A bearing fault diagnosis method based on MBCNN and Transformer encoder with DANN

Traditional data-driven approaches employed in the field of bearing fault diagnosis encounter difficulties in cross-working condition scenarios. Deep learning-based methods often demonstrate overfitting tendencies to training data, limiting generalization to new working conditions. To address this issue, we integrate domain adaptation techniques into deep learning-based methods. Specifically, we employ an adversarial learning-based approach known as DANN. We combine MBCNN and Transformer encoder as the deep learning-based method. Consequently, we propose a novel fault diagnosis method based on MBCNN and Transformer encoder with DANN. Additionally, we utilize STFT, a signal processing technique, to transform raw signals into 2D images. We demonstrate the effectiveness of the proposed method using a bearing dataset. A series of experiments across various working conditions validate its excellent diagnosis performance. The results indicate that the proposed method exhibits exceptional capability and robust performance in diagnosing bearing faults across different working conditions.