Partial Demagnetization Fault Diagnosis of Direct-Drive Drilling Motor Using Image Deep Learning

Abstract

To enhance both the accuracy and the efficiency in diagnosing partial demagnetization (PD) faults within direct-drive drilling motor, an intelligent diagnosis method that leverages bispectral imagery in conjunction with a residual network architecture, augmented by an adaptive hybrid attention module (HAM) is introduced. First, the sensitivity of motor signals to PD faults of varying degrees is analyzed, with the torque signal being identified as the primary indicator for fault detection. Then, bispectral analysis is employed to transform the original signals into visual representations as the input of the diagnosis model. Subsequently, an HAM is constructed and integrated into the residual network framework to improve the accuracy of demagnetization fault detection. Finally, a demagnetization prototype of motor is developed, and a torque measurement platform is established for conducting the experiment. The efficacy and advantage of the proposed method are substantiated through comparisons with other prevalent techniques. The results indicate that the proposed method attains an accuracy of 97.66% in recognizing various severity levels of demagnetization faults, demonstrating robustness against varying noise.