Numerical modelling of active magnetic regenerator by using artificial neural networks
We have published a research paper on numerical modelling of multi-layered LaFeCoSi active magnetic regenerator by using artificial neural networks in collaboration with researchers from University of Salerno. The article was published in the journal Applied Thermal Engineering.
The paper is available here:
One of the main problems in the framework of magnetic refrigeration regards low adiabatic temperature changes that occur in the magnetocaloric materials, which limits the widespread application of this technology. There-fore, the major effort of researchers is focused on the development of multi-layer Active Magnetic Regenerators, which allows to enlarge the temperature span of a magnetic refrigerator. The use of numerical models can help to understand the feasibility of such application with less effort in comparison with the use of experimental fa-cilities. One of the main challenges in designing a numerical model of a multi-layer Active Magnetic Regenerator is the effective incorporation of the magnetocaloric data of different magnetocaloric materials, which are fundamental to correctly optimize the configuration of such a device with the aim to improve its performance. These data are usually obtained experimentally from different measurements and their integration into the numerical model is challenging. Therefore, this work proposes a modified multi-layer Active Magnetic Regenerator numerical model based on Artificial Neural Networks to integrate the magnetocaloric properties of magnetocaloric materials, allowing an easier and a more reliable implementation of the real properties of magnetocaloric materials. The proposed model was tested simulating a four-layer and a seven-layer LaFeCoSi Active Magnetic Regenerator. The use of Artificial Neural Networks to integrate the magnetocaloric properties of magnetocaloric materials into the multi-layer Active Magnetic Regenerator allowed to improve the accuracy of the model in comparison with the commonly used technique (i.e., Curie temperature shift method) when compared to the experimental data. Indeed, the maximum error of the maximum temperature span with zero thermal load was reduced from about 13 K to 6.6 K, for the seven-layer configuration, and from about 4.1 K to 1.0 K, for the four-layer configuration. Furthermore, the new model allows to obtain more reliable simulated data about the effectiveness of each layer of the Active Magnetic Regenerator, providing a more useful tool to discuss about the optimization of its configuration. The results shows that Artificial Neural Networks can be successfully applied for integrating the magnetocaloric properties of magnetocaloric materials into a multi-layer Active Magnetic Regenerator numerical model, improving its performance. They represent an innovative way to address the problem of including magnetocaloric properties into numerical models, opening the way to other possible Machine Learning techniques as alternatives to the usual Curie temperature shifting method used in the literature to date.