Research On Intelligent Bearing Fault Diagnosis Methods Under Complex Working Conditions

As one of the key components in rotating machinery,bearings are prone to damage during operation due to long periods of high load,ultimately causing economic losses and safety problems.The normal operation of bearings can ensure the stable operation of the entire machinery,therefore,bearing fault diagnosis is of vital importance to the safe operation of machinery.However,the current fault intelligence fault diagnosis method is not only limited by the number of training samples,but also by the premise that the training and test samples need to have similar distribution.In actual production,it is usually difficult to obtain sufficient fault sample data and there are variable working conditions,resulting in the accuracy of the intelligent diagnosis method is limited.To address these problems,this paper proposes a deep learning intelligent diagnosis model based on Siamese networks and domain adversarial networks to solve the problem of diagnosing bearing faults under complex working conditions such as few-shot problem and variable working conditions,using bearings as the main research object.The main research contents are as follows:(1)Aiming at the problem of poor diagnostic performance of traditional fault intelligent diagnosis methods due to the unavailability of sufficient fault data in engineering practice,a few-shot diagnosis method based on self-attention metric learning with multi-scale feature fusion is proposed.The method takes Siamese networks as the main framework and uses the advantages of Siamese neural networks in few-shot data classification to solve the problem of insufficient fault data.A metric learning network which is a learnable nonlinear classifier is used instead of original Euclidean distance to flexibly and accurately calculate the similarity of input sample pairs.A multi-scale feature fusion module is added to the model to ensure the richness of the extracted feature information,and a self-attentive mechanism is added to improve the performance of the metric learning network.Two bearing fault experiment cases on the CWRU dataset,SQ dataset are conducted to validate the performance of proposed intelligent diagnosis model.(2)In view of the fact that the training and testing data are not guaranteed to have the same data distribution in engineering practice,which limiting the application of traditional fault intelligent diagnosis methods in practice,the proposed method is based on multi-level Bi LSTM domain adversarial adaptive fault intelligent diagnosis for variable working conditions.The method uses domain adversarial to reduce the feature differences under different operating conditions which consists of three parts: a feature extractor,a domain discriminator and a fault classifier.A multi-level Bi LSTM module is introduced in the feature extractor to take advantage of convolutional neural networks and recurrent neural networks to extract multi-level feature information,thus improving the fault identification accuracy of intelligent diagnostic methods under different operating conditions.Two bearing fault experiment cases on the CWRU dataset,PU dataset are conducted to validate the performance of proposed intelligent diagnosis model.(3)To further validate the effectiveness of the proposed method,experiments are carried out on a printing machine bearing fault test bench and a comprehensive bearing fault simulation test bench.Four different few-shot tasks were set up based on the collected printing machine bearing fault data,and the performance of the proposed few-shot diagnosis method was verified with different sample sizes.In combination with the bearing loads applied in the experiments,a total of six variable duty tasks were set to verify the performance of the proposed variable duty fault intelligent diagnosis method.Finally,based on the two proposed fault intelligent diagnosis methods,a bearing fault diagnosis software platform was developed using python software.