation, we compare the model output of the multiple reconstructions among each other. We hypothesize that the best model is the one that produces images with the highest similarity, implying a convergence in the predicted material properties and absorption values. To compare model outputs, we consider the absolute error, square error, structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and cosine similarity. We evaluate our method on two simulated tomography datasets and two, real-world, low-contrast, high-energy x-ray tomography datasets. We show our approach is more effective at determining the best model, up to an increase of 12.50% and 12.53% in SSIM and PSNR, respectively, while only requiring a fifth of the training time compared to the original approach.

CT↗
NASA NTRS2010

Multivariate Methods for Prediction of Geologic Sample Composition with Laser-Induced Breakdown Spectroscopy

Laser-induced breakdown spectroscopy (LIBS) uses pulses of laser light to ablate a material from the surface of a sample and produce an expanding plasma. The optical emission from the plasma produces a spectrum which can be used to classify target materials and estimate their composition. The ChemCam instrument on the Mars Science Laboratory (MSL) mission will use LIBS to rapidly analyze targets remotely, allowing more resource- and time-intensive in-situ analyses to be reserved for targets of particular interest. ChemCam will also be used to analyze samples that are not reachable by the rover's in-situ instruments. Due to these tactical and scientific roles, it is important that ChemCam-derived sample compositions are as accurate as possible. We have compared the results of partial least squares (PLS), multilayer perceptron (MLP) artificial neural networks (ANNs), and cascade correlation (CC) ANNs to determine which technique yields better estimates of quantitative element abundances in rock and mineral samples. The number of hidden nodes in the MLP ANNs was optimized using a genetic algorithm. The influence of two data preprocessing techniques were also investigated: genetic algorithm feature selection and averaging the spectra for each training sample prior to training the PLS and ANN algorithms. We used a ChemCam-like laboratory stand-off LIBS system to collect spectra of 30 pressed powder geostandards and a diverse suite of 196 geologic slab samples of known bulk composition. We tested the performance of PLS and ANNs on a subset of these samples, choosing to focus on silicate rocks and minerals with a loss on ignition of less than 2 percent. This resulted in a set of 22 pressed powder geostandards and 80 geologic samples. Four of the geostandards were used as a validation set and 18 were used as the training set for the algorithms. We found that PLS typically resulted in the lowest average absolute error in its predictions, but that the optimized MLP ANN and the CC ANN often gave results comparable to PLS. Averaging the spectra for each training sample and/or using feature selection to choose a small subset of wavelengths to use for predictions gave mixed results, with degraded performance in some cases and similar or slightly improved performance in other cases. However, training time was significantly reduced for both PLS and ANN methods by implementing feature selection, making this a potentially appealing method for initial, rapid-turn-around analyses necessary for Chemcam's tactical role on MSL. Choice of training samples has a strong influence on the accuracy of predictions. We are currently investigating the use of clustering algorithms (e.g. k-means, neural gas, etc.) to identify training sets that are spectrally similar to the unknown samples that are being predicted, and therefore result in improved predictions

Morris, Richard↗
DOE OSTI2025

Efficient and assured reinforcement learning-based building HVAC control with heterogeneous expert-guided training

Abstract Building heating, ventilation, and air conditioning (HVAC) systems account for nearly half of building energy consumption and $$20\%$$ of total energy consumption in the US. Their operation is also crucial for ensuring the physical and mental health of building occupants. Compared with traditional model-based HVAC control methods, the recent model-free deep reinforcement learning (DRL) based methods have shown good performance while do not require the development of detailed and costly physical models. However, these model-free DRL approaches often suffer from long training time to reach a good performance, which is a major obstacle for their practical deployment. In this work, we present a systematic approach to accelerate online reinforcement learning for HVAC control by taking full advantage of the knowledge from domain experts in various forms . Specifically, the algorithm stages include learning expert functions from existing abstract physical models and from historical data via offline reinforcement learning, integrating the expert functions with rule-based guidelines, conducting training guided by the integrated expert function and performing policy initialization from distilled expert function. Moreover, to ensure that the learned DRL-based HVAC controller can effectively keep room temperature within the comfortable range for occupants, we design a runtime shielding framework to reduce the temperature violation rate and incorporate the learned controller into it. Experimental results demonstrate up to 8.8 X speedup in DRL training from our approach over previous methods, with low temperature violation rate.

Xu, Shichao↗
DOE OSTI2023

Inverse reinforcement learning control for building energy management

Reinforcement learning (RL) based control is widely considered a promising approach in building automation and control as it has demonstrated the potential to deal with complex objectives in adjacent domains like robotics, autonomous vehicles, gaming applications, and advertisement recommendations. When applied to any environment, model-free RL learns to improve its control performance over time without requiring a control model, by receiving and then analyzing feedback from the building environment after each control action. Operational objectives are becoming increasingly complex through the simultaneous consideration of thermal comfort, carbon emissions, grid services, and indoor air quality. In this context, conventional rule-based control approaches are proving sub-optimal, mostly heuristic, and inadequate. The model-free and self learning nature of RL appears promising and attractive as it may address the scalability issues associated with advanced control approaches. However, it suffers from long training times and unstable control behavior during the early stages of its learning process, which makes it unsuitable to be applied directly to buildings. This paper addresses these issues using an inverse reinforcement learning approach (IRL), a technique utilized to learn the objective of a controller agent which is considered an expert in its respective domain. Here, we consider a rule-based control as the expert demonstrator. IRL is different from a direct imitation (i.e., direct mapping of states to actions) of control actions as it tries to find the underlying intent of an expert's policy, providing the controller with a better-generalized policy for unseen states or environments with slightly different dynamics. This approach propels the RL controller's policy to levels similar to or better than that of a rule-based policy before it starts learning by interacting with the building. This makes RL for building energy management applications more practical as it prevents the erratic and exploratory behavior in the initial training period, simultaneously speeding up the learning process when compared to applying an untrained RL agent directly to a building environment.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
NASA NTRS2011

Sparse Solutions for Single Class SVMs: A Bi-Criterion Approach

In this paper we propose an innovative learning algorithm - a variation of One-class nu Support Vector Machines (SVMs) learning algorithm to produce sparser solutions with much reduced computational complexities. The proposed technique returns an approximate solution, nearly as good as the solution set obtained by the classical approach, by minimizing the original risk function along with a regularization term. We introduce a bi-criterion optimization that helps guide the search towards the optimal set in much reduced time. The outcome of the proposed learning technique was compared with the benchmark one-class Support Vector machines algorithm which more often leads to solutions with redundant support vectors. Through out the analysis, the problem size for both optimization routines was kept consistent. We have tested the proposed algorithm on a variety of data sources under different conditions to demonstrate the effectiveness. In all cases the proposed algorithm closely preserves the accuracy of standard one-class nu SVMs while reducing both training time and test time by several factors.

Das, Santanu↗
DOE OSTI2026

MATEY: multiscale adaptive transformer models for spatiotemporal physical systems

Accurate representation of the multiscale features in spatiotemporal physical systems using vision transformer architectures requires extremely long, computationally prohibitive token sequences. To address this issue, we propose two novel adaptive tokenization schemes that dynamically adjust patch sizes based on local features: one ensures convergent behavior to uniform patch refinement, while the other offers better computational efficiency. Moreover, we present a set of spatiotemporal attention schemes, where the temporal or axial spatial dimensions are decoupled, to evaluate their baseline computational and data efficiencies and to determine whether adaptive tokenization can improve this performance. We assess the performance of the proposed multiscale adaptive model, MATEY, in a sequence of experiments. Compared to a full spatiotemporal attention scheme or a scheme that decouples only the temporal dimension, we find that fully decoupled axial attention is less efficient and expressive, requiring more training time and model parameters to achieve the same accuracy. The experiments on the adaptive tokenization schemes show that, compared to a uniformly refined model, the proposed schemes achieve comparable or improved accuracy at a much lower cost in the tested two-dimensional settings. While the asymptotic analysis suggests the potential for favorable scaling, empirical validation at substantially longer sequence lengths remains to be performed in future work. Finally, we demonstrate in two fine-tuning tasks featuring different physics that models pretrained on PDEBench data outperform the ones trained from scratch, especially in the low data regime with frozen attention.

adaptive tokenization↗
NASA NTRS2010

Atrial Fibrillation During an Exploration Class Mission

Background: A long-duration exploration class mission is fraught with numerous medical contingency plans. Herein, we explore the challenges of symptomatic atrial fibrillation (AF) occurring during an exploration class mission. The actions and resources required to ameliorate the situation, including the availability of appropriate pharmaceuticals, monitoring devices, treatment modalities, and communication protocols will be investigated. Challenges of Atrial Fibrillation during an Exploration Mission: Numerous etiologies are responsible for the initiation of AF. On Earth, we have the time and medical resources to evaluate and determine the causative situation for most cases of AF and initiate therapy accordingly. During a long-duration exploration class mission resources will be severely restricted. How is one to determine if new onset AF is due to recent myocardial infarction, pulmonary embolism, fluid overload, thyrotoxicosis, cardiac structural abnormalities, or CO poisoning? Which pharmaceutical therapy should be initiated and what potential side effects can be expected? Should anti-coagulation therapy be initiated? How would one monitor the therapeutic treatment of AF in microgravity? What training would medical officers require, and which communication strategies should be developed to enable the best, safest therapeutic options for treatment of AF during a long-duration exploration class mission? Summary: These questions will be investigated with expert opinion on disease elucidation, efficient pharmacology, therapeutic monitoring, telecommunication strategies, and mission cost parameters with emphasis on atrial fibrillation being just one illustration of the tremendous challenges that face a long-duration exploration mission. The limited crew training time, medical hardware, and drugs manifested to deal with such an event predicate that aggressive primary and secondary prevention strategies be developed to protect a multibillion-dollar asset like the International Space Station or a mission to the Moon or Mars. Learning Objectives: The audience will become familiar with the risks and challenges inherent to developing a therapeutic strategy for the treatment of atrial fibrillation during a long-term exploration class mission.

Lipset, Mark A.↗
DOE OSTI2024

Patch2Self2: Self-supervised Denoising on Coresets via Matrix Sketching

Diffusion MRI (dMRI) non-invasively maps brain white matter yet necessitates denoising due to low signal-to-noise ratios. Patch2Self (P2S) employing self-supervised techniques and regression on a Casorati matrix effectively denoises dMRI images and has become the new de-facto standard in this field. P2S however is resource intensive both in terms of running time and memory usage as it uses all voxels (n) from all-but-one held-in volumes (d-1) to learn a linear mapping Phi : \mathbb R ^ n x(d-1) \mapsto \mathbb R ^ n for denoising the held-out volume. The increasing size and dimensionality of higher resolution dMRI acquisitions can make P2S infeasible for large-scale analyses. This work exploits the redundancy imposed by P2S to alleviate its performance issues and inspect regions that influence the noise disproportionately. Specifically this study makes a three-fold contribution: (1) We present Patch2Self2 (P2S2) a method that uses matrix sketching to perform self-supervised denoising. By solving a sub-problem on a smaller sub-space so called coreset we show how P2S2 can yield a significant speedup in training time while using less memory. (2) We present a theoretical analysis of P2S2 focusing on determining the optimal sketch size through rank estimation a key step in achieving a balance between denoising accuracy and computational efficiency. (3) We show how the so-called statistical leverage scores can be used to interpret the denoising of dMRI data a process that was traditionally treated as a black-box. Experimental results on both simulated and real data affirm that P2S2 maintains denoising quality while significantly enhancing speed and memory efficiency achieved by training on a reduced data subset.

Fadnavis, Shreyas↗
DOE OSTI2023

"Safety First: Though on the Other Hand, Time is Critical”

This talk describes the deployment of research-quality software and hardware for open-road driving experiments by fleets of vehicles. Technological advancements in single-vehicle autonomy have expanded beyond the academic community, and are driven largely by industry stakeholders and manufacturers. However, as the penetration rate of cars with advanced driver assistance features increases, the impact on emergent behavior in traffic is still unknown. Compelling reasons to experiment within open-road conditions must be considered alongside the technical and human factor safety issues in deploying experimental controllers. This is even more challenging when it is necessary to deploy experimental fleets at scale. The talk will describe approaches for co-design of a research testbed for societal-scale systems. Discussion is devoted to the research-quality data gathering and control layers and their technical implementation, the different interfaces for experts in other fields to use these testbeds for research without violating safety requirements, and process and management considerations when deploying the platforms at scale when considering training time and operation complexity of human operators.

Sprinkle, Jonathan↗
DOE OSTI2023

Comparison of Machine Learning Algorithms for Natural Gas Identification with Mixed Potential Electrochemical Sensor Arrays

Mixed-potential electrochemical sensor arrays consisting of indium tin oxide (ITO), La 0.87 Sr 0.13 CrO 3 , Au, and Pt electrodes can detect the leaks from natural gas infrastructure. Algorithms are needed to correctly identify natural gas sources from background natural and anthropogenic sources such as wetlands or agriculture. We report for the first time a comparison of several machine learning methods for mixture identification in the context of natural gas emissions monitoring by mixed potential sensor arrays. Random Forest, Artificial Neural Network, and Nearest Neighbor methods successfully classified air mixtures containing only CH 4 , two types of natural gas simulants, and CH 4 +NH 3 with >98% identification accuracy. The model complexity of these methods were optimized and the degree of robustness against overfitting was determined. Finally, these methods are benchmarked on both desktop PC and single-board computer hardware to simulate their application in a portable internet-of-things sensor package. The combined results show that the random forest method is the preferred method for mixture identification with its high accuracy (>98%), robustness against overfitting with increasing model complexity, and had less than 10 ms training time and less than 0.1 ms inference time on single-board computer hardware.

03 NATURAL GAS↗
DOE OSTI2022

University Pipeline (with UNM) for Criticality Safety Professionals [Slides]

The scope of the University Pipeline is to provide students with knowledge of NCS and to inspire students to pursue a career in NCS after graduation. There are several benefits of the University Pipeline: reduced training time and costs; interested students that naturally self-sort and pursue the discipline at the university level; and a pipeline of criticality safety candidates readily available within the DOE Complex so that unexpected organizational or mission changes can be reacted to with increased agility.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2022

Pulsed Thermal Tomography Nondestructive Examination of Additively Manufactured Reactor Materials and Components (Final Technical Report)

Metal Additive Manufacturing (AM) is a promising method for cost-efficient fabrication of complex shape structures for applications in harsh environment, such as in a nuclear reactor. However, internal defects (pores) occur in high-strength AM alloys, which are manufactured with Laser Powder Bed Fusion (LPBF) AM method. Pulsed Infrared Thermography (PIT) is an efficient nondestructive evaluation (NDE) method to examine actual structures, because this method offers one-sided non-contact measurements, and fast processing of large sample areas. However, imaging of material defects, particularly defects with sizes at microscopic level, is challenging. In this report, we benchmark the performance of several Unsupervised Learning (UL) algorithms designed to enhance imaging of microscopic defects in metals with PIT. UL aims to learn the latent principal patterns (dictionaries) in PIT data to detect defects with minimal human supervision. Performance of Independent Component Analysis (ICA), Sparse Coding (SC), Principal Component Analysis (PCA) and Exploratory Factor Analysis (EFA) was compared using F-score, UL model training time and defects reconstruction time. We obtained the average F-score of 0.75, and a highest F-score of 0.89 for the EFA algorithm. Overall, EFA outperforms other UL algorithms considered in this study. In another approach, we investigate Thermal Tomography (TT), which is a computational method for reconstruction of depth profile of internal material defects from PIT nondestructive evaluation (NDE). TT algorithm obtains depth reconstructions of thermal effusivity, which has been shown to provide visualization of subsurface internals defects in metals. In many applications, one needs to determine the defect shape and orientation from reconstructed effusivity images. Interpretation of TT images is non-trivial because of blurring, which increases with depth due to heat diffusion-based nature of image formation. We have developed a deep learning convolutional neural network (CNN) to classify size and orientation of subsurface material defects in TT images. CNN was trained with TT images produced with computer simulations of 2D metallic structures (thin plates) containing elliptical subsurface voids. Performance of CNN was investigated using test TT images developed with computer simulations of plates containing elliptical defects, and defects with shape imported from scanning electron microscopy (SEM) images. CNN demonstrated the ability to classify radii and angular orientation of elliptical defects in previously unseen test TT images. We have also demonstrated that CNN trained on TT images of elliptical defects is capable of classifying shape and orientation of irregular defects. Training the CNN on irregular defect shapes instead of on elliptical shapes would make the resulting classifications more descriptive of actual defect shapes. However, this requires a much higher volume of SEM images of material defects, which are difficult to obtain because of random occurrence of defects in LPBF. To address this challenge, we developed a generative adversarial network (GAN) to augment the existing dataset of SEM defect images. The GAN model is demonstrated to create novel yet realistic defect shapes that can be used as input for simulated PTT images to train CNN. We also investigate several approaches based on Gaussian Random Circle and Bezier Curves for constructing parametric models of irregular-shape defects.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Tracing and Forecasting Metabolic Indices of Cancer Patients Using Patient-Specific Deep Learning Models

We develop a patient-specific dynamical system model from the time series data of the cancer patient’s metabolic panel taken during the period of cancer treatment and recovery. The model consists of a pair of stacked long short-term memory (LSTM) recurrent neural networks and a fully connected neural network in each unit. It is intended to be used by physicians to trace back and look forward at the patient’s metabolic indices, to identify potential adverse events, and to make short-term predictions. When the model is used in making short-term predictions, the relative error in every index is less than 10% in the L ∞ norm and less than 6.3% in the L 1 norm in the validation process. Once a master model is built, the patient-specific model can be calibrated through transfer learning. As an example, we obtain patient-specific models for four more cancer patients through transfer learning, which all exhibit reduced training time and a comparable level of accuracy. This study demonstrates that this modeling approach is reliable and can deliver clinically acceptable physiological models for tracking and forecasting patients’ metabolic indices.

60 APPLIED LIFE SCIENCES↗