InterPore2026

ML-Assisted Topology Optimization of Thermochemical Heat Storage Reactors

22 May 2026, 14:15

15m

Oral Presentation (MS15) Machine Learning in Porous Media MS15

Speaker

Dr Torben Prill (German Aerospace Center (DLR))

Description

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. Extensive research has been conducted on the subject for potential applications, including the capture of excess heat from industrial processes and the storage of energy in concentrated solar power plants. This study investigates TCES in the SrBr2 system, which offers a high energy capacity and near-perfect reversibility for medium temperature applications.
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, the powder agglomerates to a porous solid medium which expands and contracts during water uptake and release, respectively. This deformation of the bed may result in its detachment from the heat conducting surfaces, as illustrated in Figure 1, further inhibiting heat transport.
In a previous presentation [1], we presented the use of machine learning techniques to enhance heat transfer within the reactor with a non-deforming bed, which is achieved through the design of optimized heat-conducting structures. Due to the prohibitive time requirements of direct simulations, an artificial neural network surrogate model was constructed. The method entails the training of a neural network utilizing simulated data, which was generated with randomly generated fin structures. Subsequently, the trained network is used to predict the progression of the reaction over time. In this presentation, we will present the most recent findings on the use of neural networks for surrogate modeling, employing architectures based on the SinGAN [2]. Furthermore, the methodology for extending the surrogate model by a mechanical model for the deformation of the porous powder bed will be demonstrated. This enables the estimation of the powder bed/wall detachment, the resultant transport resistance, and the consequent impediment to reactor performance (see Fig. 2).
However, the primary emphasis of the presentation will be on topology optimization. The presentation will show the methodology employed to couple the surrogate model with topology optimization algorithms, which are based on the brute force, level-set (for an illustration see Fig. 3), and stochastic optimization methods [3]. These methods are employed to calculate optimal geometries for heat-conducting structures minimizing an objective function, which encodes the desired reactor performance characteristics. Finally, we will demonstrate how different objective functions give rise to different optimal geometries.