Predicting configuration performance of modular product family using principal component analysis and support vector machine

Abstract

A novel configuration performance prediction approach with combination of principal component analysis (PCA) and support vector machine (SVM) was proposed. This method can estimate the performance parameter values of a newly configured product through soft computing technique instead of practical test experiments, which helps to evaluate whether or not the product variant can satisfy the customers’ individual requirements. The PCA technique was used to reduce and orthogonalize the module parameters that affect the product performance. Then, these extracted features were used as new input variables in SVM model to mine knowledge from the limited existing product data. The performance values of a newly configured product can be predicted by means of the trained SVM models. This PCA-SVM method can ensure that the performance prediction is executed rapidly and accurately, even under the small sample conditions. The applicability of the proposed method was verified on a family of plate electrostatic precipitators.