New predictive models can improve reliability of fusion power plants | MIT News

Tokamaks are machines designed to hold and harness the power of the sun. These fusion machines use powerful magnets that contain hotter plasma than the core of the sun, and fuse the plasma atoms and release energy. If Tokamaks can run safely and efficiently, the machine can one day provide clean and unlimited fusion energy.

Today, many experimental Tokamaks are operating around the world, with more processes in progress. Most are small research machines designed to study how devices rotate plasma and utilize their energy. One of the challenges facing Tokamaks is how to safely and reliably turn off plasma currents circulating at a speed of 100 kilometers per second, with temperatures exceeding 100 million degrees Celsius.

Such a “ramp” is needed when the plasma becomes unstable. To prevent plasma from further destruction and potentially destroying the interior of the device, the operator gradually lowers the plasma current. But sometimes, rampage itself can destroy the stability of the plasma. In some machines, the slope causes abrasion and scars into the interior of the Tokamak – the slightest damage still requires a lot of time and resources to repair.

Now, MIT scientists have developed a method to predict how plasma in tokamak behaves in progressively. The team combined machine learning tools with physics-based plasma dynamics models to simulate the behavior of plasma and any instability that may arise as the plasma rises and closes the plasma. The researchers trained and tested new models of plasma data from experimental Tokamak in Switzerland. They found that the method quickly learned how plasma adjusted in different ways would develop. More importantly, this method achieves high accuracy using a relatively small amount of data. Given that every experiment run of Tokamak is expensive and therefore the quality data is limited, this training efficiency is promising.

New models, the team highlighted this week in an open visit Natural Communications Paper can improve the safety and reliability of future integrated power plants.

“To be a useful energy source, it has to be reliable,” said lead author Allen Wang. “To be reliable, we need to be good at managing our plasma.”

MIT co-authors of the study include Cristina Rea, PSFC’s lead research scientist and head of the Interference Team, as well as professors from the Laboratory of Information and Decision Systems (LIDS) OSWIN SO, CHARLES DAWSON and CHUCHU, as well as collaborators from Mark (Dan) Commanwealth Fusion Systems and Mark (Dan) Boyer from Wealth Fusion Systems and Swiss Plassma plasma plasma Center Center in Switzerland in Switzerland Land Land.

“Subtle Balance”

Tokamaks are the first experimental fusion equipment built in the Soviet Union in the 1950s. The device’s name comes from the abbreviation of Russia, which is translated into “annular chamber with magnetic coils”. As its name describes, the Tokamak is ring-shaped or donut-shaped and uses powerful magnets to rotate gas and to temperature and energy high enough that the atoms in the generated plasma can fuse and release energy.

Today, Tokamak’s experiments are relatively low in scale, and few people are close to generating the size and output required for safe, reliable, available energy. Experimental, low-energy Tokamak destruction is usually not a problem. However, as the fusion machine expands to the size of the grid scale, controlling higher energy plasma at all stages is crucial to maintaining the safe and efficient operation of the machine.

“Even during progressive times, uncontrolled plasma termination can create huge heat flux that damages the inner walls,” Wang noted. “Frequently, especially in the case of high-performance plasmas, the ramp can actually push the plasma toward some unstable limits. So, it’s a delicate balance. Now, there is a lot of attention to how instability is managed so that we can usually do these plasmas reliably and safely lower them. And in relatively few studies, and for being able to do it well.”

Let go of your pulse

Wang and his colleagues developed a model to predict how plasma behaves during Tokamak Rampdown. Wang said that while they can simply apply machine learning tools, such as neural networks, to learn signs of instability in plasma data, such tools “you will need a certain amount of data” to discern very subtle and short-lived changes in extremely high temperatures, high-energy plasmas.

Instead, the researchers paired neural networks with existing models that simulate plasma dynamics based on the basic rules of physics. Through this combination of machine learning and physics-based plasma simulations, the team found that only one hundred low-performance pulses, along with a small number of high-performance pulses, are enough to train and validate the new model.

The data they used for the new study came from TCV, Switzerland’s “Variable Configuration Tokamak” is run by EPFL’s Swiss Plasma Center (Swiss Federal Institute of Technology). TCV is a small experimental fusion experimental device for research purposes and is often used as a test bed for next-generation device solutions. Wang used data from hundreds of TCV plasma pulses, including the characteristics of the plasma, such as its temperature and energy during the rise, operation and gradual downward process of each pulse. He trained the new model on this data, then tested it and found that it accurately predicts the evolution of plasma because of the initial conditions for a specific Tokamak run.

The researchers also developed an algorithm that translates the predictions of the model into practical “trajectory” or plasmid and other plasma management instructions that can be automatically performed, such as adjusting magnets or temperatures to maintain plasma stability. They implemented the algorithm in multiple TCV runs and found that in some cases it can safely reduce the trajectory of plasma pulses, which is faster and without interference, compared to running without new methods.

“At some point, the plasma always disappears, but when the plasma disappears with high energy, we call it an interruption. Here we gradually reduce the energy to nothing,” Wang noted. “We did it a lot. We did it better in the whole way. So, we have statistical confidence in our making things better.”

This work is partially supported by the Commonwealth Fusion System (CFS), a MIT rotation designed to build the world’s first compact grid-level fusion power plant. The company is developing a demonstration Tokamak Sparc designed to produce net energy plasmas, meaning it should produce more energy than it takes to heat the plasma. Wang and his colleagues are working with CFS to investigate ways new prediction models and similar tools can better predict plasma behavior and prevent expensive disruption for safe and reliable fusion capabilities.

“We are trying to solve scientific problems to make integrations usually useful,” Wang said. “What we’re doing here is the beginning of a long journey. But I think we’ve made a good progress.”

Additional support for this study comes from the European Union framework, through the European Research and Training Programme, funded by the Swiss National Education, Research and Innovation Secretariat.