Rolling Bearing Fault Diagnosis Method of Acoustic and Vibration Multi-Level Fusion with Attention-1DCNN

Bearing faults are one of the most common causes of machine failures. However, due to the diverse nature of fault characteristics, a single sensor is insufficient for accurately identifying these features. Multi-source signals contain more comprehensive fault feature information. In order to effectively integrate multi-sensor signals and improve fault recognition accuracy, this paper proposes a rolling bearing fault diagnosis method based on acoustic-vibration multi-level fusion and Attention-1DCNN. Firstly, a single Attention-1DCNN module is employed to extract self-features from both the acoustic emission signal and vibration signal separately. Subsequently, complementary features are extracted from the fused acoustic emission and vibration signals using a feature mapping stacking method for fault identification. Finally, through softmax regression analysis, the fused features are classified into specific faults. The effectiveness of the proposed method was validated using an experimentally constructed dataset. The experimental results demonstrate that under various fault modes such as outer ring crack, inner ring crack, and rolling body pitting, the proposed method achieved 100% recognition accuracy. Compared with single-sensor or single-level fusion diagnostic methods, this approach exhibits higher diagnostic accuracy and improved robustness.