Bearing Fault Diagnosis Based on an Advanced Method: ID-CNN-LSTM

In modern industry, ball bearings are not prone to failure, once a failure occurs, the production of the factory will be shut down, which will cause economic losses. Therefore, it's crucial to research how to diagnose ball bearings. This research proposed an advanced fault diagnosis method: 1D-CNN-LSTM to classify ball bearing faults and use the ball bearing faults data from Case Western Reserve University (CWRU) to execute experiments, which is the raw one-dimensional vibration sequential data. In the experiment, the raw vibration data is first split into multiple subsequences, and input to one-dimensional convolutional neural network (1D-CNN) wrapped by TimeDistributed layer to extract features. The output of 1D-CNN is a sequence, which is input to long short-term memory (LSTM) for sequential processing. Finally, the class of bearing fault is output for diagnosis. The results indicate a good model fit and outstanding generalization and robustness on new validation data. The assessment of the training dataset indicates that it has achieved a perfect accuracy of 100%, while the validation dataset has achieved an accuracy of 99.99%, which is an exceptional outcome.