AEM Colloquium: Differentiable Computational Mechanics: Neural-Integrated and Data-Driven Modeling for Inelastic Solids and Geophysical Applications
Qizhi (“KaiChi") He, January 31, 2:30pm
University of Minnesota, Department of Civil, Environmental, and Geo-Engineering
Title: Differentiable Computational Mechanics: Neural-Integrated and Data-Driven Modeling for Inelastic Solids and Geophysical Applications
Abstract: We present a recent development in the hybrid computational framework that integrates physics-based numerical schemes with machine learning methods to address various forward and inverse problems in computational mechanics. Our focus is on applications involving complex material behaviors and coupling effects, exploring how physical laws can be effectively incorporated within these methods across varying levels of data availability. We introduce a variationally consistent physics-informed machine learning approach, termed the Neural-Integrated Meshfree (NIM) method, designed to improve accuracy and training efficiency for simulating large deformations and material nonlinearities. To this end, the NIM method employs a hybrid approximation strategy that combines neural network representations with customized basis functions. The effectiveness of the NIM method is demonstrated through a series of linear and nonlinear benchmark mechanics problems, including applications in identifying heterogeneous biological materials. We also extend this framework to model Lagrangian particle flow problems, showcasing its potential to handle complex material behaviors under extreme conditions. Additionally, in data-rich scenarios, we introduce a hybrid scheme that leverages data-driven learning models for solving coupled systems. Our results show that the proposed machine learning models can reliably learn operators to capture underlying physical processes, enabling efficient dimensionality reduction. Examples from geophysics and biology will be presented to highlight the versatility of these machine learning techniques in advancing scientific computing.
Bio: Dr. Qizhi (“KaiChi") He is an Assistant Professor in the Department of Civil, Environmental, and Geo-Engineering at the University of Minnesota (UMN). He received his M.A. in Applied Mathematics (2016) and Ph.D. in Structural Engineering and Computational Science (2018) from the University of California, San Diego. From 2019 to 2021, he worked as a postdoctoral research associate in Scientific Machine Learning Group at Pacific Northwest National Laboratory. His research focuses on developing advanced numerical methods and physics-integrated machine learning algorithms to predict complex mechanics in porous and composite material systems under extreme conditions, as well as advancing inverse modeling and data assimilation for large-scale multi-physics applications in solid mechanics, material design, and geophysics. Dr. He is a member of the ASCE/EMI technical committees on Computational Mechanics and Machine Learning in Mechanics and serves on the editorial board of Computers and Geotechnics.
