An Improved Convolutional Domain Adaptation Net work-Based Method for Motor Fault Diagnosis under Variable Operating Conditions

The accuracy of motor fault diagnosis is often limited due to varying operational conditions, which induce changes in fault characteristics and challenges in feature matching. To address this issue, this paper proposes an improved convolutional domain adaptation network for motor fault diagnosis under variable operating conditions. First, to overcome the insufficient feature extraction capability of conventional convolutional networks, a multi-attention convolutional network with multi-channel convolution is constructed to expand the receptive field and enhance discriminative feature learning. Second, based on Bayesian principles, joint distribution adaptation is refined to simultaneously align both marginal and conditional distributions between domains. Finally, the proposed method is validated by feeding both source and target do main data into the enhanced convolutional domain adaptation network for cross-condition fault diagnosis. Comparative experiments demonstrate that the proposed approach effectively extracts more discriminative fault features, while the improved joint distribution adaptation achieves superior feature alignment. The method achieves a fault diagnosis accuracy exceeding 98% under varying operational conditions, outperforming other state-of-the-art techniques and confirming its effectiveness.