Research On Fault Diagnosis Of Motor Bearing Based On Attention Mechanism

Rolling bearings are an important part of the motor's operation process,and the operating conditions are complex and diversity and the service environment is harsh,causing the bearings to be prone to failure,thereby reducing the reliability of the entire motor system.Traditional bearing fault diagnosis methods mostly use time-domain or frequency-domain signal processing methods to extract fault characteristic information from vibration signals.However,bearing fault vibration signals often exhibit characteristics such as strong noise,nonlinearity,and non-stationarity,resulting in fault characteristic information.Difficult to extract.Data-driven machine learning algorithms such as support vector machines,random forests,etc.have been applied to fault diagnosis,but machine learning algorithms are used in conjunction with other signal processing algorithms.They are only used as a classifier in the last step and cannot directly use the original signal.The end-to-end model will cause the loss of information in signal processing,which has significant limitations.Therefore,in response to the above problems,this paper has carried out the research on the fault diagnosis of motor bearings based on the deep learning technology of the attention mechanism.The specific research contents are as follows:(1)A Seq2 Seq encoder-decoder model based on attention mechanism is proposed for fault diagnosis.The Seq2 Seq model is a model that specializes in processing time series,in which the encoder adopts a bidirectional long short-term memory neural network(Bi LSTM).This structure can effectively avoid the phenomenon of "gradient vanishing".Traditional encoder-decoder may cause information loss when encoding long input sequences,making it difficult for the decoder to obtain valid data information.Therefore,in response to this problem,we introduce an attention mechanism to enhance its local feature learning ability and effectively retain context information.For this model,the data set of Case Western Reserve University was used to train and test the model.The recognition accuracy of various faults on the test set reached 98.38%,95.44%,93.25%,and 94.21%,which proved the model's performance.(2)Aiming at the problem that the Seq2 Seq encoder-decoder cannot process the sequence in parallel,the Transformer model is proposed for fault signal processing.The Seq2 Seq model based on the attention mechanism is composed of Bi LSTM structure in its codec,which determines that the Seq2 Seq model can only process information in chronological order,which greatly reduces the computational efficiency.In this paper,the Transformer model can be used to process data in parallel and accelerate the calculation.At the same time,a large number of attention mechanisms are used to promote the understanding of time series and feature extraction,and further improve the model's recognition accuracy of various faults.The results of this research prove the effectiveness of the attention mechanism for fault diagnosis,provide ideas for the intelligent diagnosis of motor faults,and lay the foundation for further intelligent motor fault diagnosis in the future.