Surrogate Models for 3D Finite Element Creep Analysis Acceleration

Jason Abdallah's presentation centered around the integration of Artificial Intelligence (AI) and Machine Learning (ML) in the acceleration of 3D Finite Element Creep Analysis for turbine blades. The presentation delved into the challenges of Multi-Disciplinary Optimization (MDO) and Robust Design (RD), highlighting the necessity for complex design models for accurate lifetime modeling of turbine blades. A key focus was on the incorporation of AI into MDO processes, proposing a surrogate modeling approach to reduce computational times dramatically. The approach involves combining traditional elastic-plastic calculations with rapid AI-based creep predictions, thus enabling faster responsiveness to field issues and integration of non-linear material behaviors into MDO applications. The presentation showcased how AI surrogate models could explore robust design under these uncertainties. The results section highlighted the effectiveness of the MLP Ensemble model, which outperformed both baseline and single models, demonstrating its ability to predict stress, creep strain, and deformation more accurately. The conclusion and outlook sections of the presentation underscored the potential of AI-augmented visco-plastic analysis in exploiting design spaces and creating robust designs.