Lifing Prediction Under Casting and Boundary Condition Deviations in Turbine Blades With 3-D Transformer-Based Neural Operators

Abstract Turbine design optimization is a critical aspect of the engineering process, ensuring efficient performance and longevity of these essential power generation components. One significant challenge in turbine design is the analysis of creep and Thermo-Mechanical Fatigue to exploit the design space and create a robust design. This paper presents a novel approach using deep learning techniques to model the relationship between geometry and boundary conditions deviations of a turbine blade and its lifing in 3-D. Incorporating AI-driven algorithms and machine learning models, this study aims to develop an efficient, automated design optimization process that adapts to the evolving requirements of turbine technologies. By leveraging simulated data and incorporating relevant material properties, the models are trained to predict creep impact under geometry deviation and different operating conditions. Specifically, using a novel process chain, geometry and boundary conditions deviations for a turbine blade design are generated. These deviations consist of stacking and rotating the blade airfoil, changing the wall-thickness of the airfoil pressure and suction side, as well as varying the cooling air supply and hot gas boundary conditions. The resulting data from this process is then used to train a deep learning surrogate and evaluated with respect to its predictive performance. The proposed AI-driven approach enables the identification of optimal design parameters that minimize the adverse effects of geometry deviation, leading to enhanced turbine performance and extended service life. By mitigating these challenges, AI-driven optimization can contribute to the development of more efficient and sustainable power generation solutions for the future.