Single-Channel Blind Source Separation and Its Application on Arc Sound Signal Processing

Welding arc sound signal is an important signal in intelligent welding diagnosis, due to its informative, noncontact, easy collected. However, due to the interference of ambient noise, the arc sound signal is highly complex and noisy, which seriously limits the application of arc sound signals. In this paper, a single-channel blind source separation (BSS) algorithm based on the ensemble empirical mode decomposition (EEMD) is proposed to purify and denoise the arc sound signals. First, EEMD is used to decompose one channel signal to several intrinsic mode functions (IMFs). Second, principal component analysis (PCA) is used to reduce the multidimension IMFs to low-dimension IMFs, which are regarded as the virtual multichannels signals. Finally, independent component analysis (ICA) separates the virtual multichannels signals into target sources. The approach was tested by simulation and experiments. The simulated results show that signals separated from mixed signal using this approach highly match the source signals that make up the mixed signal. Moreover, experimental results indicated that the source signals of arc sound were effectively separated with the environmental noise signals. The statistical characteristics of the spectrum in 5–6.5 kHz band extracted from the arc sound source signals can accurately identify the two types of weld penetrations.