Monitoring Screw Fastening Process: an Application of SVM Classification

Abstract

In industrial manufacturing, costs of production can be reduced by detecting the failures of running process. Automated systems are widely applied for detecting the failures instead of human inspection. This research develops a SVM classifier, which can be integrated in an automated system, for detecting the faults of screw fastening on the cover of hard disk. The data of screw fastening can be distinguished in 2 patterns; complete and incomplete fastening. These data are necessary to be pre-processed for developing the classifier, which 4 methods of the pre-processing data are proposed. The SVM classifier is also investigated through three popular kernel functions with the proper values of their corresponding parameters, and the results in terms of missed class percentage are compared. The experiments show that SVM performs as an efficient classifier for all pre-processing data methods and the proposed kernel functions. The SVM classifier using RBF kernel with Data 4 has the best results; therefore, it is strongly recommended for the implementation of self-automated screw monitoring machine.