Penetration States Recognition Of MAG Welding Based On The Co-sparse Representation Of Arc Sound

The monitoring of weld quality is a major research topic in the welding engineering.Usually,the arc sound contains the information about the penetration states.An experienced welder can roughly judge the weld quality by the arc sound.It is important to recognize the penetration states for the on-line monitoring of the weld quality using the arc sound signals.Analyzing the frequency spectrum of the arc sound signals acquired during the Metal Active Gas(MAG)welding process,the modifying Mel filter bank is used for extracting the Mel Frequency Cepstrum Coefficient(MFCC)features of the arc sound.Then,the co-sparse representation of the MFCC feature can be employed to build a discriminative model and recognize the different penetration states.The work of this thesis can be summarized as below:1.By using the short time Fourier transform(STFT)to analyze the frequency spectrum of the arc sound signal,a modified Mel filter bank is proposed to extract the MFCC features of the arc sound signal,where each center frequency of the Mel filters in different frequency ranges is modified to extract the feature of the arc sound signal.2.Based on the co-sparse representation model,a novel algorithm is proposed for recognizing the penetration states.The co-sparse representation of the MFCC feature can be obtained by learning the analysis dictionary from the training signal,and the representation can be employed to build the discriminative model for recognizing the penetration states to the MAG welding.