Transformer-Based Unsupervised Cross-Sensor Domain Adaptation for Electromechanical Actuator Fault Diagnosis

There have been some successful attempts to develop data-driven fault diagnostic methods in recent years. A common assumption most studies is that the data of source and target domains are obtained from same sensor. Nevertheless, because electromechanical actuators may complex motion trajectories mechanical structures, it not always be possible acquire a particular sensor position. When locations changed, diagnosis problem becomes further complicated feature space significantly distorted. The literature on this subject relatively underdeveloped despite its critical importance. This paper introduces Transformer-based end-to-end cross-sensor domain method for overcome these obstacles. An enhanced Transformer model developed obtain domain-stable features at various locations. convolutional embedding also proposed improve model’s ability integrate local contextual information. Further, joint distribution discrepancy between two minimized by using Joint Maximum Mean Discrepancy. Finally, validated an actuator dataset. Twenty-four transfer tasks designed validate adaptation problems, covering all combinations three under different operating conditions. According results, outperforms comparative terms varying