Publications

Abstract not provided.

Thermal sprayed metal coatings are used in many industrial applications, and characterizing the structure and performance of these materials is vital to understanding their behavior in the field. X-ray Computed Tomography (CT) machines enable volumetric, nondestructive imaging of these materials, but precise segmentation of this grayscale image data into discrete material phases is necessary to calculate quantities of interest related to material structure. In this work, we present a methodology to automate the CT segmentation process as well as quantify uncertainty in segmentations via deep learning. Neural networks (NNs) are shown to accurately segment full resolution CT scans of thermal sprayed materials and provide maps of uncertainty that conservatively bound the predicted geometry. These bounds are propagated through calculations of material properties such as porosity that may provide an understanding of anticipated behavior in the field.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Quasi-static structural finite-element models of an aluminum-framed crystalline silicon photovoltaic module and a glass-glass thin-film module were constructed and validated against experimental measurements of deflection under uniform pressure loading. Specific practices in the computational representation of module assembly were identified as influential to matching experimental deflection observations. Additionally, parametric analyses using Latin hypercube sampling were performed to propagate input uncertainties related to module materials, dimensions, and tolerances into uncertainties in simulated deflection. Sensitivity analyses were performed on the uncertainty quantification datasets using linear correlation coefficients and variance-based sensitivity indices to elucidate key parameters influencing module deformation. Results identified edge tape and adhesive material properties as being strongly correlated to module deflection, suggesting that optimization of these materials could yield module stiffness gains at par with the conventionally structural parameters, such as glass thickness. This exercise verifies the applicability of finite-element models for accurately predicting mechanical behavior of solar modules and demonstrates a workflow for model-based parametric uncertainty quantification and sensitivity analysis. Applications of this capability include the assessment of field environment loads, derivation of representative loading conditions for reduced-scale testing, and module design optimization, among others.

Abstract not provided.

Deep learning has been successfully applied to the segmentation of 3D Computed Tomography (CT) scans. Establishing the credibility of these segmentations requires uncertainty quantification (UQ) to identify untrustworthy predictions. Recent UQ architectures include Monte Carlo dropout networks (MCDNs), which approximate deep Gaussian processes, and Bayesian neural networks (BNNs), which learn the distribution of the weight space. BNNs are advantageous over MCDNs for UQ but are thought to be computationally infeasible in high dimension, and neither architecture has produced interpretable geometric uncertainty maps. We propose a novel 3D Bayesian convolutional neural network (BCNN), the first deep learning method which generates statistically credible geometric uncertainty maps and scales for application to 3D data. We present experimental results on CT scans of graphite electrodes and laser-welded metals and show that our BCNN outperforms an MCDN in recent uncertainty metrics. The geometric uncertainty maps generated by our BCNN capture distributions of sigmoid values that are interpretable as confidence intervals, critical for applications that rely on deep learning for high-consequence decisions.

Abstract not provided.

Contact instability may occur during discharging because of void formation. Xin Zhang et al. suggested a method to predict the conditions leading to instability. The development of solid-state batteries has encountered a number of problems due to the complex interfacial contact conditions between lithium (Li) metal and solid electrolytes (SEs). Recent experiments have shown that applying stack pressure can ameliorate these problems. Here, we report a multi-scale three-dimensional time-dependent contact model for describing the Li-SE interface evolution under stack pressure. Our simulation considers the surface roughness of the Li and SEs, Li elastoplasticity, Li creep, and the Li metal plating/stripping process. Consistency between the very recent experiments from two different research groups indicates effective yield strength of the Li used in those experiments of 16 ± 2 MPa. We suggest that the preferred stack pressure be at least 20 MPa to maintain a relatively small interface resistance while reducing void volume.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Empirically-based correlations are commonly used in modeling and simulation but rarely have rigorous uncertainty quantification that captures the nature of the underlying data. In many applications, a mathematical description for a parameter response to some input stimulus is often either unknown, unable to be measured, or both. Likewise, the data used to observe a parameter response is often noisy, and correlations are derived to approximate the bulk response. Practitioners frequently treat the chosen correlation-sometimes referred to as the "surrogate"or "reduced-order"model of the response-as a constant mathematical description of the relationship between input and output. This assumption, as with any model, is incorrect to some degree, and the uncertainty in the correlation can potentially have significant impacts on system responses. Thus, proper treatment of correlation uncertainty is necessary. In this paper, a method is proposed for high-level abstract sampling of uncertain data correlations. Whereas uncertainty characterization is often assigned to scalar values for direct sampling, functional uncertainty is not always straightforward. A systematic approach for sampling univariable uncertain correlations was developed to perform more rigorous uncertainty analyses and more reliably sample the correlation space. This procedure implements pseudo-random sampling of a correlation with a bounded input range to maintain the correlation form, to respect variable uncertainty across the range, and to ensure function continuity with respect to the input variable.

Recent advancements in micro-scale additive manufacturing techniques have created opportunities for design of novel electrode geometries that improve battery performance by deviating from the traditional layered battery design. These 3D batteries typically exhibit interpenetrating anode and cathode materials throughout the design space, but the existing well-established porous electrode theory models assume only one type of electrode is present in each battery layer. We therefore develop and demonstrate a multielectrode volume-averaged electrochemical transport model to simulate transient discharge performance of these new interpenetrating electrode architectures. We implement the new reduced-order model in the PETSc framework and asses its accuracy by comparing predictions to corresponding mesoscale-resolved simulations that are orders of magnitude more computationally-intensive. For simple electrode designs such as alternating plates or cylinders, the volume-averaged model predicts performance within ∼2% for electrode feature sizes comparable to traditional particle sizes (5-10μm) at discharge rates up to 3C. When considering more complex geometries such as minimal surface designs (i.e. gyroid, Schwarz P), we show that using calibrated characteristic diffusion lengths for each design results in errors below 3% for discharge rates up to 3C. These comparisons verify that this novel model has made reliable cell-scale simulations of interpenetrating electrode designs possible.

Battery electrodes are composed of polydisperse particles and a porous, composite binder domain. These materials are arranged into a complex mesostructure whose morphology impacts both electrochemical performance and mechanical response. We present image-based, particle-resolved, mesoscale finite element model simulations of coupled electrochemical-mechanical performance on a representative NMC electrode domain. Beyond predicting macroscale quantities such as half-cell voltage and evolving electrical conductivity, studying behaviors on a per-particle and per-surface basis enables performance and material design insights previously unachievable. Voltage losses are primarily attributable to a complex interplay between interfacial charge transfer kinetics, lithium diffusion, and, locally, electrical conductivity. Mesoscale heterogeneities arise from particle polydispersity and lead to material underutilization at high current densities. Particle-particle contacts, however, reduce heterogeneities by enabling lithium diffusion between connected particle groups. While the porous composite binder domain (CBD) may have slower ionic transport and less available area for electrochemical reactions, its high electrical conductivity makes it the preferred reaction site late in electrode discharge. Mesoscale results are favorably compared to both experimental data and macrohomogeneous models. This work enables improvements in materials design by providing a tool for optimization of particle sizes, CBD morphology, and manufacturing conditions.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Static structural finite element models of an aluminum-framed crystalline silicon (c-Si) photovoltaic (PV) module and a glass-glass thin film PV module were constructed and validated against experimental measurements of deflection under uniform pressure loading. Parametric analyses using Latin Hypercube Sampling (LHS) were performed to propagate simulation input uncertainties related to module material properties, dimensions, and manufacturing tolerances into expected uncertainties in simulated deflection predictions. This exercise verifies the applicability and validity of finite element modeling for predicting mechanical behavior of solar modules across architectures and enables computational models to be used with greater confidence in assessment of module mechanical stressors and design for reliability. Sensitivity analyses were also performed on the uncertainty quantification data sets using linear correlation coefficients to elucidate the key parameters influencing module deformation. This information has implications on which materials or parameters may be optimized to best increase module stiffness and reliability, whether the key optimization parameters change with module architecture or loading magnitudes, and whether parameters such as frame design and racking must be replicated in reduced-scale reliability studies to adequately capture full module mechanical behavior.

Deep learning segmentation models are known to be sensitive to the scale, contrast, and distribution of pixel values when applied to Computed Tomography (CT) images. For material samples, scans are often obtained from a variety of scanning equipment and resolutions resulting in domain shift. The ability of segmentation models to generalize to examples from these shifted domains relies on how well the distribution of the training data represents the overall distribution of the target data. We present a method to overcome the challenges presented by domain shifts. Our results indicate that we can leverage a deep learning model trained on one domain to accurately segment similar materials at different resolutions by refining binary predictions using uncertainty quantification (UQ). We apply this technique to a set of unlabeled CT scans of woven composite materials with clear qualitative improvement of binary segmentations over the original deep learning predictions. In contrast to prior work, our technique enables refined segmentations without the expense of the additional training time and parameters associated with deep learning models used to address domain shift.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

We offer an explanation for how dendrite growth can be inhibited when Li metal pouch cells are subjected to external loads, even for cells using soft, thin separators. We develop a contact mechanics model for tracking Li surface and sub-surface stresses where electrodes have realistically (micron-scale) rough surfaces. Existing models examine a single, micron-scale Li metal protrusion under a fixed local current density that presses more or less conformally against a separator or stiff electrolyte. At the larger, sub-mm scales studied here, contact between the Li metal and the separator is heterogeneous and far from conformal for surfaces with realistic roughness: the load is carried at just the tallest asperities, where stresses reach tens of MPa, while most of the Li surface feels no force at all. Yet, dendrite growth is suppressed over the entire Li surface. To explain this dendrite suppression, our electrochemical/mechanics model suggests that Li avoids plating at the tips of growing Li dendrites if there is sufficient local stress; that local contact stresses there may be high enough to close separator pores so that incremental Li+ ions plate elsewhere; and that creep ensures that Li protrusions are gradually flattened. These mechanisms cannot be captured by single-dendrite-scale analyses.