Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “neural operator learning”

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.

916 records · Page 51

Lithium-Ion Battery Diagnostics Using Electrochemical Impedance via Machine-Learning

Diagnosing battery states such as health, state-of-charge, or temperature is crucial for ensuring the safety and reliability of electrochemical energy storage systems. While some states, such as temperature, may be measured using cheap sensors, accurate diagnosis of battery health metrics usually requires time-consuming performance measurements, making them infeasible for use in real-world operation. These health metrics can be measured during lab-testing and then estimated on-line using predictive life models or via state observer algorithms such as Kalman filters, but these predictive methods should be supplemented by actual measurement of battery health whenever possible to ensure reliability. Rapid measurement of battery health may be done by various types of fast diagnostic techniques such as electrochemical impedance spectroscopy (EIS), which can be performed in only a few minutes and require only a fraction of the energy and power needed for a full charge and discharge measurement. But there is a substantial challenge for estimating battery health using EIS data, as EIS is sensitive to cell temperature, state-of-charge, current, and resting time in addition to health. Thus, utilizing EIS data to predict battery capacity requires correcting for all these additional variables, a task that is extremely difficult to handle analytically. This talk utilizes machine-learning methods to estimate the effectiveness of battery capacity prediction from EIS data, leveraging a data set of hundreds of EIS measurements recorded at varying temperature and state-of-charge throughout a 500-day aging study of 32 commercial, large-format NMC-Graphite lithium-ion batteries. Using EIS as input to machine-learning models is complicated by the nonlinear response of impedance to battery health, temperature, and state-of-charge, as well as the collinearity between the impedance response at neighboring frequencies, which can easily lead to overfit models. To train robust models, features from EIS data need to be extracted from the data or some subset of critical frequencies selected. Many approaches for extracting and selecting features from EIS data from electrochemical analysis and machine-learning fields were identified for analysis: using the entire raw spectra; selection of one, two, or many frequencies from the entire spectra; selecting interesting points from the EIS measurement using domain knowledge; fitting EIS with an equivalent-circuit model; calculating statistics on the raw impedance values; and reducing the dimensionality of the data using unsupervised linear (principal component analysis) and non-linear (uniform manifold approximation and projection) methods. These approaches were rigorously compared using a machine-learning pipeline approach, training linear, Gaussian process, and random forest regression models and quantifying performance using cross-validation as well as a held-out test set. An artificial neural network model trained on the raw spectra was also tested. Promising pipelines were fine-tuned via Bayesian hyperparameter optimization using cross-validation loss and training with class-specific weights to counter data set imbalance. The most reliable method for utilizing impedance in this work was the selection of two optimal frequencies through an exhaustive search, resulting in about 2% mean absolute error on test data for both Gaussian process and random forest model architectures. Interrogation of a variety of models reveals critical frequencies of 100 Hz and 103 Hz for this data set, though the optimal set of frequencies is not necessarily intuitive, i.e., the best performing models are not simply those that use impedance at frequencies that have the highest correlation to the relative discharge capacity. The best performing model is an ensemble model, which is able to predict battery capacity with 1.9% mean absolute error for unseen cells using impedance recorded at a variety of temperatures and states-of-charge.

battery↗

Explainable tokamak-agnostic forecasting of fusion plasma instability via megahertz turbulent fluctuations

Scientific applications of artificial intelligence (AI) often remain limited by device-specific training and unexplained “black-box” approaches, creating fundamental barriers to cross-system generalization. This challenge is critical for nuclear fusion, where future reactors will have limited operational data for AI training. Here, we demonstrate that our neural network, trained solely on megahertz-scale turbulence measurements from one machine (DIII-D), forecasts Type-I edge localized mode (ELM) onsets in a different tokamak (KSTAR) through zero-shot weight transfer following physics-consistent preprocessing without device-specific retraining. Through an explainable AI framework combining gradient-weighted class activation mapping with physics validation, we reveal that our network can internalize physics relationships governing the ELM instabilities rather than memorizing device-specific patterns. The network perceives spatiotemporal features that correlate consistently with independently calculated instability growth rates, magnetohydrodynamic stability limits, and pedestal structure dynamics. Statistical analyses of dimensionally-reduced saliency features reveal the identical triangular features between the saliency representations, instability growth rates, and prediction probability across tokamaks, providing evidence that our forecasting system can show tokamak-agnostic generalization. This work contributes to a foundation for explainable scientific AI systems, where cross-system developments are essential for transcending traditional domain-specific constraints.

AI↗

Vehicle energy consumption estimation using large scale simulations and machine learning methods

Vehicle energy consumption and powertrain operations for future vehicle powertrain technologies are predicted through full-vehicle simulations using Autonomie, a simulation tool developed at Argonne National Laboratory. Over 1 million vehicles, with different powertrain types (conventional and numerous electrified powertrains) and component technologies (advanced engines, batteries, electrical machines, light weighting) are simulated for outputs including vehicle fuel economy and cost. Despite the ability of Autonomic to predict fuel economy given various combination of vehicle inputs, the full-vehicle simulation is very time-consuming. It takes approximately 84 h to simulate for 33,060 vehicles even using distributed computing on a cluster of 128 worker nodes. Moreover, Autonomic is unable to populate a continuous output space through simulation since it only yields outputs with discrete values instead of an input-output relation. This paper proposes a novel large-scale learning and prediction process (LSLPP) via machine learning approaches to answer the need for a continuous space of vehicle fuel economy and a more efficient simulation process. By learning in a supervised fashion to discover structures in data, the machine learning approaches seek to train a model that can explain the data with which we can perform prediction over unseen data. As such, we are able to accomplish analytic continuation on the current discrete space of simulation results. In addition, we present a random sampling -based numerosity reduction algorithm incorporated in LSLPP that reduces simulation runs and improves efficiency of data procurement. We analytically derive a theoretical bound in order to guarantee the effectiveness of our algorithm. We also compare random sampling to stratified sampling and suggest sampling strategies. A user-friendly tool is developed to support the process. Our experimental results confirm that the proposed LSLPP is able to greatly accelerate prediction and analysis of vehicle fuel economy. The output of this study is used by the US government to evaluate the impact of its R&D funding on energy security and CO 2 emission as well as for the setting of Corporate Average Fuel Economy (CAFE) standards.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Disruption event characterization and forecasting in tokamaks

Disruption prediction and avoidance is a critical need for next-step tokamaks, such as ITER. Disruption Event Characterization and Forecasting (DECAF) research fully automates analysis of tokamak data to determine chains of events that lead to disruptions and to forecast their evolution allowing sufficient time for mitigation or complete avoidance of the disruption. Disruption event chains related to local rotating or global magnetohydrodynamic (MHD) modes and vertical instability are examined with warnings issued for many off-normal physics events, including density limits, plasma dynamics, confinement transitions, and profile variations. Along with Greenwald density limit evaluation, a local radiative island power balance theory is evaluated and compared to the observation of island growth. Automated decomposition and analysis of rotating tearing modes produce physical event chains leading to disruptions. A total MHD state warning model comprised of 15 separate criteria produces a disruption forecast about 180 ms before a standard locked mode detector warning. Single DECAF event analyses have begun on KSTAR, MAST, and NSTX/-U databases with thousands of shot seconds of device operation using from 0.5 to 1 × 10 6 tested sample times per device. An initial multi-device database comparison illustrates a highly important result that plasma disruptivity does not need to increase as βN increases. Global MHD instabilities, such as resistive wall modes (RWMs), can give the briefest time period of warning before disruption compared to other physics events. In an NSTX database with unstable RWMs, the mode onset, loss of boundary and current control, and disruption event warnings are found in all cases and vertical displacement events are found in 91% of cases. An initial time-dependent reduced physics model of kinetic RWM stabilization created to forecast the disruption chain predicts instability 84% of the time for experimentally unstable cases with a relatively low false positive rate. Instances of the disruption event chain analysis illustrate dynamics including H–L back transitions for rotating MHD and global RWM triggering events. Furthermore, disruption warnings are issued with sufficient time before the disruption (on transport timescales) to potentially allow active profile control for disruption avoidance, active mode control, or mitigation.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Neural Posterior Estimation for Scalable and Accurate Inverse Parameter Inference in Li-Ion Batteries

Diagnosing the internal state of Li-ion batteries is critical for battery research, operation of real-world systems, and prognostic evaluation of remaining lifetime. By using physics-based models to perform probabilistic parameter estimation via Bayesian calibration, diagnostics can account for the uncertainty due to model fitness, data noise, and the observability of any given parameter. However, Bayesian calibration in Li-ion batteries using electrochemical data is computationally intensive even when using a fast surrogate in place of physics-based models, requiring many thousands of model evaluations. A fully amortized alternative is neural posterior estimation (NPE). NPE shifts the computational burden from the parameter estimation step to data generation and model training, reducing the parameter estimation time from minutes to milliseconds, enabling real-time applications. The present work shows that NPE can infer parameters equally or more accurately than Bayesian calibration, even if it leads to higher voltage reconstruction errors. We also demonstrate that the higher computational costs for data generation are tractable even in high-dimensional cases (ranging from 6 to 27 estimated parameters). The NPE method also offers several interpretability advantages over Bayesian calibration, such as local parameter sensitivity to specific regions of the voltage curve. The NPE method is demonstrated using an experimental fast charge dataset, with parameter estimates validated against measurements of loss of lithium inventory and loss of active material. The implementation is made available in a companion repository (https://github.com/NatLabRockies/BatFIT).

25 ENERGY STORAGE↗

Cy-Phy ADS: Cyber Physical Anomaly Detection Framework for EV Charging Systems

Today’s large-scale Electric Vehicle (EV) infrastructures are heavily dependent on information communication technologies to maintain their operation and to support communication within sub-system components as well as the outside world. These technologies are vulnerable to various cyber and physical threats. Timely identification and mitigation of these threats are critical for improving human safety, avoiding economic losses, and preventing catastrophic system failures. By addressing this, our work presents a ResNet Autoencoder (AE) based Cyber-Physical Anomaly Detection System (Cy-Phy ADS) for detecting anomalies in EV Controller Area Network (CAN) protocol communication. It consists of four main components: Cyber-Physical Feature Extractor, ResNet AE-based Anomaly Detection Framework, Cyber-Physical Health Metric (CPHM), and Visualization Dashboard. The presented framework was trained and tested using CAN data collected from the EV charging system testbed at the Idaho National Laboratory. The presented Cy-Phy ADS compared against six widely used unsupervised anomaly detection algorithms: One Class Support Vector Machine (OCSVM), Variational Autoencoder (VAE), LSTM Autoencoder (LSTM AE), Isolation Forest (IForest), Principle Component Analysis (PCA) and Local Outlier Factor (LOF). Here the presented approach showed the highest accuracy among the compared methods. Further, the proposed approach showed comparable performance in terms of precision, F1, and False positive rate. It also showed the lowest training and inference time compared to the neural network-based baseline algorithms compared against with. Additionally, the Cy-Phy ADS has advantages such as unsupervised training, the ability to provide a holistic metric for system health characterization, and non-linear feature extraction.

99 GENERAL AND MISCELLANEOUS↗

Advanced Analytics Studies Applied to US Department of Veterans Affairs' Corporate Data Warehouse (Initial Draft)

In this report, we describe our experience of applying several advanced analytics algorithms to the US Department of Veterans Affairs’ (VA’s) Corporate Data Warehouse (CDW) electronic health records datasets during FY2017-18. While various algorithms were applied to the CDW data, the main goal of this effort was to provide useful insights into the operational aspects (as opposed to the purely clinical aspect) of this specific implementation. Since there are not many reports in the public literature on advanced analytics applied to this data, this report provides unique insight in this regard. We focused on two machine learning (ML) applications: (1) clinical pathway inference (patterns dominating clinical pathways applied to a male dataset of Stable Ischemic Heart Disease patients as use case) and (2) medical-concept representation learning’s capability to inform and refine cohort membership based on probable patient outcomes. During our study on clinical pathway inference, we used two main modeling techniques: (a) topic probabilistic modeling (latent Dirichlet allocation [LDA]) and (b) feature imbedding. We observed that the applicability of LDA to the pathway inference remains in question. LDA provides results that are intuitive and easy for humans to comprehend; however, when applied to the pathway inference, it also generates pathway patterns that consist of a few dominant pathway components. In addition, LDA provides no information regarding the relationships among components in the same pathway. The outputs of LDA are also found to categorize patients based on their trace data of clinical procedures. However, the results suggest that LDA is biased toward statistically dominant components. This makes it particularly hard to discriminate pathway subbranches. To address the issue, we designed a new pathway inference methodology that integrates temporal ordering of pathway components into LDA results, as existing feature embedding tools such as word2vec are not readily applicable to our task. In addition, we developed our own feature embedding tool customized for patient traces. We conclude this study by suggesting two approaches to use embedding representation of a component into LDA and briefly list our lessons learned and recommendations for future work. For our study on medical-concept representation learning ability to inform and refine cohort membership based on probable patient outcomes, during our empirical evaluation we compared methods from medical-concept learning to standard one of a kind (one-of-K) encoding to evaluate the change in effectiveness as done in Choi, Schuetz, Stewart, and Sun (2016). We performed two primary empirical evaluations. The first evaluation was on a curated collection of 60,000 patients with no more than 1 year of medical history included. The second was a collection of patients with no restriction to the amount of medical history included. The hypothesis across both evaluations is that representation learning is broadly useful independent of downstream processing models; thus, for each experiment, we trained three models—a logistic regression model, a two-layer neural network model, and a nearest-neighbor model—and then averaged the results. Each model was trained using fivefold cross validation. We provided evidence that medical-representation learning improves predictions of the primary diagnosis category of a short patient history through 12 experiments. In addition, we provided evidence that medical-representation learning fails to improve prediction of the primary diagnosis category of an arbitrarily long patient history through 12 experiments. We conclude this study with a brief list of our lessons learned and recommendations for future work. We conclude the report with a discussion of recent technology trends as they relate to our artificial intelligence and, more specifically, to our ML research and approaches as described previously.

97 MATHEMATICS AND COMPUTING↗

Numerical analysis of soot emissions from gasoline-ethanol and gasoline-butanol blends under gasoline compression ignition conditions

In the present work, computational fluid dynamics (CFD) simulations of a single-cylinder gasoline compression ignition (GCI) engine were performed to investigate the impact of blending two biofuels, ethanol and n-butanol, with gasoline on the trade-off between combustion phasing and soot emissions under low load conditions. Here, in order to represent market gasoline (RD5-87), a four-component toluene primary reference fuel (TPRF) + ethanol (ETPRF) surrogate (with 20% ethanol by mole; E20) was formulated using a neural network based octane predictor such that the surrogate had the same ethanol content, Research Octane Number (RON) and Octane Sensitivity (S). In addition, a novel skeletal kinetic mechanism for ETPRF and TPRF + n-butanol (BTPRF) blends, incorporating polycyclic aromatic hydrocarbon (PAH) chemistry, was developed. A three-dimensional (3D) engine CFD formulation employing the skeletal mechanism, adaptive mesh refinement (AMR), finite-rate chemistry approach, and hybrid method of moments (HMOM) was adopted to capture the in-cylinder combustion phenomena and soot emissions. The engine CFD model was validated against RD5-87 experimental data for a broad range of start-of-injection (SOI) timings (-21/-27/-36/-45 crank angle degrees (CAD) after top-dead center (aTDC)), with respect to in-cylinder pressure, heat release rate, combustion phasing, and soot emissions. The closed-cycle simulation results were analyzed to elucidate the non-monotonic trend of soot emissions versus SOI timing: SOI-36 > SOI-45 > SOI-21 > SOI-27. Thereafter, the validated CFD model was employed to simulate the combustion of a gasoline-ethanol blend with 45% (by mole) ethanol (E45) and a gasoline-butanol blend with 45% (by mole) n-butanol (B45) under the same operating conditions to study the effects of fuel composition and SOI timing on combustion phasing and soot emissions. The sooting propensity followed the trend: B45 > E20 > E45 at all SOI timings. Overall, it was observed that the autoignition propensity was primarily related to fuel chemistry. On the other hand, sooting propensity showed strong coupling with both fuel chemistry and physical properties, with greater impact of fuel physical properties at advanced SOI timings.

30 DIRECT ENERGY CONVERSION↗

Evaluation of Advanced Signal Processing Techniques to Improve Detection and Identification of Embedded Defects

By the end of 1996, 109 Nuclear Power Plants were operating in the United States, producing 22% of the Nation’s electricity [1]. At present, more than two thirds of these power plants are more than 40 years old. The purpose of the U.S. Department of Energy Office of Nuclear Energy’s Light Water Reactor Sustainability (LWRS) Program is to develop technologies and other solutions that can improve the reliability, sustain the safety, and extend the operating lifetimes of nuclear power plants (NPPs) beyond 60 years [2]. The most important safety structures in an NPP are constructed of concrete. The structures generally do not allow for destructive evaluation and access is limited to one side of the concrete element. Therefore, there is a need for techniques and technologies that can assess the internal health of complex, reinforced concrete structures nondestructively. Previously, we documented the challenges associated with Non-Destructive Evaluation (NDE) of thick, reinforced concrete sections and prioritized conceptual designs of specimens that could be fabricated to represent NPP concrete structures [3]. Consequently, a 7 feet tall, by 7 feet wide, by 3 feet and 4-inch-thick concrete specimen was constructed with 2.257-inch-and 1-inch-diameter rebar every 6 to 12 inches. In addition, defects were embedded the specimen to assess the performance of existing and future NDE techniques. The defects were designed to give a mix of realistic and controlled defects for assessment of the necessary measures needed to overcome the challenges with more heavily reinforced concrete structures. Information on the embedded defects is documented in [4]. We also documented the superiority of Frequency Banded Decomposition (FBD) Synthetic Aperture Focusing Technique (SAFT) over conventional SAFT when probing defects under deep concrete cover. Improvements include seeing an intensity corresponding to a defect that is either not visible at all in regular, full frequency content SAFT, or an improvement in contrast over conventional SAFT reconstructed images. This report documents our efforts in four fronts: 1) Comparative study between traditional SAFT and FBD SAFT for concrete specimen with and without Alkali-Silica Reaction (ASR) damage, 2) improvement of our Model-Based Iterative Reconstruction (MBIR) for thick reinforced concrete [5], 3) development of a universal framework for sharing, reconstruction, and visualization of ultrasound NDE datasets, and 4) application of machine learning techniques for automated detection of ASR inside concrete. Our comparative study between FBD and traditional SAFT reconstruction images shows a clear difference between images of ASR and non-ASR specimens. In particular, the left first harmonic shows an increased contrast and sensitivity to ASR damage. For MBIR, we show the superiority of model-based techniques over delay and sum techniques such as SAFT. Improvements include elimination of artifacts caused by direct arrival signals, and increased contrast and Signal to Noise Ratio. For the universal framework, we document a format for data storage based on the HDF5 file format, and also propose a modular Graphic User Interface (GUI) for easy customization of data conversion, reconstruction, and visualization routines. Finally, two techniques for ASR automated detection are presented. The first technique is based on an analysis of the frequency content using Hilbert Transform Indicator (HTI) and the second technique employees Artificial Neural Network (ANN) techniques for training and classification of ultrasound data as ASR or non-ASR damaged classes. The ANN technique shows great potential with classification accuracy above 95%. These approaches are extensible to the detection of additional reinforced, thick concrete defects and damage.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

A data analytical approach for assessing the efficacy of Operational Technology active defenses against insider threats

In recent years, the need for Operational Technology (OT) defenses has been recognized, serving as an additional line of defense when Information Technology (IT) defenses are bypassed. This is no longer considered an uncommon possibility when dealing with advanced persistent threat (APT) actors expected to be state-sponsored and receiving insider assistance. In these extreme adversarial situations, OT defenses aim to provide another layer of defense for the system, introduced directly at the physical process level, as described by the sensors data, the system model, and control actions. Just like IT defenses, two schools of thought, i.e., passive and active defenses, have emerged to address this challenge. In active defenses, representing the focus of this paper, known signatures, synthesized based on the system's unique characteristics, are inserted into the system. In contradistinction, passive methods rely solely on observing system behavior in search of patterns of normal behavior with deviations thereof representing abnormal behavior. In their most sophisticated implementations, both passive and active defenses rely on the use of data analytics to identify the patterns and synthesize the observed and/or inserted signatures. Past research has shown that passive defenses may be bypassed by APT actors relying on data analytics and their intimate knowledge of the system to evade detection by respecting the patterns identified by the defenders. Here this manuscript explores the use of active defenses under the assumption that the attacker has privileged access to the system, including access to the system's model and sensors data. Specifically, this manuscript assesses the ability of active defenses to remain invisible to the attackers, and discusses the associated challenges that must be addressed to ensure their resiliency against APT actors.

98 NUCLEAR DISARMAMENT, SAFEGUARDS, AND PHYSICAL P↗

Toward digital design at the exascale: An overview of project ICECap

High performance computing has entered the Exascale Age. Capable of performing over 1018 floating point operations per second, exascale computers, such as El Capitan, the National Nuclear Security Administration's first, have the potential to revolutionize the detailed in-depth study of highly complex science and engineering systems. However, in addition to these kind of whole machine “hero” simulations, exascale systems could also enable new paradigms in digital design by making petascale hero runs routine. Currently, untenable problems in complex system design, optimization, model exploration, and scientific discovery could all become possible. Motivated by the challenge of uncovering the next generation of robust high-yield inertial confinement fusion (ICF) designs, project ICECap (Inertial Confinement on El Capitan) attempts to integrate multiple advances in machine learning (ML), scientific workflows, high performance computing, GPU-acceleration, and numerical optimization to prototype such a future. Built on a general framework, ICECap is exploring how these technologies could broadly accelerate scientific discovery on El Capitan. In addition to our requirements, system-level design, and challenges, we describe some of the key technologies in ICECap, including ML replacements for multiphysics packages, tools for human-machine teaming, and algorithms for multifidelity design optimization under uncertainty. As a test of our prototype pre-El Capitan system, we advance the state-of-the art for ICF hohlraum design by demonstrating the optimization of a 17-parameter National Ignition Facility experiment and show that our ML-assisted workflow makes design choices that are consistent with physics intuition, but in an automated, efficient, and mathematically rigorous fashion.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

MPEX AI Digital Twins

All magnetically confined plasma fusion power plant concepts (Tokamak, Spherical Tokamak, Stellarator, Mirror, ...) must exhaust the heat and plasma from the core confinement region to the material walls. The primary channel for this exhaust is through a plasma divertor which directs plasma along open magnetic field lines to a material target. The Material Plasma Exposure eXperiment (MPEX) illustrated in Figure 1, is a high-power, steady-state linear plasma device designed to produce the plasma material interaction (PMI) conditions of the divertor of future magnetic confinement fusion power plants: energy flux 20MW/m 2 , ion fluence 1031/m 2 , pulse duration 106 sec. These goals of plasma exposure in MPEX are well beyond those achieved in magnetic fusion experimental devices. Successfully achieving these high power steady state conditions for long pulses requires operational control of the heating and particle sources and the plasma flux to the walls and target. The MPEX AI Hot Spot Controller, proposed in this project, will help achieve the operational milestones of MPEX. The MPEX device will begin commissioning at the end of FY26. A smaller proto-MPEX was operated for 14,666 plasma discharges and will resume operation in September of 2025 as proto-MPEX-lite, with reduced capability, to test a new window for the Helicon plasma source. The proto-MPEX data has undergone surrogate modeling with machine learning methods (R. Archibald, 2022 IEEE International Conference on Big Data). This proto-MPEX data will be used to begin development of the AI digital twins described in this white paper. The scientific mission of MPEX is to qualify materials of different composition for use in the high energy and plasma flux conditions of a fusion power plant. The materials exposed in MPEX will in some cases be exposed to high neutron fluxes at other ORNL facilities to measure the changes to their PMI properties. The targets exposed in MPEX will be transported under vacuum to a Surface Analysis Station (SAS). The SAS will be equipped with the following diagnostics: Focused Ion Beam (FIB) for trench milling, 100-400 angstrom resolution scanning electron microscope (SEM), surface mapping x-ray spectrometer, high resolution camera, and a future upgrade to a laser induced breakdown spectroscopy quadruple mass spectrometer (LIBS-QMS). The MPEX experiments will generate diverse pre- and post-exposure measurement data of detailed material properties down to the crystal grain level in 3D for post-exposure assessment of PMI damage (e.g. cracking, melting, erosion and redeposition of the material). Physics models for the PMI, and how the material composition and manufacturing impact its performance under high energy plasma exposure, need to be validated with MPEX data to guide the selection of new candidate materials. Our vision for the MPEX AI Digital Twins project is to supply experimental and physics model simulation data to train Artificial Intelligence (AI) models for data processing, analysis, operational control, PMI and materials simulation to maximize the scientific output of the MPEX device. Ultimately, an AI digital twin of MPEX material assessment metrics for tested and synthetic material types with simulated PMI will be trained by the AI Modeling Teams on the experimental and physics simulation data submitted to the American Science Cloud by this project. A purely empirical search for the best material is inefficient given the finite number of samples that can be tested on MPEX. In order to expand the material properties database for training the MPEX Material Assessment AI Digital Twin, and to gain physics understanding of the PMI processes, physics models of the material properties and PMI processes are required. The physics simulations provide detailed simulation data, like impact angles for plasma ions, sputtering yields, transport of the ionized sputtered target material in the plasma, and redeposition locations. This simulation data expands the measurement data for deeper physics understanding. The experimental data is essential to validate the PMI and material structure simulation models. The validated models can then be used to generate new simulation data of MPEX material assessments for synthetic material compositions that have not been exposed in MPEX. These predictive simulations, plus the whole experimental dataset, will be used to train the MPEX Material Assessment AI Digital Twin allowing a rapid generative AI search for new materials with reduced PMI damage by interpolating the domain of the training set. These new optimum materials can be simulated with the physics codes and/or tested in MPEX. The ability of AI neural networks to interpolate multi-dimensional parameter spaces and generate virtual data is exploited for a more efficient search for optimum materials. The advent of the Transformational AI Models Consortium (TAIMC) is an opportunity to engage with state of the art private and public AI developers to achieve the goals of the AI digital twins and AI accelerated physics models proposed in this project. Our partners at ORNL from the Advance Scientific Computing Research (ASCR) organization will collaborate in accelerating the integrated plasma material interaction simulation framework. This simulation framework will provide a platform for generating simulation data across a range of physical fidelities, including hybrid methods that produce multi-fidelity results. This data will be leveraged for AI model development, both for generation of surrogates and the automation of simulation campaigns. A part of the research below will include collaborative efforts with the TAIMC to (i) adapt data storage approaches to ensure AI-readiness, (ii) provide a protypical exemplar to inform and exercise constructed workflows, and (iii) generate and share data, using the TAIMC unified AI data standard, for foundational models that will be trained from multiple sources across the DOE complex. We will also collaborate with the TAIMC, as well as the planned AI modeling teams, to develop approaches for reducing the cost of data generation. These include tailored multi-fidelity approaches as well as fine-tuning strategies to augment general, large-scale foundational models.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

FPGA-accelerated SpeckleNN with SNL for real-time X-ray single-particle imaging

We present the implementation of a specialized version of our previously published unified embedding model, SpeckleNN, for real-time speckle pattern classification in X-ray Single-Particle Imaging (SPI), using the SLAC Neural Network Library (SNL) on an FPGA platform. This hardware realization transitions SpeckleNN from a prototypic model into a practical edge solution, optimized for running inference near the detector in high-throughput X-ray free-electron laser (XFEL) facilities, such as those found at the Linac Coherent Light Source (LCLS). To address the resource constraints inherent in FPGAs, we developed a more specialized version of SpeckleNN. The original model, which was designed for broader classification across multiple biological samples, comprised ~5.6 million parameters. The new implementation, while reducing the parameter count to 64.6K (a 98.8% reduction), focuses on maintaining the model's essential functionality for real-time operation, achieving an accuracy of 90%. Furthermore, we compressed the latent space from 128 to 50 dimensions. This implementation was demonstrated on the KCU1500 FPGA board, utilizing 71% of available DSPs, 75% of LUTs, and 48% of FFs, with an average power consumption of 9.4W according to the Vivado post-implementation report. The FPGA performed inference on a single image with a latency of 45.015 microseconds at a 200 MHz clock rate. In comparison, running the same inference on an NVIDIA A100 GPU resulted in an average power consumption of ~73W and an image processing latency of around 400 microseconds. Our FPGA-accelerated version of SpeckleNN demonstrated significant improvements, achieving an 8.9 × speedup and a 7.8 × reduction in power consumption compared to the GPU implementation. Key advancements include model specialization and dynamic weight loading through SNL, which eliminates the need for time-consuming FPGA design re-synthesis, allowing fast and continuous deployment of models (re)trained online. These innovations enable real-time adaptive classification and efficient vetoing of speckle patterns, making SpeckleNN more suited for deployment in XFEL facilities. This implementation has the potential to significantly accelerate SPI experiments and enhance adaptability to evolving experimental conditions.

47 OTHER INSTRUMENTATION↗

Using dorsal surface for individual identification of dairy calves through 3D deep learning algorithms

Advances in machine learning techniques have allowed the development of computer vision systems (CVS) that can accurately predict several phenotypes of interest for livestock operations. In this context, 3D images taken from a top-down view are particularly useful for estimating body condition score, growth development, and body biometrics in cattle. Frequently, such CVS rely on identification (ID) systems, such as electronic tags, as a way to match animal ID and the predicted phenotype. However, the same 3D images used to predict body weight and other animal biometrics could be adopted for animal recognition as well. Such alternative would optimize CVS to recognize animal ID and monitor growth development simultaneously while leveraging the same hardware infrastructure. Furthermore, this strategy could be used to recognize animals with similar color patterns. Nonetheless, growing animals are continuously changing body shape, which could limit its use as an invariant feature for pattern recognition. Thus, the objectives of this study were: (1) to compare algorithms for different 3D object representations to identify individual animals; and (2) to evaluate how short-term changes in body shape due to animal growth affect the predictive performance of these algorithms. For objective 1, the algorithms were trained (n = 4,558) and tested (n = 1,139) using images from 38 Holstein calves. For objective 2, we designed three different experiments using images (n = 2,347) from five Holstein calves taken over six weeks during their growing period, always training and testing on different weeks. Each experiment evaluated how changing a different parameter of the image capturing procedure affected the predictive ability of the trained algorithms. In the first experiment, we varied the total number of images per animal in the training set; in the second experiment, we varied the number of weeks while keeping a fixed number of images in the training set; and in the third experiment, we skipped weeks between images in the training and test sets. The F 1 score for objective (1) was up to 0.804 when testing with the last frames of each video, and up to 0.959 when using random frames for testing. For objective (2), the F 1 score was up to 0.947 for the first experiment when using 130 images per animal; up to 0.979 for the second experiment when using all five weeks; and up to 0.917 when not skipping weeks between training and testing. In conclusion, these results show that deep learning algorithms can be used to identify individual animals through their dorsal area 3D surfaces, and, from our experiments using calves in their growing period, that they are robust enough to account for changes in body shape and size, making them a promising tool for animal recognition during growth.

3D neural networks↗

Applying Improved Optical Recognition with Machine Learning on Sorting Cu Impurities in Steel Scrap

Annually, 20–55 million tons of electronic waste (e-waste) is produced worldwide (5% of all municipal solid waste). Although e-waste embodies only 2% of America’s municipal waste, it accounts for a significantly larger proportion of the heavy metals and flame retardants present in the waste stream. Currently, less than 20% of all e-waste is recycled in the United States because the heterogeneity of the feedstock limits the opportunity for reuse in high value products and processing of the waste itself is often too costly to justify handling. To address this concern, the aims of this study are: 1) to identify the major plastic and metal compositions within electronic shredder residue (ESR), 2) to formulate solvents and processing conditions to separate 90% of the plastics targeted from consumer shred ESR, and 3) to develop a process design model to estimate the cost and energy efficiency of the proposed solvent-based processing. In this study, a pre-sorted heterogeneous ESR feedstock (one where aluminum, magnetic components, and hazardous battery materials removed by an e-waste recycling facility) was used, with the major compositions of the ESR characterized. It was found that 25 wt.% of the feedstock was composed of plastics, 6 wt.% rubber, 27 wt.% printed circuit boards, 23% wire, and the remainder metals and capacitors. Within the plastic portion, polystyrene (PS, 40 wt.%), acrylonitrile butadiene styrene (ABS, 25 wt.%), and styrene-acrylonitrile (SAN, 9wt.%) were identified to compose the majority of the screened plastics using Fourier transform infrared spectroscopy (FTIR). Next, selective solvents were screened using Hansen Solubility Parameter Theory (HSP) for dissolving PS and ABS. The pre-screening results show that methylene chloride (dichloromethane, DCM) and tetrahydrofuran (THF) are capable of dissolving the most PS and ABS, while methanol (MeOH) and ethylene glycol (EG) are capable of precipitating the most PS and ABS. These solvents were subsequently used to recover polymers and remove. flame retardants within the ESR feedstock. By optimizing the dissolution time and the solvents used, the highest polymer dissolution yield (99 wt.%%) was achieved using DCM for 48 hr. Both pre-screened anti-solvents (MeOH and EG) showed the highest polymer precipitation yield (71 wt.%). In terms of flame retardant removal rate, EG was found to have a high phosphorus-containing flame retardant removal rate (up to 98%). Characterization shows that the proposed solvent-based processing can preserve a high molecular weight fraction of the polymers and effectively remove flame retardants. Cost analysis indicates that the amount of the solvent/anti-solvent recovered after the reaction would play a critical role in reducing the operating costs. The energy analysis shows that the proposed solvent-based processes can save up to 60% of the embodied energy used to manufacture plastics used in electronics (PS and ABS were used for calculations). The results from this project prove the potential of solvent-based processing to produce secondary materials (plastics and metals) from e-waste for cross-industry reuse.

36 MATERIALS SCIENCE↗

fPINNs: Fractional Physics-Informed Neural Networks

Physics-informed neural networks (PINNs), introduced, are effective in solving integer-order partial differential equations (PDEs) based on scattered and noisy data. PINNs employ standard feedforward neural networks (NNs) with the PDEs explicitly encoded into the NN using automatic differentiation, while the sum of the mean-squared PDE residuals and the mean-squared error in initial-boundary conditions is minimized with respect to the NN parameters. Here we extend PINNs to fractional PINNs (fPINNs) to solve space-time fractional advection-diffusion equations (fractional ADEs), and we study systematically their convergence, hence explaining both fPINNs and PINNs for the first time. Specifically, we demonstrate their accuracy and effectiveness in solving multidimensional forward and inverse problems with forcing terms whose values are only known at randomly scattered spatio-temporal coordinates (black-box (BB) forcing terms). A novel element of the fPINNs is the hybrid approach that we introduce for constructing the residual in the loss function using both automatic differentiation for the integer-order operators and numerical discretization for the fractional operators. This approach bypasses the difficulties stemming from the fact that automatic differentiation is not applicable to fractional operators because the standard chain rule in integer calculus is not valid in fractional calculus. To discretize the fractional operators, we employ the Grünwald--Letnikov (GL) formula in one-dimensional fractional ADEs and the vector GL formula in conjunction with the directional fractional Laplacian in two- and three-dimensional fractional ADEs. We first consider the one-dimensional fractional Poisson equation and compare the convergence of the fPINNs against the finite difference method (FDM). We present the solution convergence using both the mean L 2 error as well as the standard deviation due to sensitivity to NN parameter initializations. Using different GL formulas we observe first-, second-, and third-order convergence rates for small size training sets but the error saturates for larger training sets. We explain these results by analyzing the four sources of numerical errors due to discretization, sampling, NN approximation, and optimization. The total error decays monotonically (below 10 –5 for a third-order GL formula) but it saturates beyond that point due to the optimization error. We also analyze the relative balance between discretization and sampling errors and observe that the sampling size and the number of discretization points (auxiliary points) should be comparable to achieve the highest accuracy. As we increase the depth of the NN up to certain value, the mean error decreases and the standard deviation increases, whereas the width has essentially no effect unless its value is either too small or too large. We next consider time-dependent fractional ADEs and compare white-box (WB) and BB forcing. We observe that for the WB forcing, our results are similar to the aforementioned cases; however, for the BB forcing fPINNs outperform FDM. Subsequently, we consider multidimensional time-, space-, and space-time-fractional ADEs using the directional fractional Laplacian and we observe relative errors of 10 –3 ~ 10 –4 . Lastly, we solve several inverse problems in one, two, and three dimensions to identify the fractional orders, diffusion coefficients, and transport velocities and obtain accurate results given proper initializations even in the presence of significant noise.

97 MATHEMATICS AND COMPUTING↗