AI Tools

We are committed to advancing open and collaborative research. Several of our AI-based tools will be released as open-source resources to support the wider superconducting research community.

Research Background and Significance

Artificial intelligence offers new approaches for advancing superconducting research. Traditional physical models often have difficulty describing the complex and nonlinear relationships among temperature, magnetic field, and material properties. By learning directly from experimental data, AI can capture these interactions more effectively, improving prediction accuracy and reducing the need for extensive measurements. Integrating AI into superconducting research enables faster design, better performance optimisation, and more efficient use of experimental resources.

Research Methods

 

We apply the latest AI techniques to address key classification and regression problems in superconductivity. By combining large datasets with advanced algorithms, our research focuses on building models that can automatically learn the underlying relationships between physical variables, predict material properties, and identify patterns in multi-sensor data.

Our approaches include deep learning architectures such as transformer and graph-based models, as well as interpretable machine learning methods for analysing complex experimental datasets. By integrating domain knowledge with data-driven learning, we aim to create reliable, generalisable, and physically consistent models that can assist in material discovery, fault diagnosis, and performance prediction across a wide range of superconducting applications.

1

Research Outcomes

Our work has led to several AI-based tools that assist in superconducting research and engineering.One representative achievement is the Jc Prediction Platform, which uses a transformer-based deep learning model to predict the critical current density of high-temperature superconducting tapes across a wide range of temperatures and magnetic fields. 

This article was published on