Meshless PINN-Based Modeling of Hot Forging Processes for Gas Turbine Engine Blade Manufacturing

Blades are critical components of gas turbine engines (GTEs). Precision forging of such components remains computationally demanding because conventional finite element method (FEM) simulations require extensive computational resources and prolonged optimization cycles. The quality of forged blades strongly depends on accurate prediction of the stress-strain state (SSS). Consequently, growing research interest is focused on AI-assisted digital manufacturing approaches. This work establishes the methodological foundations for applying Physics-Informed Neural Networks (PINNs) to precision hot forging processes for GTE blades. The scientific novelty lies in integrating continuum-mechanics equations directly into a Fourier Feature Network-based PINN architecture. The proposed framework enables mesh-free approximation of severe plastic deformation during hot forging. Unlike conventional FEM approaches, the methodology avoids repeated remeshing procedures typical of large-deformation FEM simulations. Automatic differentiation and physics-constrained optimization enable continuous approximation of velocity and strain fields within deforming workpiece domains. The methodology is validated under axisymmetric hot-forging conditions using die-surface collocation points extracted from STL geometry files. One-time network training requires approximately 20 h on standard CPU hardware. After training, the PINN performs field inference over 259,200 spatial points in approximately 1.4 s. Quantitative validation against QForm UK FEM simulations yields an RMSE of 0.557 for the normalized plastic strain profile. The proposed framework is classified at Technology Readiness Level 3 (TRL 3) as a computational proof-of-concept demonstrating the feasibility of PINN modeling for precision hot forging. Compared with conventional FEM simulations, the proposed framework provides an approximately 5000-fold reduction in post-training inference time, excluding the one-time network training cost. These results highlight the potential applicability of the approach to accelerated process optimization, AI-assisted digital manufacturing, and near-real-time digital twin systems for metal forming. No experimental forging data were available for this study. Validation is performed exclusively against QForm UK FEM simulations.