Weld appearance prediction with BP neural network improved by genetic algorithm during disk laser welding

• The morphology of molten pools is measured by molten-pool-shadows.
• Four characteristics of the molten-pool-shadows are defined and extracted.
• Principal components analysis is applied to analyze the characteristics.
• BP neural network model of weld appearance is improved by genetic algorithm.

The appearance of welds is the external manifestation of welding quality. The morphology of molten pools is significantly associated with the weld appearance, but the approach to measure the morphology of molten pools during laser welding remains an outstanding challenge up to now. In this study, the shadows of molten pools were formed to describe the morphology of molten pools. Principal components analysis (PCA) is applied to analyze the characteristics of the molten pools’ shadow in order to reduce their redundancy. Then BP neural network improved by genetic algorithm (GABP) is established to model the relation between welding appearance and the characteristics of the molten-pool-shadows. The effectiveness of the established model is analyzed through two different welding speed experiments, and the results verify its prediction performance. The work provides an effective way to predict the weld appearance and assess the welding quality in real-time.