| Abstract Scope |
Metal additive manufacturing involves complex thermal behavior, while conventional numerical methods and existing machine learning approaches often suffer from high computational cost, strong dependence on training data, and limited generalization capability. This study develops a physics-informed neural operator (PINO) framework for transient heat transfer modeling during the manufacturing process. The proposed framework generalizes across varying geometries, process parameters, including laser power and scanning speed, and material properties. By incorporating an initial-condition encoder, the model further enables thermal prediction across multiple deposition layers. Results demonstrate accurate temperature-field prediction with substantially reduced computational cost while maintaining robust performance under unseen manufacturing conditions without retraining. The proposed approach provides a unified and efficient framework for generalized thermal prediction in metal additive manufacturing and supports future process optimization and control. |