Penetration prediction of narrow-gap laser welding based on coaxial high dynamic range observation and machine learning

Narrow-gap laser welding is a novel joining process for thick-walled applications. However, it is challenging to obtain accurate root weld penetration as a crucial parameter for evaluating welding quality due to the limited spatial position. The images of the front-side weld pool are one of the most effective signals for reflecting the root weld penetration. Hence, this paper proposes using a high dynamic range (HDR) camera to capture the weld pool morphology under narrow-gap laser welding. The characteristics of weld pool under different root weld penetration states are obtained and analyzed through orthogonal experiments. Furthermore, the root weld penetration states prediction model is constructed based on a backpropagation (BP) neural network. The results show that the accuracy of this prediction model is approximately 94.74 %, so the proposed method can be further applied to establish real-time root weld penetration control system of narrow-gap laser welding process.