Feature extraction and quantitative characterization of AE signals from welding process based on improved Morlet wavelet and approximate entropy

Abstract The acoustic emission (AE) signals in the welding process have the characteristics of multi-mode, multi-modality spectral and complex noise background. Combined with the time-frequency localization characteristics of Morlet wavelet, a parametric band-pass filter based on Morlet wavelet is proposed to conduct separation and extraction of the different AE source signal in the welding process. The Morlet wavelet function with adjustable parameters that is suitable for the frequency domain characteristics of AE signals is constructed. The minimum value of the information entropy is adopted as the basis for judging the optimal selection of parameters of the wavelet function. According to the frequency characteristics of the single welding excitation source signal, the wavelet separation and extraction of the mixed excitation source AE signal during welding process are carried out, and the Approximate Entropy (ApEn) is used to construct the AE signal feature vectors of the friction, arc and crack. Support Vector Machines (SVM) is applied to realize the intelligent recognition of AE signal of friction, arc and crack in the welding process respectively, which provides a new method for the realization of crack AE detection in the welding process.