Optimization of thermochemical energy storage reactors using machine learning

Abstract

Thermochemical energy storage (TCES), where thermal energy is stored in a reversible chemical reaction in a porous powder bed, is a promising technology for large-scale and long-term thermal energy storage. It has been under long-standing investigation for prospective applications, such as the capture of excess heat from industrial processes or storing energy in concentrated solar power plants, to offset their unpredictable energy generation. This study investigates TCES in the SrBr2-system, which offers a high energy capacity and near-perfect reversibility.

However, the scaling up of these reactors is hindered by the limited heat transfer from the heat source, such as reactor walls, to the powder bed. To address this challenge, heat conducting structures, such as fins, are incorporated into the bed to enhance thermal contact and shorten transport paths. Moreover, structural changes through mechanical and physical alteration of the powder bed, as well as changes in the microstructure, lead to changing heat and mass transport properties of the porous medium during cycling. Additionally, deformation of the bed can lead to detachment from the heat conducting surfaces.

Even though physical modeling these effects can be done in principle, developing and parametrizing these models is challenging due to the substantial structural changes happening on multiple scales in the reacting bed. In this contribution, we attempt to overcome these challenges through hybrid modelling, i.e. the combination of physical and data-driven methods. To this end, experimental work is carried out on the macroscale (cm) by thermochemical cycling reactive beds within reactors and measuring conversion and local temperatures inside reactors over time. In addition, imaging of the microstructure (µm) is done using µCT imaging of smaller samples, which can be used to compute effective transport parameters. Then, the available data is used to build a multi-scale model, combining data driven techniques and physical simulations.

In a second step, ML techniques are used to improve the heat transfer inside the reactor by designing optimized heat conducting structures. As direct simulations are prohibitively time consuming, we construct an ML-Based surrogate model, which is trained with a representative sample of physical simulations, and which can predict the performance of the reactor based on the structures’ geometry. This can be done either by training a neural network on simulated data or by using techniques, such as model order reduction, where the non-linearities are handled by a neural network. The surrogate model is then coupled with a topology optimization algorithm based on the level-set method, which is used to calculate optimal geometries for the heat conducting structures. Our contribution will center on the modeling techniques employed and the preliminary optimization results obtained.