Research On Fault Diagnosis Method Of Tractor And Its Key Components Based On Vibration And Noise Analysis

As a tool widely used in the field of agricultural production,the quality of tractor and its key components will directly affect the performance of the whole machine and the safety of operators.Therefore,there are a lot of studies about the failure of tractor and its key components.Tractor fault diagnosis is generally based on the experience of people,which has great requirements on the experience and sense of responsibility.Traditional tractor fault diagnosis has many problems,such as low efficiency and inaccurate judgment.Therefore,this paper proposes a tractor fault diagnosis method,which can not only finds out whether the tractor has a fault,but also finds out the location of the fault component and analyzes its specific fault type.At first,this paper studied the common methods of mechanical fault diagnosis,and analyzed the problems faced by the current research on the failure of the whole tractor.On this basis,this paper proposes a method to find out the fault of the tractor and its key components by vibration and noise signals.The main contents of the research are as follows:(1)The size of the tractor is large,so this paper use the non-contact measurement method to measure the noise of the tractor.The wavelet packet decomposition is used to decompose the tractor's noise signal into 8 frequency bands.The peak of noise,the wavelet packet energy entropy and the energy contribution rate of each frequency band are used to find out whether the tractor is faulty.After finding out the tractor is faulty,.comparing the marginal spectrum with the marginal spectrum of normal tractor to find out the frequency range where the noise changes greatly.Then comparing the frequency range with 1000 Hz and appropriately selecting the beamforming method and the method of vibration analysis to find out whether there is a fault in the component.If there is a fault,then analyzing the specific fault type and degree of the fault components.(2)For the fault components with simple structure,easy extraction of fault features,large number of data and easy installation of multiple accelerometers,a diagnosis method based on the multi-channel signal feature extraction and pattern recognition was studied.The process of time-domain feature parameters and frequency-domain feature parameters selection is analyzed based on CWRU bearing dataset.The optimization of PNN by using particle swarm optimization is also analyzed.After optimization,the model achieves recognition rate of 100% on 4 fault categories and 10 fault categories,and the recognition rate of 19 fault categories reaches higher than 98%.Compared with the single-channel vibration signal,the accuracy rate of multi-channel signal is increased by more than 10%.The recognition rate can reach 98% on the mixed data of multiple working conditions The result of the test verifies the accuracy and reliability of the fault diagnosis method based on multi-channel vibration signal feature extraction and optimized PNN.(3)In order to improve the transferability of the fault diagnosis method and meet the changeable needs of the diagnostic task in practical testing,a deep learning fault diagnosis method based on vibration waveform diagram and marginal spectrum was studied.The process of feature image extraction is also analyzed based on CWRU bearing dataset.Considering the fact that there are not enough training samples,transfer learning and a small number of target samples are used to train the pre-trained ResNet18.With a small dataset,the accuracy rate of bearing faults recognition is higher than 99%.The accuracy rate of bearing faults recognition is higher than 99% for the mixed data of multiple working conditions.This method can achieve high accuracy with a small number of accelerometers.And this method is also applied to the gearbox fault dataset of the University of Connecticut,the accuracy rate of fault diagnosis reaches 100% with a small dataset.It shows that the fault diagnosis method using time-domain and frequency-domain images of vibration signal,pre-trained convolutional neural network and transfer learning has high accuracy and adaptability.This method only requires a small dataset,and is suitable for changeable working conditions and changeable test tasks.The results show that the method proposed in this paper can identify the tractor's faults and find out the location of the fault components without prior knowledge.According to the actual test situation,the intelligent fault diagnosis method proposed in this paper can not only identify the type and degree of the fault,but also find out the specific location of the fault parts.With transfer learning,a convolutional neural network only needs a small dataset to transfer to a new diagnostic task quickly.With transfer learning,a fault diagnosis model suitable for the current diagnostic task can be quickly built.