The study of the electrical signal characteristics under different process parameters and the establishment of the relationship between electrical signal characteristics and forming characteristics are of great significance for the realisation of highly accurate and sensitive closed-loop control of welding quality. To identify the weld width

different process experiments were set up to obtain the corresponding arc signals under different melt widths. The time-frequency characteristics of electrical signals under different processing conditions were analyzed. The time-frequency characteristic values were extracted by the Variational Mode Decomposition（VMD） method

and the fusion width was classified by the Support Vector Machine（SVM） pattern recognition algorithm. The prediction models of different processing parameters and fusion width were constructed. It is proved that the prediction accuracy is 98.6111%

which lays a great technical foundation for the development of information and intelligence in the welding process.