Research On Constrained KICA Signal Separation Method And Its Application In Fault Diagnosis Of Roling Bearings

The bearing is crucial parts of the mechanical system, and its performancedirectly affects the health of the entire mechanical system. However, due to thepresence of the Rolling processing complex, poor working conditions and otherfactors, led to rolling bearings have a high failure rate. So bearing fault diagnosis isan important research topic. The measured vibration signal of rolling bearing is amixture of multiple source signals and often exhibit non-stationary and nonlinearproperties. This is seriously affects the accuracy and reliability of fault diagnosis.Independent component analysis (ICA) is based on high-level statisticalcharacteristics of signal and which is developed in recent years. ICA can separates theindependent source signals from a linear mixture signals of a plurality of sourcesignal, but it also exists the limitations for dealing with nonlinear signal. This articleuses the kernel independent component analysis (KICA) which is a new ICA methodaccompanied by the development of the kernel method and the product of thecombination of independent component analysis and kernel methods. It caneffectively capture the nonlinear characteristics of the signal and show a uniqueadvantage in the treatment of high order correlation data. Despite the introduction ofnuclear methods makes KICA have a good effect in the treatment of many nonlinearproblems of fault diagnosis, but it also brings a kernel function and kernel parameterselection problem. Kernel function and kernel parameter selection has a significantimpact on the analysis results, and there is lack of an effective means guide the choice ofparameter and kernel function. So the kernel function and kernel parameter isdetermined in the practical application only by the experience of the user. In addition,the kernel independent component analysis signal existing the uncertain, theestimation of source and used very little signal information problem. Aiming at theabove problems, this paper carried out the following work:（1）Firstly, summarizes the typical fault mechanism of rolling bearing which isthe core component of rotating machinery. Secondly, the paper carried out thevibration signal acquisition experiment of fault rolling bearing. The third discussesthe ICA, nuclear method and kernel independent component of ball bearing faultdiagnosis in order to the theoretical basis for subsequent research.（2）In order to effectively solve the problem about the determination of kernelfunction and kernel parameters of KICA in mechanical fault diagnosis, the similarity function is constructed based on improved vector cosine values. By using thesimilarity between the vectors before and after the KICA as KICA signal separationcriteria and reflect the performance of kernel functions and parameters of KICA inrolling bearing fault diagnosis. The purpose is to use the simulation conclusions toguide the choice of KICA parameters in the actual diagnosis.（3）According to the separation of KICA signal, the uncertainty in the sourcesignal estimation and signal information using a single problem put forward the faultdiagnosis model of rolling bearing based on Kernel Independent Component Analysisof constraint，aim to construct the constrained KICA signal separation using the faultdiagnosis of a priori knowledge of the reference signal and to improve the algorithmof signal separation. Researched effect of the reference signal channel number andreference signal parameters on the performance of the algorithm, and verified theextraction of feature signal in this method by the rolling bearing fault testing data,proved the effectiveness the independent component analysis algorithm based onpulse reference signal for rolling bearing fault diagnosis.