Optimized Class Decomposition for Fault Detection

Abstract

The paper proposes an innovative approach in solving the fault detection problem of sewerage treatment plant machinery. The proposed approach treats the fault detection data with the class decomposition problem, ensuring that a classification algorithm overlooks no disjunct instances. As the class decomposition technique requires heavy customization to each class of instances in every data set, Grey Wolf Optimizer is used to determine the appropriate clustering method with the appropriate setting for each class of instances. The proposed approach is tested on real-life sensor data from a sewerage treatment plant, and the results show that here proposed approach overshadows several manually proposed class decomposition methods.