Research On Fault Diagnosis Method Of Harmonic Reducer Based On Oversampling And Transfer Learning

Industrial robots are multi-joint mechanical devices that can be controlled automatically and are widely used in machining,welding,painting and other fields.Harmonic reducers are an important part of the mechanical structure of industrial robots and are responsible for the transfer of motion in industrial robots.Therefore,it is important to carry out fault diagnosis of harmonic reducers to improve the reliability and safety of industrial robots.In recent years,intelligent fault diagnosis methods have received a lot of attention from researchers because of their end-to-end fault diagnosis models.However,the prerequisite for good diagnostic results of intelligent fault diagnosis models is that a large amount of marker data is required for training.In practical engineering,equipment usually operates in a healthy state and it is difficult to obtain a large amount of data to meet the training requirements.Therefore,it is of great practical importance to conduct research on training intelligent fault diagnosis models with a small amount of fault data to achieve high diagnostic accuracy.This paper takes harmonic reducers as the research object and combines oversampling methods and transfer learning for fault diagnosis of harmonic reducers.The research focuses on the fault mechanism of harmonic reducers,the oversampling method and transfer learning to improve the diagnostic accuracy of the diagnostic model under data imbalance conditions.The main work carried out in this paper is as follows.1.A fault mechanism analysis of the harmonic reducer was carried out.Firstly,the structure and working principle of the harmonic reducer are analysed.Secondly,finite element simulation analysis is carried out on the flexible bearing,rigid wheel and flexible wheel of the harmonic reducer.The most common failure modes of the harmonic reducer are the fatigue fracture of the flexible wheel,the fatigue fracture of the outer ring of the flexible bearing and the fracture of the rigid wheel teeth.2.An adaptive synthetic minority class oversampling method(OTASMM)is proposed.Firstly,the characteristics of the original vibration signal in the time domain and time-frequency domain features are analysed,and the underlying logic of the existing oversampling methods is analysed at the algorithmic level for their insufficient ability to handle high-dimensional vibration signals.Secondly,an adaptive synthetic minority class oversampling method is proposed,which can generate high-dimensional vibration data more effectively than the existing oversampling methods.Finally,the validity of OTASMM is verified using a multilayer perceptron(MLP)as a diagnostic model with Case Western University bearing data.The validation results show that the data generated by OTASMM contains more valid features than existing oversampling methods at imbalance ratios of 300:10 and 300:20 respectively,while the quality of the generated data is more stable.3.A fault diagnosis method combining OTASMM and transfer learning is proposed.Firstly,the basic model of transfer learning,CBAM-CNN,is constructed based on convolutional neural network(CNN)and convolutional block attention module(CBAM).secondly,the synthetic data generated by OTASMM is used as the source domain data for transfer learning to diagnose machine faults,in order to address the problem that the source domain data of transfer learning is difficult to obtain.Finally,the Case Western Chu University bearing dataset was used to construct four datasets with different imbalance rates to verify the effectiveness of the proposed fault diagnosis method.The diagnostic results show that the proposed fault diagnosis method combining OTASMM and transfer learning has the highest diagnostic accuracy for different datasets compared to the control group.4.An experimental study of harmonic reducer fault diagnosis based on oversampling and transfer learning was carried out.Firstly,a harmonic reducer fault simulation test plan was developed and a harmonic reducer fault simulation test bench was built using the harmonic reducer fault modes derived from the simulation results as a guide.Then,the OTASMM and transfer learning-based fault diagnosis method was validated by constructing four data sets with different imbalance rates based on the collected raw vibration data.The validation results for diagnosing harmonic reducer faults under unbalanced data conditions show that the diagnostic method combining OTASMM and transfer learning has the highest diagnostic accuracy compared to the control group.