Mechanical acoustic fault diagnosis based on improved semi-blind extraction method

Abstract

According to statistics, about 30% of mechanical faults are caused by rolling bearing. Two or more combined failures may exist in the rolling bearing when the equipment is running. Many acoustic analyses just meet underdetermined situation because the number of microphones is less than the number of fault sources. In order to deal with these kinds of monitoring problems, a mechanical failure diagnosis method based on reference signal frequency domain semi-blind extraction is proposed. In this method, dynamic particle swarm algorithm is used to construct improved multi-scale morphological filters which are applied to mechanical failure in order to weaken the background noises; thus reference signal unit semi-blind extraction algorithm is applied to do complex components blind separation band by band, coupled J-divergence of complex independent components is employed as distance measure to resolve the permutation; finally the estimated signal could be extracted and analysed by envelope spectrum method. Comparing to the time-domain blind deconvolution algorithm based on fuzzy clustering, it has several advantages such as being more effective and more accurate. Results from rolling bearing acoustic diagnosis experiment validate the feasibility and effectiveness of proposed method.