CAE Automotive: Projection of simulation results on future product variance using machine learning and advanced analytics

Abstract

hroughout the automotive industry virtual simulations are used to ensure new models meet company internal as well as legal safety and quality criteria. For example a 30 Finite Element (FE) model of the car is built by transforming geometric CAD data into FE data. This simulation model is then used for evaluation of the car structure in various crash scenarios such as the Euro NCAP frontal impact test. This approach offers several advantages over physical testing of car models, e.g. lower cost, running a series of development cycles or availablilty of virtual validation at a point of time in development where physical car models are not yet available.
Nonetheless. the process of virtual simulation requires high technical knowhow at the OEM as well as tool developers. Last but not the least, simulation model building is a very time consuming process. This results in a limited number of executed simulations and it is essential that failure rate of the simulations be kept to a bare minimum.
The basic idea is to use historicaVstatistical simulation data as input for a Machine Learning-approach to create a knowledge-database. All OE Ms have accumulated simulation data from the past 10-15 years in SLM systems like SimManager from MSC. The knowledge retrieved from the simulation data could then be used to create suggestions for the engineer to improve the FE model to fulfil the requirements. lt is expected that the number of simulation runs for new car development projects could be reduced with this approach creating a significant business benefit.
The NTT DATA approach has already been used in the semiconductor industry to reduce development times of microprocessors. The CAE Machine Learning-method has been adapted to be used for a showcase in the automotive industry, and is presented as an example.