Visualization and Data Mining of Multi-Objective Electric Machine Optimizations with Self-Organizing Maps: A Case Study on Switched Reluctance Machines

The traditional methods used for presenting and visualizing the non-dominating solutions in multi-objective electric machine optimization problems mainly include the parallel coordinate plots, the histogram plots and the scatter plots displaying the Pareto fronts approximations. However, their visualization performances degrade with the increase of non-dominating design candidates or the increase in the number of objectives to be optimized. In particular, since the aforementioned methods cannot perform clustering or classification on the input data set, it is very difficult to locate or identify the values of all the objectives for a specific candidate, even after identifying one or two objectives of such candidate favorably meets certain design requirements, which is common in a series of 2-D plots with Pareto fronts. Moreover, the objectives of different design candidates become nearly indistinguishable in a parallel coordinate plot. As an attempt to tackle this problem in the machine design domain, this paper presents a case study that utilizes the self-organizing map (SOM) to visualize the design objectives of a high-speed switched reluctance machine. The results demonstrate that the SOM provides useful information with its intrinsic functionalities including data clustering, component-plane displays and data projections that are not offered by some conventional visualization techniques. Therefore, the SOM visualization can allow better integration of the knowledge and expertise of machine designers into specific electric machine design and optimization problems, and also assist them in the final decision-making process to choose the most appropriate designs.