Surface Temperature Prediction of Vacuum Coating Substrate Based on Neural Networks

In order to improve the prediction of front temperature during the vacuum coating process, the hybrid modeling method was proposed to integrate finite element simulation with a one-dimensional convolutional neural network (1D-CNN). Through experimental measurements, an initial correlation between the backside and frontside temperatures was established, and additional training data was generated by simulation software Ansys. The trained 1D-CNN model can effectively capture complex temperature distributions through experimental and simulated data. Evaluation results are superior to the conventional methods with a Mean Absolute Error (MAE) of 4.0783 and a Root Mean Square Error (RMSE) of 6.2497, which effectively limits the prediction error within ±5°C. By integrating physics-based simulations with data-driven learning, this approach enables accurate and real-time temperature estimation. It enhances process control and coating quality in vacuum magnetron sputtering applications.