Scientific Machine Learning

PINNs and PINOs for DED process prediction

This work develops Scientific Machine Learning models for fast, physics-consistent prediction of thermal histories in Directed Energy Deposition (DED). By reducing reliance on costly finite element simulations, these models enable efficient, meshless thermal prediction while preserving the governing heat-transfer physics, supporting improved process understanding, optimisation and decision-making in additive manufacturing.

Technical highlights:

  • Physics-Informed Neural Network (PINN): Developed a meshless physics-informed framework for DED thermal prediction by enforcing heat-transfer equations, initial conditions and boundary conditions in the loss.
  • eXtended Physics-Informed Neural Network (XPINN): Improved PINN scalability and robustness for multi-layer problems through domain decomposition.

  • Physics-Informed Neural Operator (PINO): Enabled one-shot thermal prediction across varying geometries, toolpaths, materials and process parameters without retraining.

▼ Relevant Publication(s)

Peng B, Panesar A. Multi-layer thermal history prediction framework for Directed Energy Deposition based on Extended physics-informed neural networks (XPINN) Additive Manufacturing. (2025)

Peng B, Panesar A. Application of scientific machine learning techniques to metal additive manufacturing simulation: a comparison between PINN and XPINN. In: International Solid Freeform Fabrication Symposium (SFF). Austin, Texas. (2025)

Peng B, Panesar A. Multi-layer Thermal Simulation of Directed Energy Deposition using Physics Informed Neural Networks Real-World Applications of AI and ML in Digital Engineering Seminar [Internet]. (2025)

Peng B, Panesar A. Multi-scan thermal simulation using physics-informed neural network Additive Manufacturing. (2024)

Peng B, Panesar A. Predicting temperature field for metal additive manufacturing using PINN International Solid Freeform Fabrication Symposium. p. 881–96. (2023)

Scroll to Top