Development of Computational Functions and Machine Learning Assisted Multi-Objective Optimization Tools for Supercritical CO2 Power Cycles
Date of Award
12-12-2024
Degree Name
M.S. in Mechanical Engineering
Department
Department of Mechanical and Aerospace Engineering
Advisor/Chair
Andrew Schrader
Abstract
Supercritical carbon dioxide (sCO2) power cycles offer significant advantages in thermal efficiency, component flexibility, and adaptability to various heat sources. This study develops computational functions and multi-objective evolutionary optimization tools to enhance the performance of sCO2 power cycles, specifically targeting applications with finite thermal reservoirs. Leveraging the unique properties of sCO2, including its high density and excellent heat transfer capabilities, the research aims to optimize cycle efficiency, reduce component mass, and manage operational constraints. The research focuses on four primary sCO2 Brayton cycle configurations: Direct Heating – Recuperated Cycle (DH-RC), Direct Heating – Non-Recuperated Cycle (DH-NRC), Indirect Heating – Recuperated Cycle (IH-RC), and Indirect Heating – Non-Recuperated Cycle (IH-NRC). Each cycle is analyzed for its thermodynamic performance, component interactions, and potential for efficiency improvements. The simulation functions developed for these cycles incorporate iterative processes to ensure steady-state conditions and accurate energy conservation. Key to the optimization process are machine learning models and genetic algorithms, which handle the high-dimensional data and complex design criteria associated with sCO2 cycles. The NSGA-II algorithm is employed for multi-objective optimization, focusing on maximizing cycle efficiency and minimizing the mass of the turbine and compressor. This approach allows for the generation of Pareto-optimal solutions, providing a diverse set of optimal design configurations. The study also addresses the critical components of sCO2 cycles, including compressors, turbines, and heat exchangers. Detailed simulations and regression models are developed to predict the performance and mass of these components. The optimization framework integrates these models, allowing for comprehensive analysis and design refinement. For example, the heat exchanger models utilize advanced techniques like the Logarithmic Mean Temperature Difference (LMTD) method and incorporate detailed stress analysis based on ASME boiler code equations. Significant results include the identification of optimal design criteria for each cycle configuration. For instance, the DH-RC cycle showed improved efficiency and reduced source heating requirements, making it suitable for small-scale power generation applications. The study's findings highlight the importance of recuperation in enhancing cycle efficiency and demonstrate the potential of sCO2 cycles in various energy generation scenarios, including renewable energy systems and waste heat recovery. The integration of machine learning and optimization tools has proven effective in handling the complex multi-dimensional problems inherent in sCO2 cycle design. The flexibility and robustness of these tools facilitate the exploration of diverse design spaces, enabling the identification of innovative solutions that balance efficiency, cost, and performance.
Keywords
Supercritical Carbon Dioxide, sCO2, Power Cycles, Machine Learning, Multi-objective optimization, Optimization, Brayton Cycles, Pymoo, Heat Exchangers, High temperature creep
Rights Statement
Copyright © 2024, author.
Recommended Citation
Hyland, Christopher, "Development of Computational Functions and Machine Learning Assisted Multi-Objective Optimization Tools for Supercritical CO2 Power Cycles" (2024). Graduate Theses and Dissertations. 7477.
https://ecommons.udayton.edu/graduate_theses/7477
