journal article·2026-09Open access

A multimodal machine learning approach for the integrated design of active magnetic regenerators with coordinated multilayer material layout and thermo-fluidic transfer characteristics

Asaya Fujita, Shun Muroga, Hiromi Minemawari, Takeo Yamada, Masashi Tsubaki, Kimihiro Ozaki, Kenji Hata

Materials & Design· Volume 269

Abstract

A multimodal machine learning (ML) framework is proposed for the holistic design of active magnetic regenerator (AMR). This approach comprehensively captures not only specialized assessments such as the suitability of magnetocaloric (MC) materials but also correlations among a multi-layer layout, operational parameters and the environmental temperature ( T rt ) and thermal load ( P n ). Dimensionally reduced MC feature vectors were successfully obtained from differential scanning calorimetry curves using a uniform manifold approximation and projection (UMAP). A multimodal dataset was constructed by classifying temperature span ( T span ) data across four-layer beds composed of La(Fe,Si) 13 variants under various operating parameters, treating the UMAP vectors as labels. The impact of T rt and P n was also incorporated as annotations for classification. The best predictive accuracy was achieved by a Gaussian process regression. A large no-load T span of 16.9 °C at 0.8 T, which was overlooked during the grid sampling, was found under conditions optimized for the trained four-layer layouts by the constructed ML model. Predictive optimization for untrained layouts utilizing virtual material selections constructed by combining the UMAP vectors was also effective, and the maximized no-load T span = 17.1 °C was demonstrated around the predicted conditions for a unique four-layer layout, that breaks heuristic approaches in material selection.
Source: openalex_inverted_index · Quality: baseline

Research context

OpenAlex topics: Magnetic and transport properties of perovskites and related materials · Magnetic Properties of Alloys · Advancements in Solid Oxide Fuel Cells

SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.

References (59)

Show source-provided references
1. Impacts on global ozone and climate from use and emission of 2, 2-dichloro-1, 1, 1-trifluoroethane (HCFC-123)10.1023/A:1005496900220 ↗
2. New refrigerants and system configurations for vapor-compression refrigeration10.1126/science.abe3692 ↗
3. Ozone depletion and global warming: Case for the use of natural refrigerant - a review10.1016/j.rser.2012.10.008 ↗
4. Vapor compression cycle: A state-of-the-art review on cycle improvements10.3390/cleantechnol5020030 ↗
5. A hot future for cool materials10.1007/s11708-022-0854-4 ↗
6. Present and future caloric refrigeration and heat-pump technologies10.1016/j.ijrefrig.2015.06.008 ↗
7. Caloric effects in ferroic materials: new concepts for cooling10.1002/ente.201800201 ↗
8. From the discovery of the giant magnetocaloric effect to the development of high‐power‐density systems10.1002/admt.202500545 ↗
9. Polaris: from laboratory prototypes to market-ready sustainable magnetic beverage coolers
10. J.A. Barclay, W.A. Steyert, Active magnetic regenerator, US Patent 4,332,135, 1982.
11. Magnetic heat pumping near room-temperature10.1063/1.323176 ↗
12. Description and performance of a near-room temperature magnetic refrigerator
13. Review on the developments of active magnetic regenerator refrigerators–Evaluated by performance10.1016/j.rser.2020.110247 ↗
14. The giant magnetocaloric effect of optimally prepared Gd 5 Si 2 Ge 210.1063/1.1558210 ↗
15. Energy applications of magnetocaloric materials10.1002/aenm.201903741 ↗
16. Material-based figure of merit for caloric materials10.1063/1.5004173 ↗
17. Advanced magnetocaloric materials: What does the future hold?10.1016/j.ijrefrig.2006.03.020 ↗
18. Selection of magnetic materials for an active magnetic regenerative refrigerator10.1016/0925-8388(94)90241-0 ↗
19. Sensitivity study of multi-layer active magnetic regenerators using first order magnetocaloric material La(Fe,Mn,Si)H10.1063/1.4923356 ↗
20. Study of multi-layer active magnetic regenerators using magnetocaloric materials with first and second order phase transition10.1088/0022-3727/49/34/345001 ↗
21. Interrelationship between flow profiles and the magnetic waveform and their influence on the performance of first-order active magnetic regenerators10.1016/j.applthermaleng.2022.119581 ↗
22. Experimental validation of a multiphysics modeling for a magnetocaloric bench10.1016/j.applthermaleng.2022.118415 ↗
23. Magnetic-field-induced structural phase transition in Gd 5 (S i 1.8 Ge 2.2)10.1103/PhysRevB.58.R14721 ↗
24. Caloric materials near ferroic phase transitions10.1038/nmat3951 ↗
25. Maximum performance of an active magnetic regenerator10.1063/5.0067751 ↗
26. Hysteresis design of magnetocaloric materials—from basic mechanisms to applications10.1002/ente.201800264 ↗
27. Unravelling entropic features of La(Fe,Si)13Hy-based refrigerant inside an active magnetic regenerative bed component via magnetic observations10.1016/j.scriptamat.2024.116110 ↗
28. Demonstration of hybrid regenerative cooling using active magnetic regenerative La (Fe, Si) 13 compounds and passive regenerative materials10.1016/j.jallcom.2025.179237 ↗
29. On the S (T) diagram of magnetocaloric materials with first-order transition: Kinetic and cyclic effects of Heusler alloys10.1016/j.actamat.2016.01.052 ↗
30. Large reversible entropy change at the inverse magnetocaloric effect in Ni-Co-Mn-Ga-In magnetic shape memory alloys10.1063/1.4808340 ↗
31. Caloric devices: A review on numerical modeling and optimization strategies10.1002/er.7023 ↗
32. An open source DNS solver for the simulation of active Magnetocaloric Regenerative cycles10.1016/j.applthermaleng.2018.06.007 ↗
33. High-throughput search for caloric materials: The CaloriCool approach10.1088/1361-6463/aa9bd0 ↗
34. General approach for efficient prediction of refrigeration performance in caloric materials10.1103/PhysRevApplied.22.014036 ↗
35. Application of machine learning in magnetocaloric materials: A review
36. Inverse design of materials that exhibit the magnetocaloric effect by text-mining of the scientific literature and generative deep learning10.1021/acs.chemmater.1c01368 ↗
37. Machine learning technique for prediction of magnetocaloric effect in La (Fe, Si/Al) 13-based materials10.1088/1674-1056/27/6/067503 ↗
38. Machine learning-aided design of LaCe (Fe, Mn, Si) 13H-type magnetocaloric materials for room-temperature applications10.1016/j.jallcom.2024.175746 ↗
39. Accelerated design for magnetocaloric performance in Mn-Fe-P-Si compounds using machine learning10.1016/j.jmst.2021.03.082 ↗
40. Giant room-temperature magnetocaloric effect MM′ X alloys explored by machine learning
41. Review of multi-physics modeling on the active magnetic regenerative refrigeration10.3390/mca26020047 ↗
42. The use of the first and of the second order phase magnetic transition alloys for an AMR refrigerator at room temperature: a numerical analysis of the energy performances10.1016/j.enconman.2013.02.006 ↗
43. Evaluating the effect of magnetocaloric properties on magnetic refrigeration performance10.1063/1.3525647 ↗
44. Phenomenological model for a first-order magnetocaloric material10.1016/j.ijrefrig.2019.10.003 ↗
45. Numerical study on the multi-layered magnetocaloric regenerators10.1016/j.applthermaleng.2021.118001 ↗
46. Influence of layering and curie temperature uncertainty on the performance of multilayer active magnetic regenerators10.1002/adfm.202424282 ↗
47. A comprehensive and versatile multimodal deep‐learning approach for predicting diverse properties of advanced materials10.1002/advs.202302508 ↗
48. Tabular two-dimensional correlation analysis for multifaceted characterization data10.1177/00037028241228865 ↗
49. Rational design of printable carbon nanotube transparent conductive films via data‐driven and mechanistic insights
50. Virtual experimentations by deep learning on tangible materials10.1038/s43246-021-00195-2 ↗
51. S.S. Kushwaha, M. Fuentes, A multimodal prototypical approach for unsupervised sound classification, arXiv preprint arXiv:2306.12300 (2023).10.21437/Interspeech.2023-982 ↗
52. Unsupervised deep metric learning via orthogonality based probabilistic loss10.1109/TAI.2020.3026982 ↗
53. Principal component analysis: A review and recent developments, Philosophical transactions of the royal society A: Mathematical
54. Umap: Uniform manifold approximation and projection10.21105/joss.00861 ↗
55. L. McInnes, J. Healy, J. Melville, Umap: Uniform manifold approximation and projection for dimension reduction, arXiv preprint arXiv:1802.03426 (2018).10.21105/joss.00861 ↗
56. L. Van der Maaten, G. Hinton, Visualizing data using t-SNE, Journal of machine learning research 9(11) (2008).
57. Automated elucidation of crystal and electronic structures in boron nitride from X-ray absorption spectra using uniform manifold approximation and projection10.1038/s41598-025-18580-z ↗
58. A tutorial on Gaussian process regression: Modelling, exploring, and exploiting functions10.1016/j.jmp.2018.03.001 ↗
59. The influence of the solid thermal conductivity on active magnetic regenerators10.1088/0022-3727/45/14/145001 ↗