OBTAINING MULTIPLE PROCESS PARAMETER COMBINATIONS USING A SUPERVISED CLUSTERING-OPTIMIZATION APPROACH

Abstract

Though much work has been done in input-output mapping and input parameter optimization of different manufacturing processes, very sparse work is available in obtaining multiple optimal process parameter combinations of these processes i.e. multimodal optimization. Multimodal optimization is necessary as a single optimal process parameter combination may theoretically satisfy the objective function, but it might not be applicable in real life and may lead to unstable experimentation conditions. In this paper, a new approach based on supervised clustering is presented for multimodal optimization. In the presented approach, k-means clustering algorithm is first used to cluster selected initial points. Next, the optimization is performed in a small radius surrounding these cluster centers. New cluster centers are then formed from select points of the previous iteration and the optimization results. The proposed technique aims to find multiple optimal solutions by taking advantage of the fact there is a high density of points with favorable objective value clustered around a peak or a valley. The presented approach is then applied for multimodal optimization of some benchmark problems as well as a case study of input parameters optimization of electrochemical micro-machining. The results from the benchmark problems and the case study validate the effectiveness and robustness of the proposed approach.