A multi-source frequency temporal associative network for servo motor bearing fault diagnosis under dynamic operating conditions

With the widespread application of industrial robots in smart manufacturing and industrial automation, the topic of fault diagnosis for key components of robots is gaining increasing attention. Servo drivers not only provide power and control for industrial robots, enabling them to perform complex tasks, but their rich built-in signal sources could also be adopted for fault diagnosis. This paper proposes a fault diagnosis method that integrates vibration and electrical signal information, aiming to detect faults in servo motor bearings by fully exploring the potential of built-in signals from servo drivers. By designing time-frequency features for efficiently fusing multiple physical signals, the proposed multi-source frequency temporal associative (MSFTA) network enhances the perception and collaborative processing capabilities of multidimensional information. The focus is on extracting deep features from different physical dimensions and mechanisms (such as torque, speed, current, and vibration signals) and constructing coupling and complementary mechanisms for joint analysis of time-frequency information from multiple sources. Finally, experimental results under varying speeds and load conditions demonstrate the effectiveness of the proposed method, showing promising application prospects in the field of industrial robotics.