Optimized-ELM Based on Geometric Mean Optimizer for Bearing Fault Diagnosis

Ensuring smooth machine operation and safety is crucial in most engineering plant, and fault diagnosis plays a critical role in achieving these goals. In recent years, machine learning techniques already being utilized extensively in the research of bearing fault diagnosis. One of the recent methods that has gained popularity is the extreme learning machine (ELM) method. This method offers several advantages, such as fast learning rate, generalization performance efficient, and easy to use. However, it is important to note that the ELM method can result in inaccurate diagnosis, if the values for input weight, hidden layer bias, and number of neurons are not selected properly. This paper introduces a new approach for bearing fault diagnosis, named GMO-ELM, which utilizes the ELM method and the geometric mean optimizer (GMO) to optimize ELM parameters. The proposed method was tested using sets of bearing vibration signal from Case Western Reserve University (CWRU) with four different operating conditions, including healthy baseline, outer race fault signal, inner race fault signal, and ball fault signal. Based on the result, the proposed method is able to provide 12% better performance by comparing to the conventional ELM and competitive diagnosis performance by comparing to other recent diagnosis model.