Bearing Failure Classification with Metaheuristic Adjust of the Extreme Learning Machine Model

In this study, we propose the use of an Extreme Learning Machine model with a metaheuristic adjust of parameters, to recognize mechanical vibration data from bearings, in order to determine whether the device is in good condition or experiencing some type of malfunction. Our approach involved designing a supervised Extreme Learning Machine model whose parameters are adjusted by a Grey Wolf approach. We employed a dataset from Case Western Reserve University, available on their website. In this database the signals were gathered by sampling an accelerometer at 12 kS/s, placed radially on a bearing. The data was collected for four scenarios: one with vibrational signals from a healthy bearing, and three with bearings having induced faults in the inner race, outer race, and ball, respectively. This intelligent system was developed using Python in the Spyder environment, integrated with Pytorch framework, obtaining promising results. Keywords-Extreme Learning Machine, metaheuristics, bearing failures, artificial intelligence.