Research On Mechanical Fault Diagnosis Method Based On Multi-information Fusion

Information Fusion is an important algorithm in the information age,and it has a great space for development.This paper summarizes the application of information fusion method at home and abroad and its main research methods,and proposes a fusion algorithm based on the system sub-part transmission characteristics that can improve the diagnostic accuracy in mechanical faults.Taking the mechanical vibration signals of two different cylinders on the fuselage and the cylinder head of the diesel engine,the information fusion algorithm and its value are briefly expounded by comparing with the traditional single sensor.The result of theoretical analysis and experimental verification proves that the information fusion feature based on the transmission characteristics of the system can retain the fault information in a greater degree,and has higher compression resistance.Aiming at the problem of information redundancy in multi-source and homology information Fusion,the Granger causality relationship in economics is introduced to analyze the primary and secondary relationship between multiple signals.The Granger causality test between two sets of simultaneous sampling signals from time domain and frequency domain,and then the correlation coefficient is compared and analyzed,and the conclusion of the primary and secondary points between mechanical vibration signals is obtained.The results of the analysis of two methods show that the causal relationship between multiple signals is more evident in the time domain.For the characteristic vectors of specific mechanical faults,several more commonly used classifiers are enumerated: Support vector machine,BP neural Network and a new classifier: ultra-limit learner(ELM).In the experiment of mechanical fault diagnosis with 5 kinds of operating states of diesel engines,the characteristics of the information based on the transmission characteristics of the system are used as input parameters.By using PCA and sub-band averaging method as a reduction method,the classification accuracy of three classifiers in mechanical fault diagnosis of different dimensions is studied,and the results show that the average dimensionality reduction method of the experimental neutron band is helpful to improve the diagnosis accuracy.And the ELM of two different cylinders is more excellent than support vector machine and BP neural network.