Fault detection based on squirrel search algorithm and support vector data description for industrial processes

This paper proposes a novel fault detection system by the combination of Support Vector Data Description and Squirrel Search Algorithm. This approach is capable to deal with processes or machines where the number of fault observations is small or not even available for training phase. In this work the use of classic Support Vector Data Description as well as its fast version with two kernel functions is proposed. The experimental results showed that the proposed system exhibits suitable capabilities for fault detection in complex industrial processes such as the one presented in this research. Moreover, a nonparametric statistical analysis is also included in order to compare the considered strategies and enhance the efficiency of the presented approach. Finally, a comparison with genetic algorithm approach and the one-class classifier based on support vectors is carried out which shows that the proposed algorithm outperforms traditional techniques.