Towards the digital twin of a permanent magnet

Permanent magnets are a critical component of electric motors and generators in many applications, the most important of which are wind turbines and hybrid/electric vehicles. The rapid growth of these sectors has resulted in an increased demand for high performance Nd- Fe-B-based permanent magnets but the long-term sustainability of using global resources of rare earth elements like Nd, Dy and Tb at this high rate is questionable and there is therefore a clear need to develop a rare-earth-free permanent magnet. Such materials could replace certain types of Nd-Fe-B-based magnets in applications where lower performance is required, thus alleviating the pressure on rare earth resources. A digital twin is a set of information which fully describes the structure and properties of a physical object; any information which could be obtained by inspecting the physical object could also be obtained from its digital twin. In addition to the structure of the material, the digital twin of a permanent magnet must therefore also describe its magnetic state. This is highly challenging as the magnetic state of a material depends not only on its physical structure and magnetic properties but also on its magnetic and thermal history. The digital twin of a permanent magnet has the potential to play a vital role in the development of novel permanent magnets, and in real-time monitoring of the performance of magnets in applications. Obtaining the digital twin of a permanent magnet would therefore deliver important contributions to the digitalisation of materials science, environmental sustainability, clean energy and electromobility. The digital twin of a permanent magnet comprises experimental data and simulations on both the atomistic and the microscale. As the first step, the efforts in this project will be focussed at the microscale. The rare-earth-free magnet, MnAl-C, will be taken as a model system and an enhanced micromagnetic model will be developed. Advanced characterisation combined with magnetic domain images and magnetic measurements will form the basis for the simulations. A machine learning model will then be developed and data assimilation will be employed in order to reduce the offset between predicted and measured magnetic properties. The trained model represents the microscale component of the digital twin of a MnAl-C permanent magnet.