Performance assessment of a novel fault diagnosis system based on support vector machines

Abstract

Fault diagnosis in chemical plants is reviewed and discussed, while an innovative data-based fault diagnosis system (FDS) approach is proposed. The use of support vector machines (SVM) is considered for their simpler design and implementation, and for allowing the better handling of complex and large data sets. In order to compare results with previously reported works, a standard case study such as the Tennessee Eastman (TE) process benchmark is considered. SVM achieves consistent and promising results. However, the difficulties arising when comparing SVM with previously reported results reveals the need for a systematic procedure for contrasting the performance of different FDS. Hence, general performance assessment indexes based on precision and recall of each FDS are proposed and used. In this sense, this study provides a data set and evaluation measures that could be used as a framework for future comparisons.