Loading...

Application of Machine Learning in the Development of Surrogate Models Based on the Large Eddy Simulation Approach

Abtahi Mehrjardi, Ali | 2024

0 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 58139 (45)
  4. University: Sharif University of Technology
  5. Department: Aerospace Engineering
  6. Advisor(s): Mazaheri, Karim
  7. Abstract:
  8. Gas turbines have essential applications, such as in aircraft engines and power generation systems, making the improvement of performance efficiency in gas turbines crucial. One of the most significant methods for enhancing efficiency is increasing the inlet temperature of the gas turbine, which requires the development of various cooling techniques, including film cooling. The desire for a high-precision numerical analysis method for the thermal design and analysis of film cooling, combined with the need to avoid high computational costs, serves as the primary motivation for this dissertation. Surrogate models have been developed by processing data obtained from the Large Eddy Simulation (LES) approach and utilizing artificial intelligence tools. These models enhance the accuracy of Reynolds-averaged turbulent heat flux models and keep computational costs comparable to conventional turbulence modeling methods. In this dissertation, five surrogate models for turbulent heat flux modeling have been developed. All models demonstrate improved accuracy in predicting the thermal field compared to the baseline model. Two of these five models were developed from scratch through statistical analysis and neural networks, while the remaining three models were derived from optimizing existing popular models. The principle of invariance was applied in developing and optimizing these models to ensure they remain insensitive to coordinate systems, making them applicable across all geometries and film-cooling flows. With the developed surrogate models for turbulent heat flux, the error in predicting the thermal (temperature) field was reduced by up to 58%, while the optimized models reduced the prediction error by up to 43.7%. To demonstrate the generalizability of these models, they were applied to analyze cooling on a flat plate and the leading edge of a turbine blade. The models were also employed in an internal cooling problem, predicting the heat transfer coefficient for channels with dimples. The presence of dimples was shown to reduce the maximum surface temperature of the turbine blade by up to 90 Kelvin
  9. Keywords:
  10. Film Cooling ; Turbine Blades ; Internal Cooling System ; Large Eddy Simulation (LES) ; Alternative Methods ; Turbulent Heat Flux ; Turbulent Prandtl Number

 Digital Object List

 Bookmark

No TOC