Fault detection and optimal feature selection in automobile motor-head machining process

Abstract

This work presents a comparison among four methodologies for fault detection and feature selection applied to dimensional quality control in an automobile motor-head machining process. Three of them are based on the Mahalanobis Taguchi System (MTS) while the fourth one uses Support Vector Data Description (SVDD) one-class classification in order to build an hypersphere with the minimum volume with an enclosed boundary containing almost all target objects. Moreover, Gompertz binary particle swarm optimization (GBPSO) algorithm is applied to optimize kernel hyperparameters in SVDD and simultaneously solve the feature selection problem.