Artificial Intelligence for Phase Change Materials

The continuous increase in power density of electronic and energy storage systems, particularly lithium-ion batteries, imposes increasingly stringent thermal management requirements. Effective temperature control is essential to ensure the reliability, safety, and lifetime of these systems. Among the various passive cooling solutions, phase change materials (PCMs) have attracted considerable attention due to their ability to store and release large amounts of thermal energy in the form of latent heat during solid–liquid transitions. Their integration into passive thermal management systems helps to smooth temperature peaks and maintain stable operating conditions. Nevertheless, conventional PCMs suffer from inherently low thermal conductivity, which limits their performance under high heat flux conditions. To overcome this limitation, several enhancement techniques have been investigated, including the incorporation of highly conductive nanoparticles and the addition of metallic fins. However, the optimal design of such hybrid systems remains a challenging task due to the large number of interacting parameters and their strong nonlinear dependencies. In this context, the present PhD thesis proposes the use of artificial intelligence (AI) to model, interpret, and optimize the thermal behavior of PCM-based systems. The first stage focuses on developing a predictive model based on artificial neural networks (ANNs) to estimate the effective thermal conductivity of nanoparticle-enhanced phase change materials (NEPCMs) using literature-derived datasets. The model was further analyzed using the SHAP (SHapley Additive exPlanations) method, which enabled the identification of the most influential parameters and provided a physical interpretation of the nonlinear relationships between nanoparticle characteristics and thermal response. The second stage addresses the optimization of a passive thermal management system for cylindrical lithium-ion cells incorporating a PCM enriched with copper nanoparticles and equipped with aluminum fins. A three-dimensional parametric CFD study was conducted using ANSYS Fluent to evaluate the effects of key design and material parameters such as PCM thickness, fin number, nanoparticle concentration, and PCM type. A numerical campaign of 450 configurations combining these parameters was performed to generate a comprehensive database for training an ANN-based metamodel designed to predict the maximum cell temperature and the average liquid fraction of the PCM. This surrogate model was then coupled with a genetic algorithm (GA) within a deterministic design optimization (DDO) framework to identify the optimal configuration that minimizes the maximum temperature while maintaining a controlled partial melting of the PCM. The results confirm the efficiency of the ANN-GA coupling, which enables a significant reduction in computational cost compared to direct CFD-based optimization while ensuring high prediction accuracy. The interpretability analysis also provided valuable insights into the physical mechanisms governing the system’s thermal behavior. Overall, this thesis contributes to the advancement of AI-assisted modeling and optimization of PCM-based thermal systems. It demonstrates that the integration of artificial intelligence, numerical simulation, and evolutionary algorithms provides a robust and efficient framework for the design of next-generation passive thermal management systems. Future work could extend this methodology toward multi-objective and reliability-based design optimization (RBDO) approaches, incorporating uncertainties in material properties and operating conditions to achieve even more reliable and adaptive thermal solutions.