Deep Learning-Based Intelligent Fault Diagnosis of Motor Under Information Missing Conditions: A Review

The significance of intelligent fault diagnosis technology is increasing in ensuring the operational safety and reliability of high-precision equipment. In particular, the integrity of multi-sensor data flow is directly related to the real-time and accuracy of motor fault diagnosis. Deep learning techniques, represented by generative adversarial networks, provide a new path to solve this challenge by virtue of its powerful nonlinear modeling capability and end-to-end diagnostic properties. This paper systematically reviews advancements in deep learningdriven fault diagnosis for motors under information missing conditions over the past decade. A brief introduction to the motor system is given first to dissecting the typical types of failures and their corresponding root causes from the motor system level to the component level. Next, data reconstruction methods and hidden space inference strategies to reconstruct missing data are investigated. Further, deep learning methods for label-missing scenarios are reviewed, focusing on semi-supervised, weakly supervised, and self-supervised learning frameworks. Finally, focusing on information missing scenarios, this paper summarizes the core technical challenges facing this research area and provides potential avenues for future development.