Random Forest Regressor for CHT turbine vane analysis

With increasing turbine inlet temperatures in modern gas turbines, efficient internal cooling of stator vanes is critical to ensure structural integrity and thermal performance. This thesis focuses on the optimization of the internal cooling system of the NASA C3X turbine stator vane, originally investigated by Hylton et al. The objective is to enhance thermal uniformity and reduce peak metal temperatures by modifying the distribution and geometry of ten internal cooling channels. The study is carried out in two main phases. In the first phase, a high-fidelity 3D Conjugate Heat Transfer (CHT) simulation is conducted using ANSYS Fluent. A steady-state, pressure-based solver is used to solve the RANS and heat conduction equations, and three turbulence models (k–ω SST, k–kl–ω, y–Re⁰ SST) are tested. Model results are validated against experimental data, and the k–ω SST model is selected for its better agreement. In the second phase, geometric configurations are generated using Latin Hypercube Sampling, by varying the hole positions (x, y coordinates) and radii within defined limits. Invalid samples—those violating geometric constraints such as overlapping or boundary contact—are filtered. A Random Forest Regressor is then trained to map geometric input variables to thermal performance metrics such as average blade temperature and total coolant mass flow rate. The predictive performance of the Random Forest model is assessed using standard evaluation metrics, including the coefficient of determination (R²) and the mean squared error (MSE). This work demonstrates the integration of CFD simulation and machine learning to support efficient design optimization. The framework provides a robust and flexible approach for thermal management design in high-performance turbine systems.