Simplifying the Evaluation of Many Simulation Results with Machine Learning

Jochen Garcke (Fraunhofer SCAI)

Abstract:

Nowadays, many design measures are systematically applied and numerical simulations are performed to investigate their effect on design criteria. During the development process the design variations have to be analyzed and the simulations results have to be compared and evaluated, often a time-consuming process. We present a set of tools to simplify the analysis of this data. It allows to easily analyze the impact of model variations, including a structured data representation of design changes and output behavior.

As a concrete example, we consider investigating the crashworthiness of automobiles, where numerical simulations of car crashes play a significant role. For example, one can analyze and store the CAE model variations arising in the development process. Furthermore, one can identify on the hand significant behavior modes and on the other hand anomalies, e.g. distinct or unusual variations in mesh quantities such as deformations or plastic strain.