A Bearing Fault Diagnosis Method Based on One-Dimensional Convolutional Neural Network

Bearings, as key components of the spindle in computer numerical control (CNC) machines, directly affect the efficiency and stability of production lines. Conventional methods for diagnosing bearing faults are based on pattern recognition, and rely heavily on prior knowledge. In addition, finding a straightforward and universally applicable method for extracting features presents a significant challenge for conventional techniques. Fortunately, convolutional neural network (CNN) can adaptively process signals and extract deep-level features. In light of this, this study recommends a fault diagnosis model based on one-dimensional convolutional neural network (1DCNN). This model adaptively excavates deep features from initial vibration signals without relying on prior knowledge, thereby simplifying the feature engineering process and optimizing the precision and effectiveness of fault identification. First, we collect and preprocess vibration signals from the bearings to normalize the data. Second, by utilizing the 1DCNN model along with optimization techniques, the bearings are classified into distinct states: outer ring fault, cage fault, inner ring fault and normal operation. Finally, once trained, the 1DCNN model is integrated into a maintenance system for real-time monitoring and fault diagnosis. We conduct experiments on the XJTU-SY rolling bearing accelerated life test dataset which includes vibration data from bearings with various fault types as well as data from healthy bearings. Experimental results demonstrate that our 1DCNN model has significant advantages over traditional pattern recognition models in diagnostic precision and efficiency.