CFD Modelling for Improved Components in CO2 and Ammonia Vapour Compression Systems

Abstract

Refrigeration and heat-pumping systems are shifting to natural, environmentallyfriendly refrigerants like CO2 to combat global warming. Ejectors are often used in these systems for expansion work recovery to enhance efficiency. However, ejector design is complex due to interdependent parameters and flow complexity, necessitating advanced models and tools. This thesis aims to improve CO2 ejector modeling for robust design optimization and a better understanding of system operation. To create a fundamental background for this work, key knowledge gaps on this topic are reported in an exhaustive review of CO2 two-phase ejector flow modeling. An overview of different available ejector models is reported and highlights the strengths and weaknesses of the different approaches. Other aspects, such as turbulence, non-equilibrium conditions, experimental data, and model applications are thoroughly reviewed. Different models are implemented into the ANSYS Fluent computational fluid dynamics(CFD) framework, and a comparative study of these models is performed. An experimental test campaign is conducted to validate models implemented in this work. To explore novel ejector concepts and develop improved ejector design methodologies, an algorithm for automated CFD model setup was developed to generate a database of CFD results. The Gaussian Process Regression (GPR) machine learning model is applied for modeling ejector performance trained on this database. Additionally, a numerical investigation of a novel swirl bypass concept for performance improvements of ejectors at off-design conditions is conducted. Based on the review of current CO2 ejector models, it is found that significant discrepancies between experiments and model prediction are still found. These are typically attributed to non-equilibrium thermodynamics and turbulence modeling. Based on CFD model comparisons, it was found that stable and accurate modeling of CO2 ejectors with a numerically efficient CFD method is challenging and requires further development, especially for low motive-pressure conditions. The novel two-fluid model presented in this work can predict CO2 ejector performance with appropriate parameter selection, but potential challenges for further studies include the complexity of experimental tuning and numerical instabilities. The homogeneous equilibrium CFD model was experimentally validated and reproduced mass flow rates within 2-12% and 3-50% error for the motive and suction flow rates, respectively. The GPR machine learning model algorithm was applied for modeling ejector performance with various ejector geometries and at various operating conditions with mean errors in entrainment ratio below 0.1 [-]. The algorithm was able to map ejector performance for off-design conditions, explore and optimize ejector designs, and predict local flow structures. The numerical investigation of the swirl bypass ejector indicated that designing such a concept is sensitive to the specific ejector design and operating conditions. A reduction in entrainment ratio of 2-20% is obtained when operating with a swirl bypass inlet. The flow structure inside the ejector with a swirl bypass is also investigated in detail. In conclusion, it is found that CO2 ejector modeling using CFD is a valuable tool for their design. These models in combination with machine learning have been shown to be applicable for ejector design algorithms and performance mapping. Exploration of the novel ejector concepts using CFD is considered a key way to discover novel ejector improvements. Improved modeling approaches have a direct impact on design tool accuracy, and further experimental studies and numerical developments are valuable to enhance CO2 ejector models in terms of accuracy, speed, and stability.