Revolutionizing Air-Cooled Heat Exchanger Design for Sustainable Energy
As the global demand for renewable energy grows, concentrated solar power (CSP) has emerged as one of the few sustainable technologies capable of reliable, large-scale energy storage. The recent development of the supercritical carbon dioxide (sCO2) Brayton cycle has further enhanced the cost-competitiveness of CSP plants. However, a critical challenge remains – the design of efficient, cost-effective dry cooling systems for these plants, which are often located in arid desert regions.
Conventional water-based cooling solutions are not economically viable in these environments. This has driven recent research into the optimization of air-cooled heat exchangers, also known as “dry coolers,” for sCO2 Brayton cycle CSP applications. Yet, despite these advancements, current dry cooler designs remain far from optimal, as the complex interdependencies between design parameters and their impact on system performance are not fully understood.
In this article, we present a pioneering machine learning-based framework that revolutionizes the design of dry coolers for sCO2 CSP plants. By developing a high-fidelity simulator of air-cooled heat exchanger performance and integrating it with advanced Bayesian optimization techniques, our approach can automatically generate cost-effective dry cooler designs tailored to any location worldwide. This not only reduces the lifetime cost of these critical components by up to 67% compared to existing solutions but also paves the way for more widespread adoption of sustainable CSP technology.
Mastering the Complexity of Dry Cooler Design
Designing an efficient dry cooler for an sCO2 Brayton cycle CSP plant is a multi-faceted challenge. The performance of the heat exchanger is heavily dependent on the main compressor inlet temperature, which is directly controlled by the cooler. However, this relationship is influenced by a myriad of design factors, including tube and fin dimensions, pitch, thickness, and fan configurations, among others.
Existing research has largely focused on the impact of varying power cycle set points and ambient conditions on dry cooler performance. Only a few studies have explored the effect of changing individual design parameters, such as tube diameter or fin thickness. This is because the complex, nonlinear, and non-differentiable nature of the overall heat transfer coefficient makes it extremely difficult to model the system analytically.
To overcome these challenges, we have developed a physics-based simulator that can accurately model the heat transfer between the hot sCO2 working fluid and the cross-flowing air. At the core of this simulator is the well-established Logarithmic Mean Temperature Difference (LMTD) method, which enables us to propagate the heat transfer along the finned tubes element-wise.
By simulating the pressure drop and temperature variations within the heat exchanger, our framework ensures that the output sCO2 properties always meet the required supercritical conditions. This is achieved through an iterative approach that dynamically adjusts the tube lengths until the desired output temperature is reached.
Importantly, our simulator also incorporates a detailed cost model, accounting for material, labor, and operational expenses. This allows us to optimize the dry cooler design not just for technical performance, but also for lifetime cost-effectiveness.
Harnessing Machine Learning for Optimal Dry Cooler Design
Traditional optimization techniques, such as mathematical programming or gradient descent, are ill-suited for this problem due to the complex, non-differentiable nature of the simulator. Instead, we leverage the power of Bayesian optimization, a global optimization method that can navigate high-dimensional design spaces efficiently.
Specifically, we employ the Trust Region Bayesian Optimization (TuRBO) algorithm, which has been shown to outperform other state-of-the-art Bayesian optimization methods in tackling complex, high-dimensional problems. TuRBO utilizes multiple local Gaussian process models, each with its own trust region, to balance exploration and exploitation during the optimization process.
By seamlessly integrating our high-fidelity simulator with the TuRBO algorithm, our framework can automatically explore a vast space of valid dry cooler designs, systematically identifying the most cost-effective solution for a given location. This inverse-design approach represents a significant advancement over previous work, which has only investigated changing one design parameter at a time.
Optimizing Dry Coolers for Diverse Climates
To demonstrate the versatility and effectiveness of our approach, we have applied it to optimize dry cooler designs for a range of locations around the world, spanning arid deserts and humid tropics. Using publicly available data from the National Renewable Energy Laboratory’s National Solar Radiation Database, we selected six locations with the highest mean yearly direct normal irradiance (DNI) – a key factor in CSP site selection.
The results are striking: our optimized dry cooler designs reduce the lifetime cost by up to 67% compared to recently proposed solutions. This substantial savings can be largely attributed to reductions in the size and material quantities of the heat exchanger tubes, which account for a significant portion of the overall cost.
Interestingly, the most cost-effective designs are found in locations with higher mean DNI and lower ambient temperatures, such as Waucoba Mountain, California, USA, and Antofagasta, Chile. In these environments, the cooler air more effectively cools the sCO2 working fluid, allowing for shorter overall tube lengths and lower material requirements.
To further explore the impact of ambient temperature on dry cooler design, we conducted sensitivity studies by varying the inlet air temperature while keeping the desired temperature difference across the heat exchanger constant. We also investigated the scenario where both the surface area and airflow are simultaneously adjusted to optimize the design.
The results demonstrate that the lifetime cost of the dry cooler increases exponentially with rising ambient temperatures, underscoring the critical importance of considering environmental factors in the design process. By allowing for more flexibility in the air temperature difference, our framework was able to identify even more cost-effective solutions, highlighting the value of adaptive, location-tailored designs.
Unlocking the Potential of Sustainable Energy Generation
The development of our machine learning-driven dry cooler optimization framework represents a significant advance in the quest to make concentrated solar power a truly cost-competitive renewable energy solution. By reducing the lifetime cost of these essential cooling systems by up to 67%, we have the potential to unlock the vast solar energy potential in arid regions around the world, accelerating the transition to a sustainable energy future.
Importantly, our approach is not limited to CSP applications; the modular nature of our simulator and optimization framework means that it can be easily adapted to other air-cooled heat exchanger applications, such as industrial processes or waste heat recovery systems. By providing a systematic way to identify cost-effective designs, we hope to facilitate the widespread adoption of air cooling technologies, contributing to a more energy-efficient and environmentally responsible global landscape.
As we continue to push the boundaries of sustainable energy generation, the ability to harness advanced machine learning and simulation techniques will be crucial. Our work demonstrates the power of this approach, paving the way for a new era of cost-effective, high-performance air-cooled heat exchangers that can help realize the full potential of concentrated solar power and beyond.
