Penetration quality prediction of asymmetrical fillet root welding based on optimized BP neural network

Abstract

Penetration morphology has an important influence on the weld quality. Due to the nonlinear and strong coupling characteristics of welding process, neural network is often used to predict weld formation and quality. However, the prediction method of fillet weld penetration needs further exploration due to the difficulty in quality evaluation. More intricately, fillet welds of the medium-thickness plate with one-side V-groove are structurally asymmetrical, which makes the penetration quality difficult to guarantee. In this study, the penetration quality of asymmetrical fillet welds is depicted by two characteristic quantities: penetration depth and penetration deflection. The penetration deflection can be reflected by leg length on both sides. After correlation analysis, a back-propagation neural network(BPNN) optimized by Mind Evolutionary Algorithm (MEA) is proposed. The welding current, welding speed, torch work angle and real-time molten pool width were chosen as input parameters, and the penetration of blunt edge and the leg length on both sides of the weld were chosen as output parameters. The results demonstrated that it is feasible and reasonable to predict the penetration of asymmetrical fillet welds by this model. Experimental comparison showed that the optimized model has a significant improvement in predictive performance. The prediction error of blunt edge penetration is controlled within 0.1 mm, and the prediction error rate of leg length is less than 7 %. It satisfies the accurate prediction of penetration quality of asymmetrical fillet welds, and lays the foundation for the study of penetration morphology control of asymmetrical fillet automatic welding.