Abstract: Aiming at the problem of difficulty in detecting local occlusion defects at the side wall of narrow gap GMAW,an online detection method for occlusion defects based on arc acoustic signal feature extraction and processing was proposed. By analyzing the arc morphology and arc acoustic signal characteristics of normal,critical occlusion and occlusion,it was confirmed that the arc morphology change caused by the side wall of the groove was an important factor affecting the change of arc acoustic signal. On this basis,the time-frequency analysis of wavelet packets was used,and the standard deviation between feature classes was introduced as an evaluation index to determine the sensitive features that can effectively identify the three welding states.Sigmoid support vector machine and five-fold cross-validation were used to establish a prediction model,and the experimental results show that the model can better realize the prediction classification of three welding states,and the recognition accuracy reaches 96.0 %.

Key words: narrow gap GMAW, arc acoustic signal, occlusion defects, time-frequency analysis