Thermal-profile prediction of heat-affected zone for predictive control during laser cladding

During the laser-cladding (LC) process on a metal substrate, the melt pool and the heat-affected zone (HAZ) can overheat; as a result cracks, pores, and thermal deformation develop in or near HAZ. A model predictive control (MPC) approach is a method to control process requirements by anticipating deviations from the goal. In this study, deep learning was used to develop a model (SimVP-LC) that uses a convolutional neural network to predict the thermal profile for MPC of the HAZ during the laser-cladding process. Thermal history data during the process were preprocessed to form images, remove unnecessary information, and correct for distortion. SimVP-LC can accurately predict the thermal profile of the entire substrate including the HAZ after up to 37 frames within a mean absolute error of 10 °C, whereas the average temperature of HAZ > 500 °C. A LC real-time monitoring system that uses MPC was developed and applied to actual experimental environments with WC40Ni powder on an S45C steel substrate. The width of the HAZ could be controlled below a certain level; consequently, the probability of crack creation was decreased. The model proposed here can be used for MPC to enable accurate control of the temperature and size of the HAZ during the LC process.