Detection And Recognition Of The State Of MIG Welding Process Based On The Loudness And MFCC Characteristics Of Arc Sound

Arc sound signal is usually generated by the oscillation of the molten pool and the changing of arc energy.It is an important potential source signal for quality monitoring of welding process.In engineering practice,an experienced welder can make a correct judgment on the weld state or weld penetration based on the information of the arc sound.Experiences and a large number of experimental tests have also shown that the human auditory system exhibits superior performance in listening and distinguishing.Therefore,this paper takes MIG welding arc sound as the research object and studies the loudness characteristics and MFCC characteristics.Then it establishes the classification model for the arc sound generated by different droplet transition states and different welding line energies.This method realizes the identification of the droplet transfer state and different welding line energy,provides a new idea for real-time detection of MIG welding state and weld quality.In order to collect the effective MIG welding arc sound signal,a set of MIG welding automatic welding platform and arc sound signal acquisition system was built by using relevant experimental devices and software and hardware systems,which can realize automatic control of welding process,adjustment of welding parameters and automatic acquisition function of arc sound signal.Imitating the processing of sound signals carried out by the auditory peripheral system,the characteristics of arc sound auditory loudness and MFCC are analyzed.The characteristics of arc sound loudness directly express the law of arc sound signal energy changing with time,and also reflect the stability of welding process indirectly;MFCC features reflect arc sound channel information.The frequency domain characteristics and loudness characteristics of arc sound under different droplet transfer states are studied.The results show that in the timefrequency domain,the short-circuit transitional arc sound is significantly different from the other two transitional states,while the coarse and jet transitional arc sound signal waveforms and spectral envelopes are so similar that is difficult to distinguish them.Through the calculation of loudness,we found that the characteristics of the three transitional state arc loudness are distinct.The arc sound loudness feature quantity is constructed and the support vector machine classifier is established.Which realizing the identification of short circuit transition,coarse droplet transition and jet transition.The overall recognition rate is 97.69%.Welding line energy identification based on the characteristics of arc sound MFCC was studied.We found through experiments that the welding line energy is obviously related to the width,height and penetration depth of the weld.The change of the line energy also affects the modulation characteristics of the arc channel and is reflected in the arc sound characteristics.The support vector machine classification model is established by using the MFCC feature of the arc sound.The results show that the overall recognition rate of the different line energy reaches 99.25%,and the calculation speed is fast,which meets the real-time detection requirements of welding state and quality.