基于小波包和独立分量分析的微弱多源故障声发射信号分离

针对旋转机械设备中同时存在的裂纹、摩擦等多故障源信号难以检测和分离的问题,提出了一种基于小波包分析(WPA)与独立分量分析(ICA)的多源故障信号提取方法,即首先用WPA对含噪线性混合信号降噪预处理,由db2小波基函数进行5层分解后保留62.5~187.5kHz频段信号,然后采用ICA中的FastICA算法对降噪后的混合信号分离,最后对各通道分离出的信号用收缩函数进行频段内去噪处理.对不同输入信噪比的含噪微弱裂纹和摩擦信号进行提取和分析的结果表明,该方法能有效提取出输入信噪比大于-15dB的裂纹和摩擦信号.当混合信号信噪比为-15dB时,裂纹和摩擦信号的输出信噪比分别为-1.31和-1.36dB,相关系数分别为0.62和0.63,提取效果好于结合小波包和FastICA分离方法(信噪比分别为-1.74和-2.06dB,相关系数分别为0.59和0.59)以及单独采用FastICA算法(信噪比分别为-4.57和-4.31dB,相关系数分别为0.17和0.19).因此,所提出的综合WPA和ICA的方法是一种较好的多源微弱信号提取方法.

Separation of Weak Multi Source Fault Acoustic Emission Signals Based on Wavelet Packet and Independent Component Analysis

Abstract： Multi source fault signals (such as crack and friction signals) produced from rotating machinery are difficult to detect and separate; therefore, an extraction method of multi source fault signals based on wavelet packet analysis (WPA) and independent component analysis (ICA) was proposed. The wavelet packet technology was used to reduce the noise outside the frequency band of the linear mixed signals. The signals were decomposed by db2 wavelet into five layers while the signals with the frequency band from 62.5 to 187.5 kHz were reserved. Then, the mixed signals were separated by using the FastICA algorithm. Finally, the shrinkage function was used to reduce the noise in the frequency band. By extracting the noisy weak signals with different input SNRs, the results show that this method can effectively extract the crack and the friction signals with the input SNR higher than -15 dB. Their output SNRs are -1.31 and -1.36 dB and the correlation coefficients are 0.62 and 0.63, respectively, which are higher than those obtained by using the method combined WPA and FastICA and only FastICA algorithm. The SNRs are (-1.74 and -2.06 dB) and (-4.57,-4.31 dB) and correlation coefficients are (0.59,0.59) and (0.17,0.19) for the combined method and FastICA method, respectively. Thus, the method is very suitable for extraction and separation of multi source weak signals.