Automated labelling and correlation analysis of diagnostic signals from ADITYA tokamak for developing AI-based disruption mitigation systems

Abstract

AI/ML-based data-driven methodologies are becoming increasingly effective in understanding and predicting plasma disruption in tokamaks by identifying critical signatures present in various diagnostic signals obtained from tokamaks. A high-performance ML-based disruption predictor requires large accurately labelled data. Until now, plasma shots from the ADITYA tokamak have primarily been classified (labelled) as disruptive or non-disruptive manually. Here, we present three computational techniques, namely the Sorted-array approach, the Interval comparison approach and the Threshold-Straight line method for automatic labelling of the ADITYA shots as disruptive or non-disruptive based on the plasma current dropdown time. Statistical analysis and comparison between automatic labelling and manual labelling indicate the promising potential of the proposed techniques. A correlation analysis is also conducted by incorporating plasma diagnostics such as Plasma current, Loop voltage, Bolometer, Mirnov, Hard X-ray, Soft–X-ray, Radiation from Hydrogen-alpha, ionised oxygen and ionised carbon. This comprehensive study offers valuable insights into diverse physical phenomena associated with disruptions. Furthermore, correlation analysis based on current quench time highlights the significance of different diagnostics in providing distinct signatures related to plasma disruption. The insights obtained from this work can play a pivotal role in advancing the development of data-driven disruption prediction systems for ADITYA tokamak.