Research On Fault Diagnosis Method For Rolling Bearing Under Variable Operating Conditions Based On Deep Learning Model

Rolling bearings,as one of the core components of rotating machinery and equipment,are prone to failures during long-time high-speed operation,which can affect the performance of the whole equipment and cause serious safety accidents.The traditional fault diagnosis methods mainly rely on manual feature extraction and single method,which is difficult to cope with the demand of complex,intelligent and integrated scenarios.However,deep learning has powerful adaptive feature extraction and big data processing ability,which can realize intelligent fault diagnosis under complex variable working conditions and avoid the subjective influence of expert experience.Therefore,this thesis takes rolling bearing in mechanical equipment as the research object,takes vibration signal processing method and deep learning algorithm as the main technical route,and proposes the fault diagnosis method of rolling bearing in variable working condition based on deep learning model joint signal processing,the main research contents of this thesis are as follows:(1)Aiming at the traditional rolling bearing fault diagnosis method with single feature extraction and troublesome parameter selection of pattern recognition model,the model is easy to fall into local optimization,etc.A rolling bearing fault diagnosis method based on improved convolutional neural network with joint parameter optimization SVM model is proposed.The method uses the improved CNN to automatically extract fault features and optimize the network structure and parameters,while introducing batch normalization behind the convolutional layer to simplify the process of adjusting parameters and accelerate the model learning speed;finally,the particle swarm algorithm is used to optimize the parameters of the SVM to achieve fault diagnosis.The method is verified to have good diagnostic effect through the University of Cincinnati bearing dataset and the comparison experiments of different methods.(2)In view of the problems of low diagnostic accuracy caused by the unbalanced distribution of fault data and limited samples,the method of rolling bearing fault diagnosis using the whale algorithm optimized variational modal decomposition combined with a one-dimensional convolutional network and a mixed model of long and short-term memory network is proposed.The method first uses the whale algorithm to find the optimal number of modal decomposition and penalty factor parameters for the variational modal decomposition;then uses the optimized variational modal decomposition to pre-process the bearing vibration data to achieve noise reduction and fault sample feature enhancement;finally directly inputs the one-dimensional convolutional neural network and the long and short-term memory network "end-to-end" model for rolling bearing multi-fault mode identification.Finally,the rolling bearing multi-fault mode recognition is performed by directly inputting the 1D convolutional neural network and long and short-term memory network models.The experiments use Case Western Reserve University bearing data set to verify the high diagnostic efficiency of this paper’s method under different working conditions.(3)Aiming at the characteristics of rolling bearing fault signal with nonlinearity and non-smoothness,which mainly rely on manual extraction of feature information,leading to the problems of insufficient feature extraction and lack of self-adaptation,a rolling bearing fault diagnosis method based on continuous wavelet transform and convolutional gated recurrent neural network is proposed.The method firstly uses the sliding window technique to effectively expand the data samples for the original bearing vibration data,then performs continuous wavelet transform on all samples to generate a new data set,uses convolutional gated recurrent neural network to train the sample data,and uses each layer of the network to extract feature information adaptively to achieve multi-fault diagnosis of rolling bearings.Experiments are conducted using Case Western Reserve University and Southeast University datasets to verify the effectiveness of the methods in this thesis.(4)A two-way parallel multi-scale improved residual neural network approach is proposed for rolling bearings under the influence of strong environmental noise interference,complex variable operating conditions and other factors with poor diagnosis,and reduced anti-interference and generalization.The method designs a multi-scale residual Inception module,which can effectively extract feature information,while adding an attention mechanism to solve the mutability and variability of the data,and in addition uses multiple null convolutional residual blocks to expand the perceptual field,which helps to extract more feature information and achieve accurate fault diagnosis.By comparing with other mainstream methods,it is verified that the method has better superiority.