J. S. Chen earned his undergraduate degree from National Central University (1978-1982) in Taiwan, and received master's (1986) and Ph.D. (1989) from Northwestern University. He worked in GenCorp's Research Division from 1989 to 1994. From 1994 to 2001, he held a faculty position in the Mechanical Engineering Department of The University of Iowa before moving to UCLA in 2001, where he served as the Chair of Civil & Environmental Engineering Department from 2007 to 2012. He was the Chancellor's Professor in the Civil & Environmental Engineering Department at UCLA and also Professor of Mechanical & Aerospace Engineering Department and Mathematics Department. In 2013, he joined the Structural Engineering Department of UC San Diego as the inaugural holder of the William Prager Endowed Chair. He also is the director of the Center for Extreme Events Research and a Professor of Mechanical and Aerospace Engineering at the Jacobs School of Engineering at UC San Diego.
Materials modeling is traditionally based on constitutive or material laws to describe the explicit relationship among strain, stress, and state variables based on experimental observations, physical hypothesis, and mathematical simplifications. However, the phenomenological modeling process inevitably introduces errors due to limited data and mathematical assumptions in model parameter calibration, and they rely on pre-defined functions and often lack generality to capture full aspects of material behaviors. The strategy in data-driven materials modeling is to bypass the constitutive modeling step by formulating an optimization problem to search for the physically admissible state that satisfies equilibrium and compatibility and minimizes the distance to a material dataset. In this work, we aim to develop a “model-free” Manifold Learning enhanced data-driven computing approach to overcome the curse of dimensionality and the lack of generalization in the classical constitutive materials modeling approaches. We proposed manifold learning based on deep autoencoders for noise filtering, dimensionality reduction, and discovery of low-dimensional embedding space of the high-dimensional nonlinear material data. Demonstration problems include model-free modeling of biological tissues and development of Digital Twins of musculoskeletal systems. The application to path- and history-dependent.