Kernel Extreme Learning Machine bearing fault diagnosis based on empirical wavelet transform feature extraction and Sparrow search algorithm optimization

Bearings are installed on mechanical equipment, and the environment of mechanical equipment is usually very noisy, and the vibration signal propagation path is very complex, so that the signal is always not pure enough, it is difficult to extract useful information, and the identification of faults brings some difficulties. It is proposed to use empirical wavelet transform to decompose fault signals, extract kurtosis as fault features, and input the established fault feature vector into a SSA-KELM for training and testing. According to experimental evidence, the method proposed in this article can quickly and better distinguish the fault state of the bearing at this time, which can bring many conveniences to our production and life. Compared with unoptimized kernel extreme learning machines, it has fast convergence speed and strong optimization ability.

I. INTRODUCTION

Nowadays, mechanical equipment tends to develop towards high intelligence, automation, and precision. The health status of mechanical equipment directly affects the safety of production activities and the quality of products produced. Rolling bearings, as the most crucial part of mechanical equipment, cannot operate safely without the normal operation of bearings. Therefore, in the actual production work, determine where the bearing is wrong and what kind of problem occurs can make the mechanical equipment run better, safer and more stable, and can better serve the production.