Research On The Diagnosis Of Acoustical Signals Based On Blind Deconvolution

Acoustical fault diagnosis technology has many advantages, for example, we don't need contacting with measurement object, adding quality on the devices, and we can measure signals easily in harsh conditions, etc. However, there are several difficult points to be further studied. Firstly, the observed acoustical signals are mixed signals of all the sound sources, and useful information which represents machinery condition is flooded in the mixture signals. Secondly, models of actual acoustic field are difficult to be established. Thus, an effective way to solve the problems is to separate the source signals from the mixed signals. Blind signal processing which emerged in recent years is a useful tool to feature extraction from acoustical signals.Rolling bearing is a widely applied common element and vulnerable to damage, and a large number of rotating machinery faults are related to bearing failures. Defect bearing will cause severe vibration and noise. It can even lead to destruction of equipments and casualties. Therefore, fault diagnosis and condition monitoring of rolling bearing is of great significance. The usage of acoustical signals in fault diagnosis of rolling bearing is fresh and should be paid more concern.In this dissertation, blind signal processing, especially blind deconvolution, is applied in mechanical condition monitoring and fault diagnosis. Instantaneous blind separation and blind deconvolution theories are summarized. This principles and three kinds of instantaneous blind separation methods (JADE, Informax, FastICA) and four kinds of blind deconvolution methods (Bussgang, a blind deconvolution based on FastICA, a time-domain blind deconvolution based on clustering, a time-domain blind source extraction algorithm) are researched, and acoustical signals of mechanical equipments are analyzed by using those methods. Three improved blind deconvolution algorithms are proposed:1) blind deconvolution based on wavelet preparation; 2) PSO optimized blind deconvolution; 3) RCQGA optimized blind deconvolution. Finally, Simulation and experiments are carried out to demonstration that the improved algorithms are feasible.The main experiments are made in semi-anechoic laboratory and on a test bed for rotor machine fault simulation named QPZZ-Ⅱin normal acoustical environment. Mixed sound signals of two loudhailers are used to test the above algorithms in semi-anechoic laboratory. Defect bearing noise is processed with proposed algorithms and then analyzed by envelopment method, then characteristic frequencies of outer and inner ring fault bearing can be found in envelop spectrum. Experiments results show the effectiveness of these methods. Finally, a blind signal processing toolbox which can analyzes acoustical signals by proposed algorithms, is developed based on Matlab GUIDE.