Research On Rolling Bearing Fault Diagnoise Based On Adaptive Variational Modal Decomposition And Improved Extreme Learning Machine

As common machine elements,rolling bearings perform an essential part in a wide variety of engineering processes.However,the occurrence of bearing faults may cause safety issues such as system breakdowns and economic loss.Therefore,implementing condition monitoring and diagnosis for rolling bearings is an important task,an effective fault diagnosis method is urgently needed.As vibration analysis contains extensive health information about machines and is always of great interest to researchers.Rolling bearings are taken as the research object in this paper,the mechanical structure of rolling bearings is described,several common types of rolling bearing faults are introduced according to their operational characteristics,analyzing the failure mechanism of rolling bearings,and fault diagnosis is carried out using neural networks by decomposing the vibration signals and extracting fault features.The main research content of the paper is as follows.Firstly,in the rolling bearing signal decomposition,in order to solve the disadvantages of using only time domain analysis or frequency domain analysis methods for fault diagnosis is not enough to accurately identify the fault,this paper proposes an Adaptive Variational Modal Decomposition(AVMD)method,using an Improved Sand Cat Swarm Algorithm(ISCSO)and using sample entropy as the fitness function,to optimise the Variational Modal Decomposition(VMD)method.By constructing a simulated signal in MATLAB and comparing AVMD with other commonly used signal decomposition methods,the simulation results show the effectiveness of the AVMD method proposed in this paper.Secondly,the input weights and biases are obtained randomly for the Extreme Learning Machine(ELM),and the classification accuracy needs to be improved.In order to enhance the ELM classification model,an Improved Whale Algorithm(IWOA)is proposed,using IWOA to find the optimal input weights and deviations for ELM.The whale optimization algorithm is introduced and improved by using a hybrid initialization population strategy,an improved hunting approach and a t-distribution-levy variation strategy to further expand genetic variance and the optimisation seeking performance.Simulation experiments illustrate the advantage of the upgraded algorithm and the presented model.Finally,selecting data from the bearing experimental platform of Case Western Reserve University,the vibration signal is decomposed using the AVMD method proposed in this paper,and the power spectral entropy,singular spectral entropy and energy entropy are selected as feature vectors,and the feature set is input into the IWOA-ELM model for training and testing.Experiments on different types and degrees of rolling bearing fault diagnosis prove that the model has a higher fault diagnosis recognition rate than other models.