Operational Modal Analysis And Fault Diagnosis Methods Based On ICA

Operational Model Analysis (OMA) of mechanical structure and Fault Diagnosis are themost essential and indispensable techniques in engineering application. As a statistical signaldata analysis tool, Independent Component Analysis (ICA) can effectively separate the blindsource signals from complicated signals of linear combination without prior knowledge. ICAhas been proposed as a new research method in the fields of monitoring and diagnosis. Thispaper applies ICA to gearbox operational modal parameter identification and fault diagnosis,finally combines Probabilistic Neural Network (PNN) algorithm to realize typical faultdiagnosis.Aiming at the major difficulties of OMA method based on vibration signals such as lownoise-robust ability, high subjective experience and poor accuracy in identifying parametersand more complex recognition method, the paper presents a new method of OMA on the basisof ICA. According to the analysis of the basic principle of ICA and OMA, the paper finds thatthe results of ICA calculation model and structure analysis model get the same trend, the ICAmodal parameter identification technology operates simply, and can effectively eliminatenoise interference and separate target vibration signal. In addition, the paper introduces IPI asa standard of component independence as well as falsity, the introduction of the standard ofIPI offers objective evidence to identify model.Since model characteristics change as the structural system modifies, the paper regardsFASTICA identified modal frequency as the diagnosis evidence, compares changes of anykind of model frequency on gearbox fault working condition, and achieves the goal of thefault identification. Finally, this paper introduces FASTICA analysis of characteristics offrequency domain analysis method, and applies the method to weak fault information enhancement and diagnosis of the complex gearbox. In the paper, test signals with differentrevolving speed under each working conditions are separated by FASTICA blind source inorder to analyze and extract characteristic parameters of each signal as diagnostic indexes,and then adopts the normalization signal characteristic parameters as samples of PNNalgorithm for diagnosis and analysis. The method makes full use of two algorithms, and theexperiment results show the fusion algorithm can significantly improve the accuracy of faultdiagnosis, reliability and diagnosis efficiency.