Support Vector Machines in Fault Tolerance Control

Abstract

This paper presents a new approach for quality monitoring of on‐line molded parts in the context of an injection molding problem using Support Vector Machines (SVMs). While the main goal in the industrial framework is to automatically calculate the setpoints, a less important task is to classify plastic molded parts defects efficiently in order to assess multiple quality characteristics. The paper presents a comparison of the performance assessment of SVMs and RBF neural networks as part quality monitoring tools by analyzing complete data patterns. Results show that the classification model using SVMs presents slightly better performance than RBF neural networks mainly due to the superior generalization of the SVMs in high‐dimensional spaces. Particularly, when RBF kernels are used, the accuracy of the task increases thus leading to smaller error rates. Besides, the optimization method is a constrained quadratic programming, which is a well studied and understood mathematical programming technique.