Towards Design Rule Extraction from Large Computational Datasets by Causal Correlation Analysis

Intuitive interpretation of the results from multi- objective numerical optimization of magnetically non-linear electrical machines is very challenging. The resulting designs are typically used “as they are” or tuned by trial and error, due to  lack of deeper understanding needed for the tuning in the multi- objective Optimum Design Space (ODS). The results consisting of large sets of generic and optimum designs contain invaluable information on the emerging design rules. We recommend causal correlation analysis for design rule extraction.