Bearing fault diagnosis using extreme learning machine based on artificial gorilla troops optimizer

Abstract

Bearing diagnosis is important to ensure smooth machinery operation and safety. Machine learning methods have been used widely in bearing diagnosis study, and one of the recent methods used is an extreme learning machine (ELM). The ELM method offers ease of implementation, rapid learning rate, and better generalization performance. However, the ELM method may lead to inaccurate diagnosis due to inappropriate value selection for neuron number, input weight, and hidden layer bias. Therefore, this paper proposed a new bearing diagnosis using ELM-based gorilla troops optimizer (GTO) method, named as GTO-ELM. The GTO method was used to select an optimized parameter for the ELM method. Two sets of bearing datasets from experimental work and online database were used in this study for evaluation on the proposed method. Both datasets have four different type of operation condition which are normal (baseline), inner race fault, outer race fault, and ball fault. Based on diagnosis result, the proposed GTO-ELM was able to surpass whale optimization algorithm (WOA)–ELM in term of convergence speed and conventional ELM methods in term of diagnosis performance with almost 10–12% better performance.