Design Support Technology for Efficient Discovery of Structural Knowledge in Car Body Development

(1) Nonlinear Sparse Modelling with Evolutionary Factor Extraction and Selection Probability of Factors

Abstract

We have developed the interactive design support technology in order to efficiently obtain finding of lightweight-car-body structure with high performance. The challenge with this technology is nonlinear contribution analysis with a small number of samples, a large number of variables and strong nonlinearity. In addition, complex results by nonlinear contribution analysis makes it difficult to provide engineers knowledge in weight reduction. To solve these problems, we propose the nonlinear contribution analysis method that combines the technique to derive non-linear basis functions using the evolutionary computation and sparse modelling using selection probability of factors. Additionally, we also apply the post processing that excludes factors difficult to interpret based on engineering viewpoint to analysis result. This method indicates high accuracy and robustness with a small number of samples, and the post processing decreases the number of terms in the model while maintaining high accuracy and improves interpretability. As the application result to car body structure problem, this method is shown to be effective in discovering bottlenecks and knowledge in weight reduction.