Plenary and Semi-Plenary Speakers

Abstract

Adaptive mesh refinement (AMR) is one of several techniques for dynamically modifying the spatial resolution of a simulation in particular regions of the spatial domain. Block-structured AMR specifically refines the mesh by defining locally structured regions with finer spatial, and possibly temporal, resolution. This combination of locally structured meshes within an irregular global hierarchy is in some sense the best of both worlds in that it enables regular local data access while enabling greater flexibility in the overall computation.

Originally, block-structured AMR was designed for solving hyperbolic conservation laws with explicit time-stepping; in this case the changes to solution methodology in transforming a single-level solver to an AMR-based solver are relatively straightforward. AMR has come a long way, however, and the more complex the simulation we want to perform, the more we gain from leveraging trusted AMR software frameworks.  In this talk I will give an overview of block-structured AMR for different types of applications and will focus on a few key exemplars for how to think about adaptivity for multiphysics simulations.

Bio

Ann Almgren is a senior scientist in the Applied Mathematics and Computational Research Division of Lawrence Berkeley National Laboratory and the Department Head of Berkeley Lab's Applied Mathematics Department. Her primary research interest is in computational algorithms for solving partial differential equations in a variety of application areas. Her current projects include the development and implementation of new multiphysics algorithms in high-resolution adaptive mesh codes that are designed for the latest hybrid architectures. She is a SIAM Fellow, serves on the editorial boards of CAMCoS, IJHPCA and Phil. Trans. A., and co-leads LBL's Computing Sciences Area Mentoring Program. In 2023 she was awarded the Berkeley Lab Director's Award for Exceptional Scientific Achievement. Prior to coming to LBL she worked at the Institute for Advanced Study in Princeton, NJ, and at Lawrence Livermore National Lab.

Abstract

In the context of scientific and industrial applications, we often have to deal with unstructured spatial and temporal data obtained from numerical simulations and the real world. The data is usually in the form of a mesh or a point cloud. In this context, graph neural networks (GNNs) have proved to be effective tools to reproduce the data behavior; however, depending on the physical nature of the datasets, variations of vanilla GNNs have to be considered to ensure accurate results. Furthermore, when only a point cloud is available, one is faced with the question of how (and whether) to build a corresponding graph.

In this presentation we go over general challenges in the use of GNNs in computational mechanics and fluid dynamics. Special attention will be given to particle-accelerator simulations; a computationally demanding class of problems for which rapid design and real-time control are challenging. We propose a machine learning-based surrogate model that leverages both graph and point networks to predict particle-accelerator behavior across different machine settings. Our model is trained on high-fidelity simulations of electron beam acceleration, capturing complex, nonlinear interactions among particles distributed across several initial state dimensions and machine parameters. Our results show the model’s to accurately track electron beams at downstream observation points, outperforming baseline graph convolutional networks. This framework accommodates key symmetries inherent in particle distributions, enhancing stability and interpretability. We also go over the extension of these architectures to autoregressive tracking across multiple timesteps. This research offers a powerful approach to reducing computational demands in particle-accelerator simulations, contributing to advancements in real-time optimization and control. This work had been performed at Stanford in collaboration with SLAC.

Bio

Marta D'Elia is the Director of AI and ModSim at Atomic Machines and an Adjunct Professor at the Institute for Computational and Mathematical Engineering (Stanford University). She previously worked at Pasteur Labs, Meta, and Sandia National Laboratories as a Principal Scientist and Tech Lead. She holds a PhD in Applied Mathematics and master's and bachelor's degrees in Mathematical Engineering. Her work deals with development and analysis of machine-learning models and optimal design and control for industrial applications, especially in the context of MEMS and robotics. She is an expert in nonlocal modeling and simulation, optimal control, and scientific machine learning. She is an Associate Editor of SIAM and Nature journals, a member of the SIAM industry committee, the Vice Chair of the SIAM Northern California section, and a member of the NVIDIA advisory board for scientific machine learning.

Bio

Roger Ghanem holds the Tryon Chair of Stochastic Methods and Simulation at the University of Southern California where he is also Professor in the Departments of Civil & Environmental Engineering and Aerospace & Mechanical Engineering. Dr. Ghanem research is in the areas of stochastic analysis and computational science. For the past thirty years, he has worked on scientific, mathematical and algorithmic aspects of uncertainty quantification with applications from across science and engineering. His recent work addresses challenges presented by modeling errors and complex interacting systems such as those exhibiting multiscale and multiphysics behaviors. Dr. Ghanem has co-authored over 200 journal articles related to stochastic systems and predictive science. He has supervised the research of over 20 postdoctoral associates and 40 PhD students.

Dr. Ghanem is the President of the International Association for Structural Safety and Reliability (IASSAR) and has previously served as President of EMI, on the Executive Council of USACM, as Chair of the SIAM-SIAG on Uncertainty Quantification, and on USNCTAM. He is fellow of AAAS, SIAM, USACM, IACM, and EMI and Distinguished Member of ASCE. Dr. Ghanem is the recipient of a number of awards acknowledging his research and teaching contributions. His research has been supported by NSF, ONR, AFOSR, DARPA, DOE, and a number of industries including GM, GE, Daikin, Taisei, Takenaka, and Shimizu.

Abstract

Architected materials (or metamaterials) have become not only a popular solution for many applications that require materials with specific or extreme properties, but they also continue to present challenges for computational mechanics across scales. Besides the need for techniques that accurately predict a material’s properties based on its small-scale architecture (the forward problem), an even bigger challenge are methods that enable the optimization of the effective material performance through a careful design of the small-scale architecture (the inverse problem). We will discuss opportunities for and solutions provided by computational multiscale modeling for both problems. This includes the efficient simulation of architected materials by on-the-fly homogenization, the efficient prediction of fracture properties by coarse-graining, the design of novel spatially graded metamaterials for wave guidance, and the generative design of architected materials by methods of machine learning. For each example, we will demonstrate how advanced and targeted computational mechanics techniques have enabled new directions in modeling and design, and how this field is still offering many challenges to be addressed.

Bio

Dennis M. Kochmann received his education at Ruhr-University Bochum in Germany and at the University of Wisconsin-Madison. After postdoc positions at Wisconsin and the California Institute of Technology, he joined the California Institute of Technology as Assistant Professor of Aerospace in 2011 and in 2016 was promoted to Professor of Aerospace, a position he held through 2019. Since April 2017 he has been Professor of Mechanics and Materials at ETH Zurich, where he also served as Deputy Head of Department. His research focuses on the link between microstructure and properties of natural and architected materials, which includes the development of theoretical, computational, and experimental methods to bridge across scales from nano to macro. In recent years, this also includes the use of machine learning to tackle problems in multiscale material modeling and computational mechanics. His research has been recognized by, among others, IUTAM’s Bureau Prize in Solid Mechanics, GAMM’s Richard von Mises Prize, ASME’s T.J.R. Hughes Young Investigator Award, an ERC Consolidator Grant, an NSF CAREER Award, and IACM’s John Argyris Award. He also enjoys teaching and has received Caltech’s Graduate Student Council’s Teaching Award as well as ETH’s Golden Owl.

Abstract

Leveraging recent advances in machine learning (ML) and artificial intelligence (AI), we show that data-driven methods are able to capture the mechanical response of a variety of tissues with unparalleled accuracy. While it is not surprising that ML methods are able to interpolate material response, we show how to build constitutive models that have the desirable flexibility of ML tools but are also designed to satisfy physics constraints a priori through the careful design of the ML architecture. We showcase data-driven methods for tissue mechanics from hyperelasticity, to damage and non-equilibrium visocoelasticity, to tissue growth and remodeling. We also discuss how to extend data-driven frameworks to capture the distribution of material response across individuals, or even the heterogeneous distribution of material properties across a single biological tissue sample. Implementation into popular finite element packages such as Abaqus contributes to the further advancement of ML and AI in computational mechanics and bring these tools closer to clinical application. Example applications we demonstrate include skin expansion and reconstructive surgery.

Bio

Dr. Buganza-Tepole is an Associate Professor of Mechanical Engineering at Columbia University. He obtained his Ph.D. in Mechanical Engineering from Stanford University in 2015 and was a postdoctoral fellow at Harvard University before joining Purdue as a faculty member in 2016. He has been a Miller Visiting Professor at UC Berkeley and a Harrington Fellow at The University of Texas. His group studies the interplay between mechanics and mechanobiology of soft tissue, with skin as a model system. Using computational simulation, machine learning, and experimentation, his group seeks to characterize the multi-scale mechanics of tissues to understand the fundamental mechanisms of mechano-adaptation in order to improve clinical diagnostics and interventional tools.

Abstract

Inspired by the remarkable success of machine learning (ML) in fields such as computer vision and language modeling, the scientific community is increasingly excited to harness its potential for addressing societal challenges. However, realizing this potential requires a paradigm shift in data-intensive scientific discovery, as ‘black box’ ML models often fail to generalize to unseen scenarios and may produce results that conflict with established scientific understanding of the underlying phenomena.

This talk presents an overview of a new generation of machine learning algorithms, where scientific knowledge is deeply integrated in the design and training of machine learning models to accelerate scientific discovery. These knowledge-guided machine learning (KGML) techniques are fundamentally more powerful than standard machine learning approaches, and are particularly relevant for scientific and engineering problems that are traditionally addressed via process-guided (also called mechanistic or first principle-based) models, but whose solutions are hampered by incomplete or inaccurate knowledge of physics or underlying processes. While this talk will illustrate the potential of the KGML paradigm in the context of environmental problems (e.g., Ecology, Hydrology, Agronomy, climate science), the paradigm has the potential to greatly advance the pace of discovery in any discipline where mechanistic models are used.

Bio

Vipin Kumar is a Regents Professor and holds William Norris Chair in the department of Computer Science and Engineering at the University of Minnesota. His research spans data mining, high-performance computing, and their applications in Climate/Ecosystems and health care. His research has resulted in the development of the concept of isoefficiency metric for evaluating the scalability of parallel algorithms, as well as highly efficient parallel algorithms and software for sparse matrix factorization (PSPASES) and graph partitioning (METIS, ParMetis, hMetis). He has authored over 400 research articles, and co-edited or coauthored 11 books including two widely used text books "Introduction to Parallel Computing", "Introduction to Data Mining", and a recent edited collection, “Knowledge Guided Machine Learning”. Kumar's current major research focus is on knowledge-guided machine learning and its applications to understanding the impact of human induced changes on the Earth and its environment. Kumar’s research on this topic has been funded by NSF’s AI Institute, BIGDATA, INFEWS, STC, GCR, and HDR programs, as well as ARPA-E, DARPA, and USGS. He also served as the Lead PI of a 5-year, $10 Million project,  "Understanding Climate Change - A Data Driven Approach", funded by the NSF's Expeditions in Computing program (2010-2015).

Kumar has been elected a Fellow of the Association for Advancement of Artificial Intelligence (AAAI), the American Association for Advancement for Science (AAAS), Association for Computing Machinery (ACM), Institute of Electrical and Electronics Engineers (IEEE), and Society for Industrial and Applied Mathematics (SIAM). Kumar's foundational research in data mining and high performance computing has been honored by the ACM SIGKDD 2012 Innovation Award, which is the highest award for technical excellence in the field of Knowledge Discovery and Data Mining (KDD), the 2016 IEEE Computer Society Sidney Fernbach Award, one of IEEE Computer Society's highest awards in high performance computing, and Test-of-time award from 2021 Supercomputing conference (SC21).

Abstract

Predicting the evolution of microstructurally small cracks (MSCs) is a long-standing, scientifically challenging problem with critical implications for materials design and structural prognosis. In fatigue applications, fatigue cracks are known to spend a majority of their total lifetime—upwards of 90%—being “microstructurally small” (that is, having a size on the order of that of the dominant microstructural feature). The ability to accurately predict the earliest stages of MSC evolution is critically needed to make accurate estimates of the remaining useful life of components, to explain observed scatter in total fatigue life among a population of components, to aid in the development of crack-resistant microstructures, to enable digital twin technologies, and to qualify new materials like additively manufactured metals. However, the evolution of cracks at microstructure scale remains challenging to predict and simulate due in part to the computational expense associated with resolving the requisite microstructural features in 3D while simultaneously capturing the formation and evolution of a discontinuity (i.e., crack). This talk will highlight some of the recent advancements toward addressing this challenge, including the implementation of new damage models within computationally efficient fast Fourier transform (FFT) codes and the integration of high-fidelity, physics-based modeling with machine learning. The new modeling developments serve to accelerate predictions of realistic MSC evolution in 3D microstructures across a broad range of loading regimes, enabling high-throughput predictions of strength, ductility, and fatigue life for a variety of materials.

Bio

Dr. Ashley Spear is a Presidential Scholar at the University of Utah, where she is an Associate Professor in the Department of Mechanical Engineering with adjunct appointments in Materials Science & Engineering and the Kahlert School of Computing. She directs the Multiscale Mechanics & Materials Laboratory, which specializes in integrating physics-based modeling, data science, and experiments to examine deformation, fatigue, and fracture in a wide range of materials. She received her B.S. in Architectural Engineering from the University of Wyoming and Ph.D. in Civil Engineering from Cornell University. She is the recipient of the Constance Tipper Medal from the International Congress on Fracture, the Young Investigator Award from the Air Force Office of Scientific Research, the TMS Early Career Faculty Fellow Award, the ASTM International Additive Manufacturing Young Professional Award, and the National Science Foundation CAREER award.

Abstract

In this talk, I’ll present our work on advancing computational methods for multiphysics modeling, focusing on complex interactions between multiple physical phenomena, including fluid-structure, shock physics, and multiphase flows. My research emphasizes developing numerical approaches that integrate various discretization methods, e.g., mesh-based and meshfree, to accurately capture dynamic interactions across complex interfaces in multiphysics systems. A key component of this work includes the modified immersed finite element method and coupled particle-based techniques, designed to model deformable structures and multiphase interactions. To address the stability and convergence challenges inherent to multiphysics simulations, we developed partitioned coupling algorithms that ensure robust simulations and modular coupling framework. These methods enable adaptive, efficient interface tracking, essential for capturing rapid changes across coupled domains. I will discuss applications of these methods such as blood flow and soft-tissue modeling, shock physics in materials, and multiphase flows in environmental and aerospace systems. This talk is to provide an overview of these computational frameworks, highlighting how they improve simulation fidelity and coupling solution strategies for a range of complex engineering applications.

Bio

Prof. Lucy Zhang is a Professor in the Department of Mechanical, Aerospace & Nuclear Engineering and Associate Dean for Research Innovations, Partnerships, and Workforce Development at Rensselaer Polytechnic Institute (RPI). She is a Fellow of ASME. She received her B.S. from Binghamton University and obtained her M.S. and Ph.D. from Northwestern University, IL. She began her academic career as an assistant professor in Mechanical Engineering at Tulane University but relocated to RPI in 2006 due to Hurricane Katrina, where she was subsequently promoted to Associate Professor and then Professor. Her research interests are building advanced and robust computational tools and software for accurate and efficient multiphysics and multiscale simulations that can be used for engineering applications in biomechanics, micro and nano-mechanics, and defense projects. She served as a Program Director in the Mechanics of Materials and Structures (MoMS) and Biomechanics & Mechanobiology (BMMB) programs within the CMMI Division at the NSF. In addition to her research, she co-hosts the podcast series This Academic Life, sharing life experiences of academics in STEM fields.