Imbalanced Multidomain Generalization Method for Electromechanical Actuator Fault Diagnosis Under Variable Working Conditions

Electromechanical actuators (EMAs) play a critical role in the more/all electric aircraft, which is considered the next-generation aircraft. Due to the complexity and variability of the EMA working environment, along with the scarcity of fault data, the monitoring data of EMA exhibit in-domain data imbalance and divergent label distributions across domains. To address this issue, a novel imbalanced multidomain generalization method is here developed for EMA fault diagnosis. Specifically, a new loss function BoDA introduced to achieve the out-of-distribution generalization through aligning and calibrating across imbalanced multidomain data during training. The Lion optimizer is used to ensure the training loss converges to optimality while maintaining a minimal memory footprint. In addition, a two-stage training method is used to improve the classification performance of the proposed method under imbalanced data distributions. Experimental results demonstrate that the proposed method achieves superior diagnostic performance than several state-of-the-art methods on imbalanced single/multidomain EMA datasets.