“Our recent work has demonstrated that AI can be a very useful tool to help nuclear reactors run more secure, safe, and more economical.”

Dr. Yang Liu
Assistant Professor, Texas A&M University

Challenge

Nuclear engineering has earned its reputation as one of the most rigorous and safety-driven disciplines in the world. That rigor is not accidental. It is the result of decades of hard lessons, disciplined processes, and systems designed to minimize risk in environments where failure is not an option.

“Nuclear systems have always demanded caution, because history has shown what’s at stake when systems fail.”

That same history, however, has also shaped a control paradigm that is fundamentally conservative by design. Traditional nuclear control systems prioritize determinism and human oversight, but they were never designed for modern computational workloads, particularly advanced analytics and artificial intelligence.

At the same time, the external context is changing rapidly:

• Global energy demand is rising
• Small Modular Reactors (SMRs) are emerging as a path to more flexible, deployable nuclear power
• AI has moved from theory into everyday engineering practice

For the next generation of nuclear reactors, this creates a tension. 

Small Modular Reactors demand highly reliable, cyber-secure, and automated control systems, as operators are rarely on site. Texas A&M used the COPA 500 to demonstrate how SMRs can be co-piloted by AI while maintaining industrial-grade safety and reliability. But introducing AI into nuclear control raises a fundamental question:

What kind of control system can support advanced AI while preserving the rigor, safety, and trust of nuclear engineering demands?

At Texas A&M, this question became the core research challenge:

Could an AI-assisted control system be deployed on a real physical nuclear facility, operate in real time, and meet the standards of reliability and safety required in a high-stakes nuclear environment?

“Essentially, if we want to demonstrate the system is working, we need to demonstrate on the physical facility.”
— Dr. Yang Liu, Assistant Professor, Texas A&M University

Solution

An open, production-grade control platform built on the COPA 500 controller

To explore this challenge, the Texas A&M research team set out to build a control environment capable of supporting AI-assisted operation of a small modular reactor test facility, without compromising the principles of nuclear safety.

Their approach rested on several key requirements:

1. Human-centered control, not automation for its own sake
The goal was never to replace operators. Instead, AI would augment human decision-making by increasing speed, foresight, and situational awareness.

2. Physics-informed AI, not generic models
Rather than using off-the-shelf AI tools, the team developed custom physics-informed subroutines, augmenting large language models with domain-specific nuclear physics.

“We are not saying that we want to replace the human. We are saying that the AI models that we are building will give the human more speed…What we are doing is augmenting the model with custom physics-informed subroutines… so you’re able to get it to predict the temperature, to tell you what reactivity you need to input into the system.”
— Zavier Ndum Ndum, Graduate Researcher, Texas A&M University

3. A control system capable of real-time, production-grade operation
Demonstrating AI in simulation was not enough. The system had to operate on a physical facility, with real sensors, real control loops, and real consequences.

“Essentially, if we want to demonstrate the system is working, we need to demonstrate on the physical facility.”
— Dr. Yang Liu, Assistant Professor, Texas A&M University

Building such a system from scratch proved difficult. Developing a new control platform while simultaneously ensuring reliability, safety, and cybersecurity created friction and slowed progress.

That changed with the introduction of the COPA 500 controller.

The COPA 500 provided an open, modern control foundation that allowed the research team to focus on their core work — nuclear physics and AI — rather than rebuilding control infrastructure.

“For high-stakes environments like nuclear engineering, the system has to be absolutely perfect. COPA 500 just changes the landscape entirely, allowing the researcher or end user to focus on what they do best.”
— Timothy Triplett, Senior Control Systems Engineer

Crucially, the COPA 500 was built using mature, industry-proven components, including technologies from established automation vendors. This allowed Texas A&M to combine:

• The reliability of traditional industrial control systems
• With the flexibility and compute capability of a modern IT-style architecture

“We’re getting the proven reliability of an industrial control system with the benefits of a modern IT architecture.”
— Bob Hagenau, Co-leader, COPA Team

The platform also enabled secure remote monitoring and control, continuous data supervision, and tight integration between AI models and real-time control logic — capabilities that are difficult or impossible to achieve on legacy nuclear control systems.

 Outcome

Demonstrated, AI-assisted nuclear control operating safely on a physical facility

Over the course of the past year, the Texas A&M team successfully demonstrated that AI-assisted nuclear control is not just theoretical. It can operate safely, reliably, and securely on a real system.

Key outcomes include: