Machine Learning and Reduced Order Modelling for HPC Intensive CAE Application


Machine learning and Reduced Order Modelling (ROM) techniques are interpolation methods exploiting data sets derived from existing virtual or experimental setup. They are essential to the concept of Digital Twins since they provide the missing link for both rapid re-design and operational evaluations in real-time. While the starting point is a DOE-type design with sufficient space filling properties, ML/ROM are different from response surface methods (RSM) where the approximation functions (surrogate model), often representing variations of a scalar entity within the design space, are imposed by the shape and nature of the fitting surface or a prescribed equation which also affects the number of runs required. ML/ROM techniques exploit the known physical behavior represented by the “modal” contents or similitudes expressed in terms of clusters of responses (ROM) in conjunction with various simple model prediction techniques (ML). These techniques provide completely new frontiers for cost effective and accelerated design and optimization of products or processes. In what follows we shall provide state-of-the art solutions to typical applications in Crash & Safety, CFD, multi-physics and optimization. All have employed in one way or another. The software package ODYSSEE (CAE & A-EYE) has been exploited and will be presented via various industrial applications. We shall demonstrate how ML/ROM can exploit HPC with highest efficiency and for real-time design and optimization.


Dr. Kambiz Kayvantash, Sr. Director of AI/ML Applications for Design and Engineering, Hexagon Manufacturing Intelligence

Dr. Kambiz Kayvantash is the Sr. Director of AI/ML Applications for Design and Engineering at Hexagon Manufacturing Intelligence. With 40 years of industrial and academic experience he is currently European Expert for AI/ML & road transport safety and continues to provide lectures at various academic institutes.