Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : September 25, 2019 - September 27, 2019
In this study, we construct an analysis and visualization method that can be easily understood by a person who does not have much knowledge of optimization using the data set obtained from Mazda's multi-objective car structure optimization problem. First, we perform filtering according to the gauge thickness of feasible solutions using hierarchical clustering, and then focus on the solution pair between clusters and visualize them with parallel coordinate plots. As a result, it is confirmed that even if the solution pair have no weight difference, not only the difference in thickness combination but also the constraint conditions close to the feasibility boundary are different. This implies that if there are multiple ways of combining design variables with the same objective function, it is necessary to filtering the data set in advance for extracting design information.
