Research On Fault Diagnosis Of Train Rolling Bearing Based On IGA-BP And H-ELM

In recent years,along with the arrival of ?Made in China 2025?,China is in the construction of "13th five-year plan" and "One Belt And One Road".The popularization of high-speed trains and the development of technology have become ever more mature.Increasing the speed of operation and ensuring safety has become the primary factors.Rolling bearings in trains play a critical role in mechanical fault components.Its operating conditions directly affect the overall performance of trains.At present,China's train construction plan appears to be "heavy construction,light management",and the cost of service is high.The accuracy and efficiency of fault diagnosis of rolling bearings in trains are relatively low.Therefore,detection of bearing fault conditions and the research of diagnostic methods are of great significance.This paper takes the rolling bearing in the train as the research object,and studies the fundamental factors of the diagnostic accuracy and time efficiency.At present,among the various fault diagnosis methods,the traditional feature extraction efficiency is low,and the non-stationary signal processing is not good.The BP neural network and genetic algorithm(GA)in the classification method have been proved to have good diagnostic results,but there are still defects in the processing of large amounts of data:(1)the model parameters are difficult to determine;(2)the training time during diagnosis is relatively Long;(3)easy to fall into local minimum points;(4)there is no universal feature extraction method.Extreme Learning Machine(ELM)has a high training speed,but directly processes the original signal for fault diagnosis.The final experimental classification has poor results and must be extracted.In this paper,bearing fault detection is introduced,bearing fault detection,feature extraction,fault status classification and other aspects of bearing fault diagnosis is introduced in detail.Research on its corresponding method was carried out.The main tasks are as follows:(1)Based on wavelet packet decomposition.The feature extraction method for vibration signal of rolling bearing is proposed.For the traditional Fourier transform processing vibration non-stationary signal effect is poor,wavelet decomposition of the high-frequency part of the signal is not subdivided,the proposed wavelet packet decomposition,the first layer of wavelet packet decomposition of the original vibration signal layers of high and low frequency information;The wavelet packet coefficients are reconstructed and the signals of each frequency band are extracted.The energy characteristics of the original fault signal are finally obtained.The experimental results demonstrate that the method can well characterize the characteristics.(2)A fault diagnosis method for rolling bearing based on improved GA optimization BP is proposed.For the eigenvalues of wavelet packet decomposition,it is not intuitive to directly judge the bearing fault classification effect,which needs to be supervised and classified.Firstly,the classification is introduced using the traditional BP neural network and LM-BP;then the genetic algorithm of the optimization method of LM-BP is introduced,which is introduced in detail,and the selection operation of the roulette game based on GA itself has certain randomness.To ensure the diversity of the population and ultimately affect the global optimization performance,the selection operation of the genetic algorithm is improved.The final experimental results show that the improved genetic algorithm(IGA)optimization LM-BP neural network accuracy is higher.(3)Based on H-ELM rolling bearing fault diagnosis method.At present,feature extraction is required before classification for various types of diagnosis,and the GA-BP optimization time is longer.The layered over-limit learning machine(H-ELM)is proposed to process the original vibration signal of the bearing and obtain the final classification result.The first is unsupervised multi-layer coding,which uses feature mapping of an ultra-limit learning machine automatic coder(ELM-AE)and introduces the L1 norm to make the learned features more sparse;secondly,multi-layered thinking is more able to map the original representations.The low-dimensional characteristics of the signal;the ultimate are the classification of supervised features.(4)Conduct an experimental comparison.For the method proposed in this paper,the experimental data of the United States Case Western Reserve University was compared.According to the accuracy and time efficiency of the comparison between the two,the final experimental results show that H-ELM is not only more accurate than IGA-BP,and from the time efficiency point of view,using H-ELM method classification results better time,based on H-ELM's train rolling bearing fault diagnosis can be applied and promoted in real life.