关键词: FastICA, 独立成分分析, 经验模式分解, 盲源分离, 奇异值分解

Abstract: Blind source separation (BSS) is an effective method to diagnose multi-fault of mechanical equipment. Empirical mode decomposition (EMD) is a potent tool to analyze non-stationary signals and it can decompose nonlinear and non-stationary signal into a set of linear and stationary intrinsic mode functions. In blind source separation of mechanical fault signals, single-channel mechanical signal separation is an ill-conditioned problem. Here, by combining the respective advantages of BSS and EMD, a blind source separation method based on empirical mode decomposition is proposed to solve the difficult problem of single-channel mechanical signal separation. This algorithm is composed of three steps. Step 1 is empirical mode decomposition of single-channel mechanical signal and its combining with intrinsic mode functions. Step 2 is to estimate mechanical sources number by singular value decomposition (SVD). Step3 is to recombine multi-channel mixed mechanical signals according to estimated sources number and separate mechanical sources with FastICA algorithm. This method is applied to the simulation research of bearing and gear in order to correctly separate their source signals. Simulation research indicates that it can well solve the difficult problem of source number estimation and blind source separation of single-channel mechanical signal.

Key words: Blind source separation, Empirical mode decomposition, FastICA, Independent component analysis, Singular value decomposition