CMT Welding Process Monitoring Based On Acoustic Signal Multi Feature Fusion

With the rapid development of science and technology,the requirements for the quality of welded joints in the engineering field are becoming more and more demanding.The traditional MIG welding technology is far from meeting the standard of high strength requirements at the welded joint,so a new environmental protection welding technology has come into being.Cold metal transition technology(CMT)can overcome the technical shortcomings of traditional MIG welding to a considerable extent in terms of automation,energy conservation,and environmental protection,and improve the work efficiency of MIG welding technology.Since CMT welding is technically impossible to simulate the simulation process at this stage,it is necessary to conduct a real-time inspection of the welding process.In addition,research on CMT welding process detection based on arc sound is mainly focused on the use of a single feature analysis part of the feature extraction algorithm for multi-feature fusion is less studied,therefore,the analysis of the acoustic signal through a variety of features and combined with multi-feature fusion algorithm,to explore the impact of acoustic signal changes on the weld forming is of great significance.In view of the above issues,this paper investigates the practical detection of CMT welding acoustic signals as follows.Firstly,based on the CMT welding equipment and high-precision microphone combined with the sound field principle,sound detection theory designed a CMT welding sound monitoring system,the arc sound signal during the welding process through the sound monitoring system for acquisition and preliminary processing to obtain the accurate sound signal.After the successful construction of the CMT welding sound monitoring system combined with the CMT welding test platform to carry out prefabricated defects in the CMT welding test,to achieve the purpose of prefabricated welding defects in welding,the use of electric drills and other tools to prefabricate defects in the plate.Secondly,the combination of three time-frequency domain methods(short-time fourier transform,mel spectrum,wavelet transform)to analyze the welding sound signal: for the shorttime fourier transform,by comparing a variety of window functions to select the optimal window function to match the sound signal,and use the window function to optimize the shorttime fourier feature extraction method to obtain the welding process,defects and sound signal short-time fourier time-frequency relationship;for the Meier spectrum,by Design Meier filter,the acoustic signal noise filtered out after the amplification of the defect information,the welding process,defects,and acoustic signal mel spectrum time frequency relationship;for wavelet transform,select the wavelet base with the highest similarity to the arc acoustic signal,the welding defects,and acoustic signal wavelet time frequency relationship.Finally,a multi-feature fusion algorithm for a linear fusion of the three time-frequency domain analysis results,and then combined with neural networks to build a single feature and multi-feature based on the CMT welding defect recognition model.After the comparison of the two defect recognition models,verify the feasibility and reliability of the multi-featured defect recognition model proposed in this paper.The test results show that the multi-feature defect recognition model in the CMT welding process recognition is better than the single-feature defect recognition model,its average recognition rate can reach 98%,12% higher than the average recognition rate of a single feature extraction algorithm.In this paper,the multi-featurebased defect recognition model can more accurately identify and monitor the weld condition.