| With the continuous development of modern transportation,aviation,high-end precision mechanical equipment,instrumentation and other fields and the continuous improvement of the complexity of mechanical equipment,rolling bearings as an important part of the equipment,state monitoring of the running state of bearings is an important guarantee for the safe and normal operation of the whole system.As vulnerable parts,when the rolling bearing runs abnormally,mechanical equipment can not run or even damage.Therefore,the fault diagnosis of rolling bearings has been a hot research topic at home and abroad,and the research topic also has theoretical value and practical significance.This paper takes the improvement of variable mode decomposition as the research content,mainly analyzes the feature extraction and pattern recognition of rolling bearing fault.In rolling bearing fault diagnosis,since the decomposition effect of variable modal decomposition(VMD)is affected by its core parameters,the determination of core parameters requires a lot of professional experience,and the wrong setting of parameters will lead to unsatisfactory signal decomposition results.In this paper,we propose to optimize the VMD algorithm with particle swarm algorithm(PSO)to solve the problem that the core parameters need to be set artificially.The advantage of particle swarm algorithm is that it is easy and simple to implement,and does not have a lot of parameter settings and fast search speed.While PSO-VMD can find out the optimal solutions of core parameters,the particle swarm algorithm itself has the problems of insufficient global search ability in the early stage and easy to fall into local optimum in the later stage.To address these problems,a feature extraction method of Improved Particle Swarm(Improved PSO)Variable Modal Decomposition algorithm(IPSO-VMD)is proposed,which effectively makes up for the shortcomings of PSO algorithm and better balances the pre and post search ability of PSO algorithm,and greatly improves the overall search ability and convergence speed of PSO algorithm.The improved PSO algorithm optimizes the core parameter set of VMD,and the optimal parameter set is substituted into the VMD decomposition algorithm to decompose the bearing fault signal to obtain several components.The components with large correlation kurtosis values are selected as feature vectors to provide effective data for pattern recognition.Then the convolution is studied from the aspects of pattern recognition neural network(CNN),in order not to destroy the continuity of one-dimensional signal and coupling,to maximize the reduction of the input signal,the fault features compared with traditional convolution neural network,this paper according to the properties of one dimensional feature vector to construct a one-dimensional convolution kernel neural network(ODCNN).On this basis,the parameters of ODCNN model are optimized,and the changes of learning rate value are improved adaptively to maximize the advantages of convolutional neural network.Then the feature vectors processed before are divided into training set and test set,and the ODCNN model is trained with the training set,so as to predict and diagnose the test set.The simulation results show that the IPSO-VMD model converges faster than PSO-VMD and can successfully extract feature vectors,while the optimized ODCNN model is superior to the common CNN network in feature extraction and pattern recognition.Through theoretical and experimental verification,the IPSO-VMD-ODCNN model proposed in this paper can diagnose rolling bearing faults more accurately and efficiently.Finally,the GUI interface of rolling bearing fault diagnosis system is designed based on MATLAB platform,so that users can intuitively and conveniently complete the fault diagnosis process of rolling bearing.A series of fault diagnosis links such as bearing data loading,signal decomposition and pattern recognition can be completed through the human-computer interaction interface. |
PDF Full Text Request