Department of Mechanical Engineering Graduate Seminar

PRESENTATION: Differentiable Finite Element Method - A Paradigm Change in Computational Mechanics. Traditional finite element method is not formulated as a differentiable mapping. It can predict an output from an input, but cannot, at least not directly, tell how the output would change if the input were altered. Modern AI computations hinge on differentiable mapping, that is, functions that produce not only the output, but also the derivatives. This is enabled by automatic differentiation (auto-diff), a modern computation infrastructure. Lack of differentiability in classical FEM has been a big burden in finite element code development, and is a major tumbling block that hinders the integration with AI and other components of engineering analysis. In this presentation, we will introduce differentiable finite element method (diff-FEM), focusing on its impact on both code development and computational architecture. We show that diff-FEM enables seamless integration with inverse analysis and design-optimization, and can be naturally pipelined to AI computation.

PRESENTER: Dr. Jia Lu is a professor in the Department of Mechanical Engineering, The University of Iowa. He joined UI in 2001 after a brief employment at ANSIS Inc. Dr. Lu’s research interest has been in computational mechanics, soft tissue constitutive modeling, nonlinear elasticity, and inverse problems. He has published over 80 journal papers in these areas.