Multi-Scale modelling and design of thermal energy storage (TES) devices based on cementitious materials.

Abstract

This thesis investigates the potential of using geopolymer concrete as an alternative to ordinary Portland cement (OPC) for thermal energy storage (TES) systems, particularly for high-temperature applications. OPC concrete, the conventional material proposed in TES systems, exhibits thermal degradation at elevated temperatures, limiting its suitability for high-temperature applications. Geopolymer concrete, on the other hand, offers several advantages over OPC concrete for TES, including superior thermal stability, higher heat capacity, and lower environmental impact. To evaluate the potential of geopolymer concrete for TES, a broad research approach was employed, combining numerical modeling, experimental validation, and machine learning optimization. A 2-D numerical model was developed to simulate the thermal performance of TES prototypes made with OPC and geopolymer-based materials. The model successfully demonstrated the superior thermal performance of geopolymer concrete compared to OPC concrete, particularly at high temperatures. Experimental validation of the numerical model was conducted using real TES prototypes made of OPC and geopolymer concrete. The experiments confirmed the superior thermal stability and storage capacity of geopolymer concrete, with temperature differences up to 30-40°C and storage capacity up to 2-3.5x higher than OPC concrete. To further optimize the design and performance of TES systems based on geopolymer concrete, a 3-D computational model was developed. This model enabled systematic evaluation of design choices and operating parameters to maximize the performance of TES systems for up-scale approaches. Finally, machine learning techniques were employed to optimize the design and performance of TES systems based on solid materials. A decision tree machine learning (ML) model was trained to predict TES performance metrics based on a dataset generated from the validated numerical model. The ML model was then used in conjunction with multi-objective optimization to identify Pareto optimal solutions that balanced objectives such as efficiency and pressure drop for up-scale design.