Study On Fault Diagnosis Of Rotating Machine Based On Blind Source Separation

Rotating machinery play an important role in many engineering fields, such aspetroleum, chemical, metallurgy, electric power, aerospace, etc. Therefore, fault diagnosishas important significance for rotating machinery in engineering. In practical application,the mechanical vibration signals are non-stationary, and are often faced with variousinterruptions or condition of fault coupling. How to extract the expected fault feature fromthe observed signals effectively is the bottleneck of mechanical fault diagnosis.Blind Source Separation is a new method of signal recognition, which is developedrapidly in recent years. With systematically studied the lack of traditional BSS algorithms,this paper presented three improved algorithms applid in different conditions. And put thealgorithms into the practical rotating machinery fault diagnosis.Firstly, in view of the fixed-step EASI algorithm is unable to resolve thecontradiction between the convergence speed and the error in the steady state, this paperproposes a variable step-size EASI algorithm based on kurtosis. Computer simulationresults show that the convergence and steady-state performance of the proposed methodoutperforms the fixed-step EASI algorithm, and it is a primely solution to the fixed-stepEASI algorithm of resolving the contradiction between convergence speed and the error inthe steady state. Then the proposed method is applied to the separation of mixed faults ofgear box, and the results further verify its effectivity.Secondly, aiming at the problem that in the observation conditions ofunderdetermined, the traditional blind source separation methods produce poor result, anunderdetermined blind source separation method of machine faults based on Time SeriesDecomposition is proposed. The simulation results confirm the feasibility and validity ofthe proposed method. Finally the proposed method is applied to the separation of mixedfaults of rolling bearing, its fault features are fully detected and the effectiveness of theproposed method is verified.Thirdly, the traditional independent component analysis is too difficult to solve theproblems of underdetermined blind source separation and statistically correlated sources separation existed in mechanical fault diagnosis. Assuming some subcomponents ofcorrelated machine vibration sources are independent, a novel blind source separationmethod based on subband extraction of ensemble empirical mode decomposition isproposed to solve the problem of single-channel statistically correlated mechanical signalsseparation. The simulation and experiment testify the validity of the proposed method.Finally, a BSS package using MATLAB and Visual C++software was designedbased on the MFC. This package made the operation of BSS simulation and applicationsconveniently.