Study On Fault Diagnosis Of Shearer Rolling Based On Deep Transfer Learning

Nowadays,the degree of mechanization,informatization and smart of production facility of colliery in China is accelerating,but the casualty rate of coal mine electromechanical accidents still shows an upward trend.As a crucial mechanical equipment of colliery working face,the security and reliability of shearer is very crucial to come true safe and efficient coal exploiting.The thesis takes the rolling bearing,the pivotal component of shearer,as the study target,and uses the relevant methods of signal preprocessing technology,deep learning and migration learning,for purpose of realizing the early weak fault diagnosis of coalcutter antifriction bearing,which is in strong noise and varying duty circumstances,so as to make the intelligent fault diagnosis technology more suitable for engineering practice.The research carried out in this thesis is as follows:A denoise way based on EEMD is studied.Aiming at the many interference factors of the vibration signal when the shearer is working,using the EEMD to decomposes the original vibration signal to get a series of IMF components,and a comprehensive index is proposed to screen and reconstruct the IMF components,so that the reconstructed vibration signal has the characteristics of high correlation with the decomposed signal and obvious shock characteristics,so as to achieve the purpose of noise reduction.The denoised one-dimensional vibration signal is converted into a two-dimensional grayscale image as the input of the subsequent fault diagnosis model.The initial minor failure diagnosis of rolling bearing under single working condition based on deep learning is researched.The VGG16 fault diagnosis model is built,the difference between the output value of the model and the actual value of the label is measured by using the cross entropy loss function,the model arguments are imporved by using the Adam,and the techniques of batch normalization and random deactivation are selected to prevent the network from over fitting.Using the pre trained VGG16 fault diagnosis model to diagnose the early weak fault of rolling bearing under single working condition,the average diagnosis accuracy can reach about 97%,and the model parameters can be retained for the construction of fault diagnosis model based on deep migration learning later.The initial minor failure diagnosis of rolling bearing under changing conditions based on DTL is researched.Aiming at the problem that the diagnosis accuracy of the traditional model decreases under changing conditions,the pre trained VGG16 fault diagnosis model is used as the feature extractor and the model parameters are retained.The MK-MMD is introduced to quantitatively calculate the data distribution difference between the training set and the test set,and the fault diagnosis model based on deep migration learning is optimized together with the classification loss,a little bit of target domain data is used to fine tune the model parameters.Using the trained model to diagnose the early weak fault of rolling bearing under variable working conditions,the average diagnosis accuracy can reach about 94%,and the diagnosis accuracy of the model for five health states can reach more than 90%,which can effectively meet the needs of intelligent diagnosis of early weak fault of shearer rolling bearing in production practice.Through the comparative analysis with other models,the superiority of the method in this paper is further verified.At the same time,relevant suggestions are also put forward for the maintenance strategy of the shearer.