A Hybrid Fault Diagnosis Method Based on Transformer and CNN

The elevator data of the welding production line has significant timing characteristics. There are more normal data and less abnormal data. Therefore, it becomes more complicated to diagnose the abnormality of the welding production line hoisting system through data analysis. This paper proposes a hybrid fault diagnosis method that combines a Transformer with a convolutional neural network. Firstly, the Transformer is used to extract the long-range dependencies in the data and identify complex faults that may not be adjacent. Then, the convolutional neural network extracted local or spatial features from the data. Finally, the multilayer perceptron maps the extracted features to the output space. The model innovatively merges the Transformer's global context analysis with the convolutional neural network's local feature detection precision, boosting fault diagnosis accuracy and efficiency. The proposed hybrid model was applied to the actual vibration data set of the hoists condition monitoring system of the white welding production line, and the results were contrasted with those of the support vector machine method, the long short-term memory network method, and the back propagation neural network method. Experimental findings demonstrate that the hybrid model achieves an accuracy of 99.98%, which is suitable for the fault diagnosis of the driving end bearing of the body of the white welding production line.