Bead geometry prediction based on SVM

Bead geometry is one of the important factors of the joint strength and related properties, and is also one of the important contents of welding quality control. Support vector machine (SVM) model of CO₂ welding bead geometry prediction was established by using welding current, arc voltage and welding speed as network inputs. Bead width, bead penetration depth and bead height were predicted by using linear kernel function, polynomial kernel function, radial basis kernel function and exponential radial kernel function. The results show that the bead geometry can be effectively predicted by using SVM model of exponential radial kernel function. The prediction model can provide the basis for on-line control of welding quality.