I. Introduction
Modern equipment is becoming more and more complex, precise, intelligent, and automated [1]. As a critical component of modern equipment, rotary machinery always works under harsh environment and varying duty, which leads to a great risk of failures [2]. The sudden breakdown of rotary machinery may cause enormous economic loss or even severe safety accident. Therefore, it is crucial to develop rotary machinery fault diagnosis technique for the operating reliability and the safety of equipment [3], [4]. To effectively diagnose the failure mode of rotary machinery, scholars have conducted many researches. The mainstream fault diagnosis methods can be roughly separated into four categories: signal processing-based method [5], [6], feature extraction-based method [7]–[9], traditional machine learning method [10]–[12], and deep learning method [13], [14]. Among these fault diagnosis methods, deep learning-based intelligent fault diagnosis methods have shown outstanding performance in recent years.