Machine learning is a type of artificial intelligence which helps in face recognition, language identification, translation. Apart from this, now machine learning can help in bringing the clean fusion energy, which helps in lighting the stars, to the Earth.
Now, a group of researchers from Princeton Plasma Physics Laboratory(PPPL) are taking the help of machine learning for creating a model to enable rapid control of plasma. It is the state of matter which consists of free electrons, ions and is responsible for fusion reactions. The biggest examples of such reactions are sun and many other stars which are themselves giant plasmic balls.
Scientists under the leadership of physicist Dan Boyer have trained neural networks which is the essential core of any machine learning software on the dataset produced by National Spherical Torus Experiment-Upgrade at PPPL. This model is quite accurate in reproducing the predictions of the behaviour exhibited by the particles produced by the neutral beam injection (NBI). It is used in fueling the NSTX-U plasmas and reaching upto million degrees of temperature.
The predictions are conventionally done with the help of NUBEAM. It is a program to incorporate the information about the impact made by the beam on the plasma. The calculations are performed a hundred times each second to determine the behaviour of the plasma. But since each calculation takes minutes to complete, researchers can know the result only after the experiment is over.
This problem is solved by the machine learning software as it reduces the time of calculation to less than 150 microseconds. As a result, the outcomes will be visible to the scientists during the experiment. The plasma control system will be able to make better decisions on how to control the injection of the beam for efficient performance.
With such fast evaluations, the operators will be able to make the needed adjustments for the experiments. Boyer, who is also the principal author on a paper of Nuclear Fusion commented that the rapid modelling capacities can guide the operators in changing the NBI settings for the next experiment.
Along with scientist Stan Kaye, he generated a database with NUBEAM results for a specific range of conditions resembling the ones during the initial NSTX-U operations. This was used by the scientists in training a neural network for predicting the effects of the beam on plasma like heating. After that, it was implemented by software engineers on a computer for controlling the experiment and finding out the calculation time. Scientists plan on expanding this modelling approach for other plasma phenomena.