Engineering topics
Publications and source records attributed to Dhulipala, Som LakshmiNarasimha.
Novel uncertainty quantification approach for machine learning based physics informed graph neural networks capable of quickly computing bulk energy system contingencies from a large search space with tolerance to changing grid topologies.
Poster for the LDRD presentation event.
MOOSE is an open-source computational platform for constructing multi-physics models and executing them in a massively parallel fashion. It has a stochastic tools module (STM) for forward/inverse uncertainty quantification (UQ) and surrogate modeling. This presentation details some recent developments to the STM with respect to the implementation of adaptive, active learning, and multifidelity Monte Carlo methods for forward UQ of computational models. Specifically, the adaptive Monte Carlo methods include Markov Chain Monte Carlo (MCMC)-driven algorithms like adaptive importance sampling and parallelized subset simulation for statistical QoI estimation, rare events analysis, and stochastic gradient-free optimization. The active learning methods include Gaussian Process (GP) surrogates and their training via Adam optimization, design of acquisition functions, and integration with samplers like Monte Carlo, adaptive importance, and parallelized subset simulation. These active learning methods are also designed to work in a batch mode, wherein, the required calls to the full computational model are executed in parallel whenever a user-specified batch size is met. The multifidelity methods in STM are broadly divided into two categories: hierarchical, where a defined hierarchy exists among the low-fidelity models, and peer, where all the low-fidelity models are treated equally. A GP surrogate is used to learn the differences between the low- and high-fidelity models in both multifidelity categories, and acquisition functions from the active learning classes are used to decide whether to rely on a low-fidelity model or call the expensive high-fidelity model. Alongside the software description and usage, applications are also presented to nuclear engineering computational models including a TRISO nuclear fuel particle, a reactor pressure vessel, and a heat-pipe microreactor.
When sampling for Bayesian inference, one popular approach is to use Hamiltonian Monte Carlo (HMC) and the No-U-Turn Sampler (NUTS). However, HMC and NUTS can require numerous numerical gradients of the target density and can prove slow in practice. We propose Hamiltonian neural networks (HNNs) with HMC and NUTS for solving Bayesian inference problems [1, 2]. Once trained, HNNs do not require gradients of the target density while sampling. Moreover, they satisfy important properties such as perfect time reversibility and Hamiltonian conservation, making them well suited for use within HMC and NUTS because stationarity can be shown. We also propose an HNN extension called latent HNNs (L-HNNs), which predict latent variable outputs. Compared to HNNs, L-HNNs offer improved expressivity and a reduction in integration errors. Finally, we propose employing L-HNNs in NUTS with an online error monitoring scheme to prevent degeneracy of the sampling in regions of low probability density. We demonstrate L-HNNs in NUTS with online error monitoring by using several example cases involving complex, heavy-tailed, and high local curvature probability densities. Overall, L-HNNs in NUTS with online error monitoring satisfactorily inferred these probability densities. Compared to traditional NUTS, L-HNNs in NUTS with online error monitoring improved the effective sample size (ESS) per gradient by an order of magnitude.
The current practice for natural phenomena hazards (NPH) risk assessment of nuclear facilities is to compute the risk for each hazard independently and then compound the total risk as a combination of single hazard risks. This state of practice does not consider correlations between hazards and the cascading impacts to structures, systems, and components (SSCs), and could thus underestimate the NPH risk or overestimate the nuclear facility safety. Events such as the Fukushima Daiichi accident have highlighted the importance of multi-hazard risk considerations to nuclear power plants (NPPs) that quantify the cascading damage effects to SSCs in the risk models. Moreover, the current fleet of NPPs in the United States is aging; these NPPs are now expected to operate well beyond their initially planned design life. Aging-related deterioration can potentially decrease the capacity of critical structures such as containment vessels to withstand NPH. Such aging considerations may not be adequately accounted for by the current NPH risk assessment guidelines. This paper presents a preliminary modeling and simulation of a representative reinforced concrete containment vessel subjected to seismic mainshock and aftershock considering concrete degradation due to alkali silica reaction. The broader aim is to develop multi-hazard time-dependent fragility functions that could be subsequently used in the probabilistic risk assessment (PRA) model. The multi-hazard component comes into play due to the consideration of damage to the containment vessel under seismic loads and concrete degradation. Consideration of concrete degradation also brings into play the time-dependent nature of the containment vessel response. The response of the containment vessel under varying degrees of concrete degradation to seismic loads is investigated. The results presented are simulated using the Multi-hazard Analysis for STOchastic time-DOmaiN phenomena (MASTODON) software for seismic analysis and the Blackbear software for concrete degradation and damage modeling. Both software are open source and developed within the Multiphysics Object-Oriented Simulation Environment (MOOSE).
The Monte Carlo method delivers an unbiased estimate of the probability of failure. However, the variance of the estimate depends on the number of evaluated samples. This number must be very large for estimations of a low probability of failure. If the evaluation of each sample is computationally expensive, the crude Monte Carlo simulation strategy is impracticable. Therefore, subset simulations are used to reduce the required number of evaluations. Subset simulations require a Markov Chain Monte Carlo sampler, such as the random walk Metropolis-Hastings algorithm. The algorithm, however, struggles with sampling in low-probability regions, especially if they are narrow. As a consequence, advanced Markov Chain Monte Carlo simulations have been developed. In particular, the Hamiltonian Monte Carlo method explores the target distribution rapidly. Driven by the idea of Hamiltonian dynamics, this sampler provides a non-random walk through the target distribution. The incorporation of subset simulation and Hamiltonian Monte Carlo methods has shown promising results for reliability analysis. One downside of the Hamiltonian Monte Carlo method is that gradient evaluations are computationally expensive, especially when dealing with high-dimensional problems and evaluating long trajectories. We show that integrating Hamiltonian neural networks in Hamiltonian Monte Carlo simulations significantly speeds up the sampling task. Furthermore, the enhancement of adaptive trajectory length within the Hamiltonian Monte Carlo results in the efficient proposal of the following states. Based on this recent enhancement, we provide a fast sampling strategy for subset simulations using Hamiltonian neural networks to replace the evaluation of the gradient and significantly speed up the Hamiltonian Monte Carlo simulation.
To assess the reliability of critical technologies like nuclear plants and infrastructure systems and improve the robustness of design, engineers have to quantify the uncertainties surrounding the system behavior accurately. However, the complexity of the problem can make standard reliability analysis algorithms prohibitively expensive, primarily due to the high computational cost of estimating the system response at each iteration. This cost can be greatly reduced by using multi-fidelity modeling and machine learning to build a surrogate model to replace the expensive response function. We propose a general and robust method for building surrogates from multiple Low Fidelity (LF) models coupled with machine learning to retain accuracy. Our framework first constructs “Corrected Low Fidelity models” (CLFs) by coupling a High Fidelity (HF) model inferred Gaussian Process correction term with each of the LF models. It then uses the correction terms to assign model probabilities to each of these CLFs in an explainable way before using them to assemble the final surrogate. No assumptions are made about the type of the LF models or their correlation with the HF model. The proposed surrogate modeling framework is used within the subset simulation algorithm (a variance-reduced MCMC-based reliability analysis algorithm) for enhanced efficiency. Additionally, an active learning step is added to the algorithm to adaptively decide when the surrogate is not sufficiently accurate, at which point the HF model is called and used to refine the surrogate. Through a frame buckling example, our method is shown to be highly efficient at reducing the expensive HF model calls while accurately estimating the failure probability.
The U.S. Department of Energy (DOE)’s Nuclear Energy Advanced Modeling and Simulation (NEAMS) program aims to develop predictive capabilities by applying computational methods to the analysis and design of advanced reactor and fuel cycle systems. This program has been providing engineering-scale support for the development of BISON, a high-fidelity and high-resolution fuel performance tool. Fuel behavior in a nuclear reactor is governed by a complex network of mechanisms interacting with various other physics aspects in the reactor system. Any model developed to represent the fuel behavior will likely be idealized resulting in uncertainties in their predictions compared to the observed data. As such, this report was motivated by the need to identify the sources of uncertainties and quantify and propagate them through the fuel model outputs. Such quantification of uncertainties will establish a level of model trustworthiness, identify approaches to improve the model trustworthiness, and even guide optimal experiment design for maximal information gain. To accomplish the uncertainty quantification for computational models, this report has relied on the Bayesian framework which provides probabilistic treatment of models their inputs and outputs. The current state-of-the-art on performing Bayesian Uncertainty Quantification (UQ) for nuclear engineering models using High Performance Computing (HPC) resources have been reviewed. Implementation of capabilities for massively parallel Bayesian UQ in Multiphysics Object-Oriented Simulation Environment (MOOSE) is discussed. Several verification cases are discussed to verify the accuracy of the quantified uncertainties using the developed computational capabilities in MOOSE. Then, the problem of quantifying the uncertainties in TRI-Structural isOtropic (TRISO) fuel silver release is addressed. For the first time, the uncertainties arising from the TRISO Fission Gas Release (FGR) model due to model inadequacy and experimental noise are quantified. Also, the Bayesian capabilities are applied to the calibration of the MATPRO creep model, a widely used model in several fuel assessment cases. The impact of the prediction uncertainties in the MATPRO model on the fuel cladding behavior as part of the TRIBULATION assessment case (which is an integral effects case) is investigated. This report concludes with a discussion on the future work for the UQ for computational models.
The Tristructural isotropic (TRISO)-coated particle fuel is a robust nuclear fuel proposed to be used for multiple modern nuclear technologies. Therefore, characterizing its safety is vital for the reliable operation of nuclear technologies. However, the TRISO fuel failure probabilities are small and the computational model is time consuming to evaluate them using traditional Monte Carlo-type approaches. In the paper, we present a multifidelity active learning approach to efficiently estimate small failure probabilities given an expensive computational model. Active learning suggests the next best training set for optimal subsequent predictive performance and multifidelity modeling uses cheaper low-fidelity models to approximate the high-fidelity model output. After presenting the multifidelity active learning approach, we apply it to efficiently predict TRISO failure probability and make comparisons to the reference results.
Although the no-u-turn sampler (NUTS) is a widely adopted method for performing Bayesian inference, it requires numerous posterior gradients which can be expensive to compute in practice. Recently, there has been a significant interest in physics-based machine learning of dynamical (or Hamiltonian) systems and Hamiltonian neural networks (HNNs) is a noteworthy architecture. But these types of architectures have not been applied to solve Bayesian inference problems efficiently. We propose the use of HNNs for performing Bayesian inference efficiently without requiring numerous posterior gradients. We introduce latent variable outputs to HNNs (L-HNNs) for improved expressivity and reduced integration errors. We integrate L-HNNs in NUTS and further propose an online error monitoring scheme to prevent sampling degeneracy in regions where L-HNNs may have little training data. We demonstrate L-HNNs in NUTS with online error monitoring consider several complex high-dimensional posterior densities and compare its performance to NUTS.
Earthquakes affect numerous nuclear power plants around the world and safe earthquake design requires that the physics of reactors subjected to earthquakes is well understood. This project attempts to understand the physics of molten salt reactors (MSRs) subjected to earthquakes using coupled multiphysics tools in MOOSE. A beyond-design-basis scenario of an unprotected pump trip (ULOFA) during an earthquake is analyzed.
Pool-type Molten Salt Reactors (MSRs) are a promising advanced nuclear reactor concept relying on passive physical behavior to provide increased safety characteristics. The high heat capacity, high boiling point, unpressurized liquid salt contains the nuclear fuel and as it passes through the vessel achieves criticality and produces nuclear power. The salt then carries away the heat produced along with the delayed neutron precursors towards heat exchangers and primary pumps before entering the core. This forms a highly coupled multiphysics problem between neutronics and fluid flow, which makes modeling these reactors challenging. The safety/licensing case heavily relies on modeling and simulation to prove their safety, and significant effort has been expanded by the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program to develop the appropriate simulation tools. Their safety/licensing will require a thorough analysis of their behavior during earthquake-based transients. Earthquakes are a transient condition commonly experienced in many regions, and are safely survived by dozens of reactors every year. Nonetheless, significant conservatisms were introduced with limited predictive modeling and simulation available. The first-of-a-kind capability developed by this research will allow for engineers to analyze ranges of design-basis and beyond-design basis accident scenarios to assess the integrity of the reactor vessel and surrounding system, possibly allowing for limiting the costly conservatisms in the design. This projects uses MASTODON, Griffin, and Pronghorn in the NEAMS ecosystem to model the coupled multiphysics problem posed by an earthquake in a molten salt reactor. It marches through a number of number of coupling schemes, from uncoupled simulations to tightly two-way coupled simulations. The appropriate level of coupling for these simulations will then be determined based on the accuracy in the quantities of interest (QoI) and computational effort involved in each scheme. The QoIs chosen for this paper include the mass flow rate across the heat exchanger (for Pronghorn simulations), power output of the reactor (for Griffin simulations), and the maximum Von Mises stress (for MASTODON simulations). This summary paper reports on the current progress of this project, with 2D tightly coupled multiphysics results.
The U.S. Department of Energy’s Nuclear Energy Advanced Modeling and Simulation (NEAMS) program aims to develop predictive capabilities by applying computational methods to the analysis and design of advanced reactor and fuel cycle systems. This program has been providing engineering-scale support for the development of BISON, a high-fidelity and high-resolution fuel performance tool. This study was motivated by the need to incorporate more physics-based models in BISON in order to foster tri-structural isotropic (TRISO) applications. This document details the integration of new modeling capabilities in BISON, including (1) development of pyrolytic carbon (PyC) and silicon carbide (SiC) layer anisotropic thermal and mass transport capabilities, (2) verification of the mass diffusion solution in TRISO modeling, (3) calibration of fission product diffusivity using Advanced Gas Reactor (AGR) experiments, (4) improved fission product release modeling by developing compact diffusion modeling capabilities, and (5) documentation of accelerated failure analysis on the BISON website. Improvements made to the diffusion models and parameters were documented and validated against AGR-1 and -2 experiment data.
Orientation and completion for well pairs that have been subjected to multi-zonal stimulation play a critical role in the long-term performance of an Enhanced Geothermal Reservoir. Enhanced geothermal systems often rely on preferential flow along fractures between well injection and production locations. Modeling this preferential flow using discrete fracture networks (DNF) relies on stochastic realizations of the DFN based on geological sampling. Here we present the development of a stochastic optimization methodology to determine well completion options in a discrete fracture network based on using parallel subset simulation. Stochastic optimization will provide insight into regions where placements of the injection and production wells are optimal. An example optimization of well-pair location optimization based on a deterministic-stochastic DFN model representing FORGE follows a discussion of the theory.
Modeling the recovery of interdependent critical infrastructure is a key component of quantifying and optimizing societal resilience to disruptive events. However, simulating the recovery of large-scale interdependent systems under random disruptive events is computationally expensive. Therefore, we propose the application of Deep Operator Networks (DeepONets) in this paper to accelerate the recovery modeling of interdependent systems. DeepONets are ML architectures which identify mathematical operators from data. The form of governing equations DeepONets identify and the governing equation of interdependent systems recovery model are similar. Therefore, we hypothesize that DeepONets can efficiently model the interdependent systems recovery with little training data. We applied DeepONets to a simple case of four interdependent systems with sixteen states. DeepONets, overall, performed satisfactorily in predicting the recovery of these interdependent systems for out of training sample data when compared to reference results.
The Grizzly code is being developed to address degradation issues in nuclear reactor structures and components. For light-water reactors, Grizzly currently has capabilities to simulate degradation processes and their effects on structural integrity in two key areas: reactor pressure vessels (RPVs) and reinforced concrete structures. This report documents improvements made to Grizzly’s ability to address both of these structural systems. For RPVs, the reduced-order models (ROMs) used in fracture mechanics calculations have been expanded to allow their application over a broader range of the parameter space than was permitted by the previous models. The ROMs currently used in Grizzly for the evaluation of flaws that are fully embedded within the RPV (as opposed to surface-breaking flaws) are based on a model that is known to be conservative, indicating higher stress intensity factors than would be obtained from direct simulations. A more accurate model that eliminates these excess conservatisms has been recently included in the American Society of Mechanical Engineers Boiler and Pressure Vessel Code but was not applicable for flaws near the RPV surface, which is where the most critical flaws are usually located. That model has recently been extended for increased applicability in this near-surface region. The ROMs for embedded flaws in the Grizzly code have been expanded to include these recent extensions, which permit their use in a much broader set of cases than previously possible. Direct 3D simulations have been used to check these ROMs and have shown good agreement in most cases, although there are still some cases that need further investigation. There are considerable benefits to using these these more accurate and less conservative ROMs for embedded flaws. On a benchmark probabilistic fracture mechanics problem tested here, the conditional probability of fracture initiation computed for a population of flaws in a single plate in an RPV decreased by over a factor of 3. To address aging in reinforced concrete structures, a capability to simulate multiple degradation mechanisms, including alkali-silica reaction and radiation-induced volumetric expansion has been developed in Grizzly over the past several years. This had previously been demonstrated on laboratory-scale specimens but not on full-scale nuclear concrete structures with reinforcement. To demonstrate the applicability of Grizzly to the analysis of large-scale structures of interest, a full 3D model of a representative reinforced concrete structure, including a complex arrangement of reinforcing bars, was developed and demonstrated in Grizzly.
This report summarizes a joint effort between Argonne National Laboratory, Idaho National Laboratory, and Los Alamos National Laboratory to develop and deploy constitutive models targeted at predicting the life of Grade 91 alloy components subjected to high temperature environments typical of those that structural components in advanced nuclear reactors would experience. Two distinct, but complementary constitutive modeling approaches have been taken here. The first employs a phenomenological viscoplastic model for which parameters have been calibrated based on experimental data for a wide range of Grade 91 alloy that has undergone a variety of processing. A Bayesian approach was used to derive distributions of uncertain parameters for this model based on this data set. The second approach is a reduced order model suitable for engineering-scale analysis that is based on the results of a large set of mesoscale simulations. Mesoscale models allow for the microstructure and composition of a particular alloy to be directly taken into account in the computation of the viscoplastic response, but are computationally expensive, which makes it impractical to directly call those models for the material constitutive response in an engineering-scale simulation. The reduced-order representation of the response of the underlying model used here allows for an engineering-scale model to take into account the characteristics of the underlying microstructure, while only incurring a reasonable computational expense. Both of these approaches have different strengths, and are applicable for different parts of the design/analysis process. The phenomenological models can be readily parameterized based on a set of experimental data for a given class of materials and used for scoping calculations. Once a specific material is chosen and adequately characterized, the reduced order models can accurately predict the response of that specific alloy, and because the models are based on predictive models of the underlying microstructure, they can be used to more confidently predict the response under conditions in regions where there is limited experimental data. Both of these models have been integrated in the Grizzly code, which is used here to perform proof-of-concept uncertainty quantification analyses of a simple component under prototypical conditions. The built- in stochastic analysis capabilities in the MOOSE framework that Grizzly is built on are used here to run large sets of simulations for this uncertainty quantification analysis. As would be expected, because the reduced order models are developed for a much more tightly defined alloy, they predict tighter distributions of the time to failure than the phenomenological models, which are calibrated to a broader set of data. Also important is that these simulations demonstrate that a reduced order modeling approach can be successfully deployed to propagate uncertainties from the material scale to practical engineering-scale component simulations.