Fault Diagnosis Of Shearer Arm Drive System Based On Deep Transfer Learning

Electric traction shearer is the core equipment of fully mechanized coal mining face,which directly affects the safety and efficiency of underground production.With the construction of high mining and high efficiency mine and the mining of thin coal seam,it is urgent to monitor the working conditions of shearer and obtain the warning information of failure in time,so as to ensure the safety of production and predict maintenance.The traditional fault diagnosis method takes time and effort to extract fault features,and the typical fault data under shearer condition is few and the label data is insufficient,which makes fault diagnosis very difficult.Deep learning can automatically extract data features through unsupervised or semi-supervised feature learning algorithms and hierarchical feature extraction,but requires a large amount of data to understand potential data patterns.Transfer learning is used to solve the problem of insufficient training data,which has been widely concerned in recent years.Therefore,this thesis studies the deep transfer learning fault classification method,and transfers the fault diagnosis knowledge obtained after training the transmission fault data of the simulation platform to the fault diagnosis application of the shearer,which is of great significance to realize the equipment fault diagnosis under the small sample.This paper analyzes the fault mechanism and vibration characteristics of shearer rocker transmission system,determines different transfer learning strategies to realize fault classification of shearer rocker transmission system according to the known and unknown geometric parameters of mining equipment.The model recognition and classification method based on model transfer learning is proposed to solve the problem of fault diagnosis when the structural parameters of equipment transmission system are unknown.The network parameters are transferred to the shearer fault diagnosis model by constructing pre-trained deep convolution neural network mode.The thesis focuses on the problems of vibration signal interference,such as non-linearity and non-stationarity,when the coal mining machine is working in the coal mine.The thesis studies the noise reduction method based on singular value decomposition,and proposes an effective rank order determination method for singularity detection.After the singular value decomposition of the noise signal determines the effective rank order,the noise reduction signal is obtained after processing the singular value representing noise.By converting the one-dimensional original signal into a two-dimensional time-frequency distribution image,it serves as the input image data set of the model.Aiming at the problems of low typical fault data and insufficient labeled data under the condition of coal mining machines under coal mines,a fault diagnosis method of rocker arm transmission system based on deep transfer learning is proposed.Construct a pre-trained deep convolutional neural network model,transfer its network parameters to the shearer fault diagnosis model,lock the low-level network without participating in the update,fine-tune the high-level network parameters to achieve weight update and model optimization until the error is minimized,and obtain the transfer fault diagnosis model.By studying the effects of algorithms such as regularization,network sparsity optimization,and layer data distribution feature optimization on model training accuracy,the model optimization scheme is determined to reduce overfitting and improve generalization ability.And build a fault simulation experiment platform to verify the effectiveness of the algorithm.Aiming at the problem of complex fault data classification,a multi-label classification method based on deep transfer learning is proposed.The association and difference between a single fault diagnosis model and a composite fault diagnosis model are studied.The multi-label classification method is combined with the deep transfer learning method to improve the network structure of a single fault diagnosis model to realize the classification of composite fault data and verify the feasibility of the method.Finally,the vibration signal data of shearer’s rocker arm transmission system are collected for verification.The experimental results show that the proposed fault diagnosis method based on deep migration learning can realize the fault diagnosis knowledge transfer between different equipment monitoring data,and identify the shearer fault and obtain high diagnostic accuracy,which verifies the feasibility of the method.Compared with the traditional intelligent diagnosis method,the convergence speed is fast and the diagnostic accuracy is high.This method can realize high-precision equipment state recognition and classification based on laboratory data and a small number of application data samples.