Magnetic refrigeration is a highly investigated topic with great potential in cooling and heating applications. Its promising efficiency and environmentally friendly operation make it an attractive and resilient solution. The modeling of such thermodynamic systems is a central research focus, due to the high costs of manufacturing and testing real-life designs. To achieve the maximum capabilities of magnetic refrigerators, the tuning of their design parameters is essential. The complexity of magnetic refrigeration applications results in a high-dimensional design space that is difficult to solve analytically.

In this thesis, a surrogate model-based optimization framework was developed and validated for near-room-temperature Active Magnetic Regenerators (AMRs) that balance second-law efficiency against magnet mass. The framework combines a Multi-layer Perceptron (MLP) surrogate model with a genetic algorithm to efficiently explore a design space defined by more than eight parameters: length, width, and height of the regenerator, number of magnetocaloric material (MCM) layers, individual layer thicknesses, Curie temperature per layer, porosity of MCM layers, applied magnetic field, and void spaces. The surrogate model approximates a computationally expensive 1-D thermodynamic AMR model, reducing evaluation time and paving the way to multi-dimensional complex optimization AMR problems. The results demonstrate that with sequential model-based optimization (SMBO), the model can predict with higher accuracy the efficiency of each design, eventually leading to various design configurations with high efficiency and within the desired cost limits. The effect of SMBO is captured every training round by the mean absolute error (MAE) metric.

The optimal AMR configuration identified through this framework for a 15 K temperature span features 8 MCM layers with Curie temperatures spanning 274.8 K to 291.8 K, a regenerator geometry of 68 mm × 25 mm × 29 mm (length × width × height), 23% porosity for the MCM blocks, and operates under a 1.3 T magnetic field. This configuration represents a practical balance between hermodynamic performance and manufacturing feasibility for near-room-temperature magnetic refrigeration applications.