tLOSS: a collaborative machine learning platform for predicting AC losses in HTS devices

This work describes an open access platform for collaborative data-driven modelling of AC losses in high-temperature superconducting (HTS) devices, as opposed to computationally intensive, time-consuming numerical methods. The platform is being developed in the frame of the Portuguese project tLOSS. HTS devices are usually modelled and simulated by the Finite Element Method (FEM), due to its accuracy and ability to address multiphysics problems. Yet, the non-linearity of HTS properties or huge width-to- thickness ratios, makes FEM extremely computationally intensive, leading to unreasonable processing times in the optimization of devices, when, e.g., thousands of configurations need to be simulated and assessed. This is still a major impediment to HTS technologies dissemination. There is an opportunity for the development of data-driven based approaches, from the field of Artificial Intelligence (AI), for modelling and simulating the behaviour of HTS devices. Data-driven approaches use data (from experiments and/or simulation) to learn patterns in it. The built computational models can be used for obtaining predictions of distinct conditions instantly. The ability of these approaches to build generalised models with adequate accuracy depends on the availability of large amounts of data from diverse devices and operating conditions.