ST40 electromagnetic predictive studies supported by machine learning applied to experimental database.

Area of Science:

• Nuclear Engineering
• Plasma Physics
• Machine Learning Applications

Background:

• Nuclear fusion power plants face risks from plasma disruptions due to high stored energy.
• Mitigating and predicting these disruptions is crucial for machine integrity and availability.
• Current methods for disruption characterization and mitigation are under active development.

Purpose of the Study:

• Investigate the causes and effects of plasma disruptions in the ST40 device.
• Develop preliminary predictive analyses of ST40 plasma scenarios using machine learning.
• Map the controllable operational space concerning plasma displacement and internal parameters.

Main Methods:

• Utilized machine learning techniques on an experimental database of ST40 plasma pulses (2021-2022 campaign).
• Classified common features within disruptions and mapped operational parameters.
• Validated machine learning classification by benchmarking numerical plasma dynamics reconstruction with experimental diagnostics data.
• Performed predictive simulations of plasma column displacement using MAXFEA.

Main Results:

• Identified common features associated with plasma disruptions in ST40.
• Mapped controllable operational parameters related to plasma displacement and internal states.
• Successfully predicted disrupted plasma configurations and simulated vertical displacement.

Conclusions:

• Machine learning provides a powerful tool for understanding and predicting plasma disruptions in fusion devices.
• The findings offer insights for the next ST40 experimental campaign and future fusion device designs.
• Accurate prediction and mitigation of disruptions are essential for the viability of fusion power plants.