Research On Signal Feature Extraction Technology Of Non-stationary Rolling Bearing

In recent years,with the rapid development of science and technology in China,all walks of life in society are constantly expanding the scale of production,in order to meet the needs of enterprises,industrial production in the use of machinery and equipment gradually large-scale,sophisticated,intelligent and complex,and its failure rate is also increasing,so the device for real-time status monitoring and fault diagnosis has important practical significance.Rolling bearings as the most common but most easily damaged important parts of rotating machinery,its health will be directly related to the safety of machinery and even the entire industry.With the deepening of relevant research,the analysis and processing technology for smooth signal has been very mature,but in actual engineering,due to the influence of noise and other external factors,most of the vibration signals generated when a fault occurs are complex non-smooth time-varying signals,this paper takes rolling bearings as the research object,takes bearing fault diagnosis as the research objective,and carries out the relevant research work for the non-smooth rolling bearing signal feature extraction technology.Although the traditional time-frequency analysis method can decompose non-stationary signals from multiple angles,the algorithm has its own shortcomings that make it difficult to achieve the required accuracy.Therefore,this paper adopts the local feature scaling(LCD)method,which can adaptively decompose a complex non-stationary signal into a series of ISC components according to the characteristics of the signal itself,and then filter the decomposition results by the correlation coefficient criterion in order to facilitate the extraction of fault features.To address the feature extraction problem,this paper uses a combination of multidimensional features to extract the signal features.As screening the effective ISC components may result in the loss of some of the effective information,this paper extracts the time and frequency domain features of the original signal on the one hand,and the time and frequency domain energy features of the effective ISC components on the other hand,so as to ensure that the extracted feature information can more comprehensively characterise the fault state of the rolling bearing.This paper uses bearing data from Case Western Reserve University as data support,and the extracted multidimensional feature vector set is input into a support vector machine model(SVM)for fault identification,and experiments are conducted to compare the effectiveness of two optimisation algorithms,particle swarm algorithm(PSO)based on population intelligence optimisation and sine cosine algorithm(SCA)based on heuristic optimisation,on traditional support vector machines.The three models,SVM,PSO-SVM and SCA-SVM,are compared in terms of their respective accuracy and training time with the same data set input.The analysis of the experimental results shows that the SCA-SVM model has certain advantages in terms of fault identification accuracy and efficiency.