RNN-based Method for Fault Diagnosis of Grinding System

At present, the fault diagnosis of grinding system is evaluated by human being, which causes low efficiency, low accuracy, high cost and casualties easily. The traditional method has the unsatisfied performance on the classification ability of sample dataset with high dimension and temporal correlation. Aiming at the above problems, a RNN-LSTM based deep learning method is proposed in the paper, which realizes the intelligent fault diagnosis of grinding system. The dataset is batched for inputs of LSTM networks, and the temporal correlation of dataset is extracted, which is used to analyze the fault classification of feature vectors input of before and after time. We conduct the comparison of RNN-LSTM based networks and autoencoder based networks via simulation. It is concluded that the RNN-LSTM based method is obviously superior to the method with autoencoder for the dataset with high dimension and high temporal correlation, and the result of the error rate is less than 3%.