Physics-Informed Machine Learning for Engineering Applications

About the Webinar

Modeling complex physical systems governed by partial differential equations (PDEs) is a fundamental challenge across many civil engineering domains. Traditional numerical methods like finite element analysis can struggle with high-dimensional parametric PDEs or cases with limited training data. Physics-informed machine learning (PIML) provides a powerful alternative by combining neural networks with the governing physics described by PDEs. This webinar explores the core methodology of PIML and its applications through hands-on training. PIML embeds the known physics directly into the neural network architecture, either as hard constraints or via additional loss terms derived from the PDE residuals. The neural network then approximates the unknown solution while inherently satisfying the specified physical laws. We illustrate PIML techniques through examples of modeling nonlinear PDEs like Burgers’ equation describing fluid flows and heat flow. We will also discuss inverse problems estimating the PDE parameters. The flexibility and physics-grounding of PIML make it a broadly applicable tool for diverse civil engineering disciplines.

Presenter

Krishna Kumar is an assistant professor of the Fariborz Maseeh Department of Civil, Architectural, and Environmental Engineering and an affiliate member at the Oden Institute of Computational Sciences at UT Austin. Krishna received his PhD in Engineering at the University of Cambridge, UK, in multi-scale and multiphysics modeling. Krishna’s work involves developing AI-accelerated numerical methods. His research is on physics-based machine learning techniques, such as graph networks and differentiable simulators, to solve inverse and design problems and application of AI-accelerated and DiffSims to develop autonomous construction robots on the Moon and Mars. He is the PI and Director of Chishiki AI, an NSF-funded AI in Civil Engineering Ecosystem.