Application of neural network-based hardness prediction method to HAZ of A533B steel produced by laser temper bead welding

Temper bead welding is one of the effective repair welding methods instead of post weld heat treatment. With the development and popularization of the laser welding, the advanced laser temper bead welding repair technique has been developed recently. However, it is always laborious works to determine the optimum welding condition for laser temper bead welding. Therefore, in the present study, the hardness prediction system for laser temper bead welding has been constructed using a neural network. Thus, the appropriate welding conditions can be selected before the actual repair welding. Firstly, the high cooling rate in heat affect zone (HAZ) of laser temper bead welding has been measured. Secondly, the hardness database are prepared by the experiment. Thirdly, on the basis of experimentally obtained database, the neural network-based hardness prediction system for laser temper bead welding has been constructed. With it, the hardness distribution in HAZ of laser temper bead welding was calculated based on the thermal cycles numerically obtained by FEM. The predicted hardness was in good accordance with the experimental results. It follows that the new prediction system is effective for estimating the tempering effect during laser temper bead welding.