Research On Fault Diagnosis Method Of Train Wheelset Bearing Based On Variational Mode Decomposition

The wheelset bearing is one of the core components of the bogie,which plays an indispensable role in load transfer and motion conversion.Wheelset bearings are subjected to a variety of alternating loads during service.Once local damage occurs on the surface of a component,the failure will gradually expand under the action of dynamic loads,which will lead to the degradation of the performance of the whole transmission device,and may even induce major accidents.Therefore,the research on fault diagnosis of wheelset bearings has undeniable practical application value.The service environment of wheelset bearings is complicated.Affected by factors such as wheel-rail noise and complex transfer path,the collected vibration signal often contains a variety of signal components,which makes it difficult to effectively separate the weak fault information of bearings in the early stage.To solve the problem that the fault feature information of single fault,compound fault,and multi-source fault of wheelset bearing is difficult to be accurately obtained,the fault diagnosis methods of wheelset bearing based on variational mode decomposition(VMD)are proposed.The main research contents are as follows:(1)The traditional parameter-adaptive VMD method can adaptively determine the key parameters of VMD by the powerful optimization ability of the meta-heuristic algorithm,but it also brings a new problem,namely,the setting of a fitness function.Considering that the Gini index,as a novel sparsity index,has excellent fault impact identification ability.Therefore,it is introduced into the parameter-adaptive VMD method,and an improved parameter-adaptive VMD method is proposed by taking the local maximum envelope Gini index as the fitness function of the particle swarm optimization algorithm.This method can accurately extract the weak single-point fault information of the wheelset bearing from the complex vibration signal.(2)In the parameter-adaptive VMD method,it is necessary to optimize the mode number K and select the optimal component,leading to the algorithm’s low computational efficiency.To solve this problem,an efficient adaptive single-mode VMD method is proposed.There is a U-shaped convergence phenomenon in the singlemode VMD algorithm,and the U-shaped region related to fault impact is very wide.Therefore,taking full advantage of the convergence characteristics of VMD,a new population position initialization strategy is established for the traditional particle swarm optimization algorithm.Then,taking correlated kurtosis as the objective function,the proposed adaptive single-mode VMD method is extended to the field of composite fault diagnosis.Through the analysis of simulation and experimental signals,it is proved that this method can efficiently extract specific fault information under strong interference.(3)Variational mode extraction(VME)is a new signal processing method derived from the theoretical framework of VMD.It can extract a single component from a multi-component signal,and the accuracy of fault information extraction is better than that of VMD.However,similar to VMD,the performance of VME is also limited by its initial parameters,including initial center frequency and penalty factor.To obtain these key parameters adaptively,the improved grey wolf optimizer with better global optimization performance is used to optimize VME.At the same time,considering that the fault feature coefficient has a clear directivity and can accurately evaluate the content of specific faults in the signal,the corresponding fitness function is established.Simulation and experimental analysis prove that the proposed method is suitable for both single fault diagnosis and compound fault diagnosis of wheelset bearings,and it is a high-precision fault information extraction method.(4)No matter calculating the correlated kurtosis or fault feature coefficient,it is necessary to input accurate fault characteristic frequency.If the rotating speed information deviation is large,the final diagnosis result will be seriously affected.Given this deficiency,and from the perspective of engineering application,a fault diagnosis method with high computational efficiency,good versatility,and,no prior knowledge is needed.Therefore,inspired by the convergence characteristics of VMD,the convergence characteristics of VME are studied deeply.According to the monotonicity of the iterative trajectory of the center frequency of the mode,the VME convergence tendency diagram is proposed.The number of potential sub-signals and their optimal initial center frequencies can be determined automatically by the VME convergence tendency diagram.Then,combining the VME convergence tendency diagram with the empirical formula of the penalty factor,a completely data-driven adaptive decomposition method is proposed.This method is suitable for the diagnosis of single fault and compound fault of wheelset bearings and the separation of multisource faults of bearing-tread and has a very fast computational speed.In this paper,the signal decomposition methods for adaptive extraction of bearing fault information are studied by taking the wheelset bearing under constant speed condition as the research object.By analyzing the vibration signal at the axle box,it can diagnose whether there is a fault in the wheelset-bearing system.The research results have a certain theoretical reference value for the research of bearing fault diagnosis and play a positive role in promoting the development of intelligent train operation and maintenance technology.