Research Of Rotary Machinery Fault Diagnosis Based On Blind Source Separation And HHT

In the process of state detection and fault diagnosis of rotary machinery, thesituation of multi-interference sources or multi-faults mixed to each other has alwaysbeen the significant research content in rotary mechanical vibration signal processing.Traditional signal processing method is hard to separate and extract aliasing ofmultiple vibration signals accurately. Since the1990’s, high-profile Blind SourceSeparation method does not require too much priori information, only throughobserving signals can restore the unknown signal source in the aliasing signal, whichprovides a new way for multiple fault signal separation. However, traditional BSSmethod requires the observation channel number is greater than or equal to thenumber of source signals when dealing with rotary mechanical vibration signal, evenappear special circumstance of single observation channel in the practical application,and thus the traditional BSS method becomes invalid in such case. Aimed atovercoming this shortage, this paper will combine Hilbert-Huang Transform with itsunique advantage to research on single channel BSS method and its application in thevibration signal processing of rotary machinery. The main works of this paper are asfollows:（1）The BBS algorithm based on differential evolution is proposed in this paper. Thealgorithm uses negentropy of the signal as objective function and differentialevolution algorithm to solve the optimal separation matrix, and adjust the crossoverprobability and mutation factor according to the adaptive state of the signal. Thisalgorithm overcomes the shortcoming such as slow convergence speed and easily fallinto regional optimal value of the traditional optimization algorithm effectively andthus realize blind source separation of mixed signal. The simulation results show thatboth the separation effect and convergence speed of this algorithm is good.（2）The basic concept, basic algorithm and improved algorithm of Hilbert Huang-transformation are expounded in this paper. Because modal mixing phenomenon oftenhappens when the core part of Hilbert-Huang transform carry on empirical modedecomposition, therefore N.E.Huang proposed ensemble empirical modedecomposition. According to the uniform distribution property of white noise, adding a certain amount of white noise to the signal, and then carry on the empirical modedecomposition for many times. Furthermore, average the intrinsic mode functioncomponents in an ensemble way, and thus overcoming the modal mixing effectively.The contrast and analysis suggest that the ensemble empirical mode decomposition isbetter than empirical mode decomposition.（3）Research the fault diagnosis of single channel blind source separation methodbased on the EEMD-BSS deeply. Firstly, this algorithm uses EEMD to increase thedimension of single channel observation signal, and transforms the underdeterminedproblem into the positive definite problems; Secondly, Dominant Eigenvalue Methodis adopted to determine the number of signal source, and thus the effectiveness of theBSS can be ensured; Finally, using to improved algorithm of BBS to processmulti-dimension signal and obtain an effective estimation of the source signal.Simulation experiments and the measured rotor failure experiments show that themethod can separate the rotary mechanical multi-faults signal under the circumstanceof single observation channel effectively and separation effect is good.