Research Of Aluminum Alloy Pulse GTAW Penetration Recognition Method Based On Welding Sound Signal Frequency Characteristics

Arc weld pool penetration state is one of the most important factors that determine the quality of the weld.Research shows that the welding arc sound characteristics closely associated with penetration status.Experienced welders can judge weld quality with welding sound signal,mainly based on the loudness and pitch.Currently,the research of welding quality control based on the time domain signal feature extraction and speech processing technology has made some achievements.But because of the welding process is affected by many factors,coupled with the complexity of the sound signal analysis in frequency domain,the research about welding quality control based on sound frequency characteristics has been the emphasis and difficulty in this field.In this paper,welding sound signal was acquired at the welding automation control platform with sound sensor system.First,the generation process of welding sound was analyzed.The result showed that amplitude of sound was decided by the rate of current change.The sound generated at the current rising edges is similar with that at generated the falling edges.In addition,the sound signal acquired in our experimental environment mainly consisted of the part generated from pool oscillation and arc shock,the noise can be ignored in analyzing the relationship between sound signals and welding penetration status.The channel model analysis showed that the closed channel formed between the arc and the molten pool strengthen the low-frequency part of the sound,the open channel formed between the arc and narrow seam strengthen the high-frequency portion of the sound.In this paper,one new method to abstract frequency features about penetration was proposed.At first,the total energy in short time was chosen to be the measure,and FFT and statistical methods was used to focus the sensitive frequency band in 0-7000 Hz.Then the sensitive frequency to penetration status was narrowed to the 2500-5000 Hz by 3 layers wavelet packet decomposition and reconstruction.The minimum sensitive frequency was defined to be 3164-4335 Hz by FFT with 1024 points.At last,the short-time average amplitude,standard deviation,total energy and logarithmic average energy in 3164-4335 Hz were calculated to comprise the frequency feature.The SVM based on the feature vector consisted of the short-time average amplitude,standard deviation,total energy and logarithmic average energy in 3164-4335 Hz was established to distinguish different penetration status,the recognition rate reached 99.46%.Then PSO was chosen to optimize the parameters,the recognition rate reached 100%.At last,the software system of PSO-SVM penetration status recognition based on arc sound frequency characteristics was established,the system could achieve the sound signal pre-processing,feature extraction and penetration status recognition based PSO-SVM.