Seminar talk on Thu. 02.07.26 by Prof. Pin Zhang (NUS) on "Knowledge-Informed Neural Networks for Computational Geomechanics"

Abstract:

This study presents recent advances in integrating domain knowledge of geomechanics with neural networks, referred to as knowledge-informed neural networks (KINNs), for computational geomechanics. Three bespoke KINN frameworks are introduced: (i) a thermodynamics-informed learning framework for the automatic discovery of constitutive relations of geomaterials directly from data; (ii) a stress-point integration scheme that requires neither labelled data nor explicit evaluation of  the Jacobian matrix, while still rigorously enforcing the consistency condition; and (iii) a variational principle-based neural network for solving boundary-value problems. Overall, this study demonstrates the potential of KINNs to bridge conventional geomechanics knowledge and modern neural-network methodologies. The underlying concept of KINNs is general and holds significant promise for the development of a broad range of innovative methods and applications in computational mechanics and engineering practice.

Bio:

Pin Zhang is a Presidential Young Professor in the Department of Civil and Environmental Engineering at the National University of Singapore (2024-present). Before NUS, he was a Royal Society Newton International Fellow at the University of Cambridge (2022-2024). He obtained a PhD from the Hong Kong Polytechnic University (2020-2022), including one year at the University of Oxford as a visiting scholar (2021-2022). His research focuses on computational geomechanics and intelligent geo-infrastructure, with particular emphasis on scientific machine learning for geomechanics, sustainable geomaterials, tunnelling, and underground geotechnics. He has received some international honors, including the ISSMGE Bright Spark Lecture Award, the ICE (UK) Early Career Award (Géotechnique) the HKIE Ringo Yu Award, the TUST Best Paper Award, and the ASCE-JEM Editor’s Choice Award,