Physics-Informed Neural Network (PINN) – Based modelling and post-processing investigation in Laser Powder Bed Fusion (LPBF) of aluminum alloys

Laser Powder Bed Fusion (LPBF) shows great potential for producing complex metal components, yet faces challenges in thermal modelling, parameter optimization, and performance enhancement. Traditional numerical methods like FEM and CFD accurately capture temperature evolution and melt pool dynamics but suffer from high computational costs, limiting their use for fast prediction. Data-driven machine learning models are efficient but depend on large datasets and lack physical consistency. For high-performance aluminum alloys like Scalmalloy® (Al–Mg–Sc–Zr), systematic investigations remain limited, hindering broader industrial application. This study addresses these issues through two main stages: pre-process temperature field modelling with parameter calibration, and post-process performance optimization. In the modelling stage, a three-dimensional transient heat transfer model was developed based on a Physics-Informed Neural Network (PINN) framework. By embedding the heat conduction equation and boundary conditions into the loss function, the model achieves high-accuracy temperature prediction under limited data. A double-conical volumetric heat source describes energy distribution in keyhole-mode melting, offering both tunability and interpretability. The PINN model reduces computational cost by several orders of magnitude compared to CFD while maintaining high accuracy. For calibration, Pearson correlation and inverse optimization identified lower cone height (Hₗ) and top radius (Rₜ) as key parameters controlling melt pool depth and width. Empirical mappings between these parameters and laser variables enabled self-adaptive calibration. Validation against experiments and CFD showed average relative errors below 5% in melt pool geometry, with millisecond-level prediction speed. In the post-processing stage, Scalmalloy® components were investigated under machining (M), shot peening (SP), and combined treatment (M+SP). Shot peening induced hardening up to ~200 μm depth and compressive residual stresses to ~300 μm. Low-intensity SP (150 mm, 4 bar, 20 s) balanced surface smoothness and stress induction, while high-intensity peening caused surface damage. Fatigue testing showed ~55–60% life enhancement, with M+SP achieving the best overall performance.