Faster Fusion Reactor Calculations Thanks to AI
Fusion reactor technologies are well positioned to contribute to our future electricity needs in a safe and sustainable manner. Numerical models can provide researchers with information about the behavior of the fusion plasmaas well as valuable insights into the effectiveness of reactor design and operation. However, modeling the large number of plasma interactions requires a number of specialized models that are not fast enough to provide data on reactor design and operation.
Aaron Ho from the Fusion Science and Technology group in the Applied Physics Department at Eindhoven University of Technology has investigated the use of machine learning approaches to accelerate the numerical simulation of turbulent nuclear transport in plasma. Ho defended his thesis on March 17th.
The ultimate goal of fusion reactor research is to generate a net power gain in an economical manner. To achieve this goal, large, complicated devices were designed. However, as these devices become more complex, it becomes increasingly important to take a predict-first approach to their operation. This reduces operational inefficiencies and protects the device from serious damage.
To simulate such a system, models are required that can capture all the relevant phenomena in a fusion device, are accurate enough that predictions can be used to make reliable design decisions, and are fast enough to quickly find workable solutions.
Model based on neural networks
For his PhD thesis, Aaron Ho developed a model to meet these criteria using a model based on neural networks. This technique effectively allows a model to maintain both speed and speed accuracy at the expense of data collection. The numerical approach was applied to a reduced-order turbulence model, QuaLiKiz, which predicts the amounts of plasma transport caused by microturbulence. This particular phenomenon is the dominant transport mechanism in tokamak plasma devices. Unfortunately, its calculation is also the limiting speed factor in current tokamak plasma modeling.
Ho successfully trained a neural network model with QuaLiKiz scores while using experimental data as training input. The resulting neural network was then coupled into a larger integrated modeling framework, JINTRAC, to simulate the core of the plasma device.
The simulation time has been reduced from 217 hours to just two hours
Neural network performance was assessed by replacing the original QuaLiKiz model with Ho’s neural network model and comparing the results. Compared to the original QuaLiKiz model, the Ho model took into account additional physics models, duplicated the results with an accuracy of 10% and reduced the simulation time from 217 hours on 16 cores to two hours on a single core.
In order to test the effectiveness of the model outside of the training data, the model was used in an optimization exercise using the coupled system in a plasma ramp-up scenario as a proof-of-principle. This study provided a deeper understanding of the physics behind the experimental observations and highlighted the usefulness of faster, more accurate, and more detailed plasma models.
Finally, Ho suggests expanding the model for further applications such as controllers or experimental design. He also recommends extending the technique to other physical models, as it has been found that the turbulent transport predictions are no longer the limiting factor. This would further improve the applicability of the integrated model in iterative applications and allow the necessary validation efforts to bring its capabilities closer to a truly predictive model.