Deep Learning-Assisted Numerical Simulation of Laser Cladding for Cold-Rolled Rolls

Cold-rolled rolls play a crucial role in steel plate rolling, and their surface performance directly affects production efficiency and product quality. Due to the harsh working environment, the roll surface is prone to wear, and frequent roll replacements lead to significant economic losses. Therefore, an efficient repair method is urgently needed. Laser cladding, as an advanced surface repair technology, can effectively repair surface defects and extend the service life of rolls. However, direct experimentation leads to material waste. The temperature field is a key factor influencing cladding quality. To address this, this paper proposes a hybrid modeling approach that combines finite element analysis (FEA) with deep learning, specifically Long Short-Term Memory (LSTM) networks, to optimize temperature field control during the laser cladding process. Considering the large size of cold-rolled rolls, traditional FEA simulations are time-consuming. This study first obtains fundamental temperature field data through finite element numerical simulations and then uses the LSTM model to predict temperature evolution. To improve prediction accuracy, an Improved Particle Swarm Optimization (IPSO) algorithm is employed to optimize the hyperparameters of the LSTM model. This approach effectively reduces the computational time for temperature field simulations, enhances production efficiency, and extends the service life of cold-rolled rolls, providing a new technical solution for laser cladding repair.