Research On Machinery Fault Diagnosis Based On Vibration Signal Feature Extraction And Machine Learning

As human society enters the industrial revolution,rotating machinery has spread all over society and injected new vitality into the vigorous development of various industries.Rolling bearing is an indispensable component for rotating machinery,and its reliability greatly affects the normal operation of the whole device.Therefore,taking rolling bearings as the object of study,using existing technologies and means to quickly and accurately diagnose the operation status of the mechanical equipment has certain engineering and use value.As a common fault diagnosis method,vibration signal analysis method has been widely used due to its high practical value and reliable diagnosis results.In this paper,the fault diagnosis of rolling bearings in rotating machinery is realized by extracting fault features from vibration signals and using machine learning methods.The primary work of this paper is as follows:(1)Aiming at the parameter selection problem of variational mode decomposition(VMD),an adaptive method of VMD parameters based on sparrow search algorithm(SSA),namely SSA-VMD,is proposed.This method takes the minimum mean envelope entropy as the objective function to optimize the parameters.Based on SSA-VMD and support vector machine(SVM),a rolling bearing failure method is proposed(SSA-VMD-SVM).Firstly,the parameters of VMD are optimized by SSA,and the original signal is decomposed by the optimized VMD to obtain several modal components.Then,the fuzzy entropy of each mode is calculated separately,and the obtained results are used as characteristic values.Finally,the obtained eigenvalues are input into SSA-SVM as eigenvectors to classify the samples.Two published data sets are used to verify the feasibility of SSA-VMD-SVM,and compared with the same type of methods,SSA-VMD-SVM shows certain advantages in terms of running speed and accuracy.(2)In view of the problem of converting one-dimensional time series into two-dimensional images,Markov transition field theory(MTF)is introduced to realize the conversion between data.A fault diagnosis method is proposed based on MTF and convolutional neural network(CNN),namely MTF-CNN.This method uses quantiles to discretize the time series,so that each element in the series has its corresponding quantile bin.The corresponding Markov matrix is obtained through the Markov chain,and the time axis is introduced into the Markov matrix to obtain the Markov transform field.The MTF image can be obtained by taking the elements in the matrix as pixels.By optimizing the structure of model and the number of neurons,a CNN model suitable for rolling bearing fault diagnosis is proposed.By converting different types of original vibration signals into corresponding MTF images,CNN model is used to classify the converted images.The MTF-CNN is verified through the data set,and it can be seen from the result of experiment that this method can classify faults well to a certain extent.In addition,MTF-CNN has certain advantages compared with the same type of methods.