The Brain of Fusion Power Plants

The Brain of Fusion Power Plants

How to Tame the Energy of the Stars

Fusion research vs. the power industry

There are two very distinct “fusion worlds”. On the one hand, there are research machines — tokamaks, stellarators, and a handful of alternative designs — where fusion physics is refined and advanced. On the other hand, there are future fusion plants — facilities that will operate safely, quietly, and reliably, providing gigawatts of power to the grid. Both involve hot plasmas confined within magnetic traps, but the plasmas differ. Additionally, they are viewed as distinct entities from the control system’s perspective, which must coordinate hundreds of subsystems, manage highly nonlinear plasma, protect expensive equipment, and ensure reliable operation over decades.

We primarily focus on tokamak- or stellarator-based FPPs, as they are the best-known and most extensively studied type within magnetic confinement fusion machines and are considered the most promising for future energy production.

Control of research machines

Research fusion machines constantly face sudden plasma behaviors, including instabilities, edge-localized modes (ELMs), neoclassical tearing modes (NTMs), and major disruptions. In a tokamak, we address these issues in real time using fast magnetic coils for rapid position and shape control, along with high-power microwaves or neutral beams to target specific regions of the plasma.

Sometimes, experimenters aim to push limits by intentionally driving the plasma into new, occasionally unstable regimes to see what happens. Plasma disruptions are allowed in research machines. They’re annoying if they happen unintentionally and can damage components, but they also provide valuable physics data. The control system’s primary role is to safeguard the hardware and to capture data for analysis of what went wrong.

No fuel cycle, no electric power. For research tokamaks, fueling involves a gas tank connected to a vacuum vessel, plus some pellets. Heat is dissipated into the cooling water and simply heats the atmosphere. There’s no tritium breeding blanket, no turbine, no grid connection.

Stellarators behave differently from tokamaks. They don’t need a significant plasma current, so they’re inherently more stable against some tokamak-specific disruptions. Control is more about fine-tuning a complicated 3D configuration, controlling impurities, and sustaining long pulses rather than violently suppressing significant instabilities.

Either way, a research control system is optimized for flexibility and experimentation, not 24/7 reliability.

Control of fusion power plants

A fusion power plant (FPP) is designed for reliable, cost-effective electricity generation. Economic viability mandates on-site tritium breeding within blanket modules and strict inventory tracking for regulatory compliance. The FPP must also function reliably on the electrical grid, responding to load commands and enduring grid faults.

Consequently, the control problem shifts from refining and advancing physics to maintaining the stable, long-term operation of the complex system.

Plasma control becomes conservative and predictive. Critically, plasma disruptions, which could damage components and threaten economic viability, must be prevented. Under normal operation, the FPP’s plasma is not pushed to its limit. Engineers select safe operating ranges where dangerous instabilities are inherently unlikely. If instabilities such as NTMs, ELMs, or disruptions occur, the control system automatically reduces power, reshapes the plasma, or backs off. If a disruption still appears unavoidable, a dedicated mitigation system activates to slow it down and minimize damage (for example, by injecting gas or pellets in large quantities) — but this should be a rare emergency, not a routine occurrence.

The control system must incorporate the fuel cycle and tritium breeding. In an FPP, fueling is the visible component of a closed tritium loop that includes blankets that breed tritium from lithium, storage, purification, and delivery to injectors, and extraction systems that remove tritium from coolants and processing beds, all under strict inventory limits [1]. The control system must consider mass balance: if we change plasma density or operating power, we change how fast we consume and breed tritium.

The management of heat removal and power conversion becomes paramount. Instead of merely preventing wall overheating, an FPP must transport hundreds of megawatts of thermal power through intricate coolant loops, supply power to turbines and generators, and comply with grid on-ramp rates, frequency support, and fault ride-through [2]. These loops are governed similarly to a highly advanced fission or chemical processing facility.

The design prioritizes reliability and maintainability. It is not feasible to modify the software for every new operational regime or experiment through a simplified procedure. Consequently, software changes are infrequent and subject to rigorous verification. All critical components, including controllers, networks, and sensors, are fully redundant. Furthermore, the control system is required to coordinate remote maintenance robots to enable safe entry into high-radiation zones for tasks such as removing blanket modules and exchanging divertor cassettes [3].

A stellarator FPP keeps all of this, but plasma control is more placid. There’s less need for ultra-fast current-profile and vertical stability control. Instead, we focus on keeping the 3D configuration optimized and the wall/divertor in a comfortable long pulse. This characteristic aligns well with the operational goal of continuous, month-long runs, resembling the steady-state plant operation more closely than a pulsed device.

Integrated control system

FPP’s integrated control system (ICS) is not a singular, monolithic controller but rather a hierarchical structure composed of functionally discrete control layers (Fig. 1). Each layer possesses its own distinct temporal scale and operational mandate, yet all are integrated under centralized supervision. Essential controllers must incorporate redundancy to mitigate component failures, and communication networks must employ deterministic, robust industrial protocols coupled with clearly defined failover strategies. Hardware platforms should be built using established industrial control standards and selected based on criteria such as longevity, maintainability, and demonstrated reliability.

We can represent an FPP ICS as four main layers:

1. Plasma control and protection

This layer contains the plasma within its magnetic trap. It safeguards essential equipment by managing plasma shape, position, and vertical stability through ultra-fast control loops, predicting disruptions, and activating mitigation systems. It also quickly protects the magnets from quenches and overloads. At this layer, kinetic control of the burning plasma is maintained, ensuring the required temperature, density, and fusion power.

It builds on FPGAs and real-time CPUs with deterministic, high-bandwidth links to diagnostics and actuators. Simple, tightly verified algorithms are used here: proportional-integral-derivative or model-based controllers. Sometimes it is justified to use reinforcement learning (RL) or machine learning (ML) controllers, wrapped in conservative safety envelopes.

Minimal external interfaces: only receives commands from slower supervisory layer; doesn’t communicate directly to plant technological or external networks.

2. Plant control

This layer behaves a lot like the control system of a conventional nuclear or large chemical plant. It controls all the “big hardware” around the plasma, so-called technological systems: magnet and heating power supplies, cryogenic plant, primary and secondary coolant loops, heat exchangers, turbines and generators, vacuum systems and pumping, tritium production, and complete fuel cycle.

On this layer, ICS manages the full spectrum of operating modes, including startup, ramp-to-power, steady operation, ramp-down, shutdown, and maintenance cooldown. This layer provides comprehensive human-machine interfaces in the control room, extensive alarming, and data-trending capabilities.

It builds on industrial distributed control system (DCS) and programmable logic controller (PLC) platforms using robust, standard communication (e.g., time-sensitive Ethernet, fieldbuses). There is a strong separation between the nuclear island (blankets, vacuum vessel, coolant) and the conventional island (turbines, transformers).

3. Safety and interlock

The safety and interlock layer is a specialized, highly reliable redundant controller that regulators really care about. It monitors crucial variables: pressures, stresses, temperatures, radiation levels, tritium inventory, and containment boundaries. Enforces hard limits: trips heating systems or kills the plasma when limits are approached, and starts emergency cooling or isolation sequences. It provides defence-in-depth: if plasma control fails, interlocks step in; if interlocks fail, passive safety takes over.

The safety layer builds on special-certified equipment (safety integrity level-certified DCS/PLC), which uses elementary hardwired logic and is entirely separate from the operational or optimization codes. It uses multiple redundant channels and often diverse hardware vendors to avoid common-mode failures.

ITER’s control architecture already uses a layered approach (plasma control system + central interlock system + central safety system), and an FPP will extend that philosophy to cover the whole power plant and fuel cycle [4].

4. Supervisory

The supervisory layer constitutes the highest level, providing oversight for all preceding layers and directing the plant’s comprehensive operation in accordance with economic and operational imperatives. It chooses operating points to maximize net electric output, component lifetime, and tritium breeding margin. Supervisory also coordinates with the power grid to provide load-following or ancillary services and schedules ramp-ups and shutdowns. Maintenance planning uses data from sensors and digital twins to predict when components will need replacement and schedules remote handling campaigns.

This layer builds on conventional IT servers, cloud systems, and data lakes.

The path forward

As we move toward demonstration power plants and eventually commercial fusion, the control system will evolve from experimental tool to operational necessity. Here’s how the community is preparing:

Simulation and modeling: Developing plasma and plant simulators enables testing control algorithms before commissioning, reducing maintenance time and risk.

Cross-project collaboration: Sharing control algorithms, frameworks, and best practices across fusion projects speeds up progress. Open-source control frameworks offer a shared foundation.

Integration with FPP design: Control considerations must influence reactor design from the start, not be added later. STEP (UK’s Spherical Tokamak for Energy Production) exemplifies this approach, with plasma control challenges explicitly shaping design choices [5].

Workforce development: Building and operating fusion control systems requires expertise spanning plasma physics, control theory, software engineering, and industrial automation. Training programs must prepare the next generation of fusion engineers.

From lab to infrastructure

If we stacked the control diagrams of today’s research machines following an FPP, we’d notice that fast plasma control loops look vaguely similar — that’s where today’s research and development transfers most directly to a plant. Everything else around them explodes in complexity: fuel cycle, breeding blankets, coolant networks, diagnostics, grid interface, safety systems, and remote maintenance.

Research tokamaks and stellarators are undergoing continuous testing: they are highly dynamic, sensitive to control inputs, and subject to frequent modifications. An FPP, in contrast, requires extreme reliability, a deeply hierarchical control structure, and an uncompromising focus on safety and availability.

Designing the ICS with these realities in mind, we turn fusion labs into miniature artificial suns that have been operating nonstop for months and power cities.