In 2023, Idaho National Laboratory (INL) and Idaho State University unveiled the industry’s first near real-time digital twin of a nuclear reactor. The virtual replica of the 5-Wth AGN-201 research reactor uses cloud-based monitoring and advanced machine learning to predict reactor performance dynamically.
Digital twins in the nuclear sector—real-time virtual models of reactors—have long-held promises to enhance safety, efficiency, and precision by allowing operators to monitor and predict system behaviors without physical intervention.
 |
|
1. According to the U.S. Nuclear Regulatory Commission (NRC), a digital twin, as part of a digital twin system, is a virtual representation of an entity, process, or system synchronized at a frequency and fidelity sufficient to maintain state concurrence. Essentially, it leverages various types of models, data, and frameworks to produce knowledge and insights about the represented entity, process, or system to fulfill an intended purpose. Source: NRC |
According to regulators, digital twins (Figure 1) could offer a real-time, synchronized view of plant conditions, supporting swift, data-driven responses to anomalies and reinforcing operational safety and efficiency. As key benefits to the nuclear industry, digital twins hold the potential to streamline diagnostics and preventive measures, furnishing operators with more insights for efficient, informed decision-making. Still, developing effective digital twins has been challenging, owing to the sector’s unique demand for secure, reliable data integration, complex sensor networks, and stringent regulatory compliance—all of which require high levels of accuracy and coordination.
Since 2020, Idaho National Laboratory (INL), a hotbed for nuclear innovation, has pursued several digital twin initiatives, exploring models that simulate reactor misuse scenarios, enhance safeguards in advanced reactors like sodium fast reactors and pebble-bed reactors, and other efforts that sought to optimize sensor placement and data flows for real-time monitoring and security compliance in new reactor designs. To add to that foundational research, in 2023, Ryan Stewart, an INL digital engineer, recommended using the Aerojet-General Nucleonics nuclear reactor (AGN-201) for some of the team’s planned demonstrations.
A Groundbreaking Replica
Built in the 1960s at Idaho State University’s (ISU’s) College of Science and Engineering in Pocatello, Idaho, the compact 5-Wth AGN-201 reactor—one of only five still operating around the world—has served as a crucial teaching and research tool for scientists and students. The reactor’s core consists of nine fuel disks made from enriched uranium dioxide mixed with polyethylene, forming a compact 9.45-inch by 10.08-inch reactor core. Safety is central to its design: a polystyrene thermal fuse separates core sections if temperatures rise, halting the reaction safely. The reactor also features control rods and a central irradiation facility. As INL explains, the reactor’s simplified structure provided an ideal backdrop for creating a digital model without the complexity and risks associated with larger commercial reactors.
In This Issue
Full issueStewart’s recommendation evolved into a full collaboration between INL and ISU, bringing together researchers, engineers, and ISU nuclear engineering students with the intent to create an integrated, cloud-based digital twin. By mid-2023, the team had developed a sophisticated, real-time digital twin that captures reactor data, processes it through advanced machine learning models, and streams it securely via cloud infrastructure. The digital twin has since been put to work in a range of operational scenarios, allowing researchers to monitor and predict reactor performance in real-time continuously.
INL says the digital twin is “a first in the industry.” The experiment has served as a “key step in building a digital twin of a larger reactor system and has helped highlight many potential pitfalls and problems that such an endeavor might face,” project researchers added. It has “also shown the great promise that a cloud-first approach has when creating digital twins.”
The Technology Behind the Twin
According to INL, the AGN-201 digital twin is composed of several components. Essentially, it is powered by a robust data architecture that connects the physical reactor to its digital counterpart to allow for real-time, seamless data flow. To replicate the physical AGN-201’s characteristics digitally, the twin includes models of control rods, thermal safety mechanisms, and key temperature points.
At the core of the setup is the Data Acquisition System (DAS), which serves to monitor the physical reactor parameters continuously. That data feeds into Jester, an open-source tool developed by INL with an extensive plugin system, “which allows it to work on a myriad of different systems and with different file types,” the lab explains.
Jester then transfers data from sensor systems to the DeepLynx, the backbone of the digital twin. DeepLynx is an open-source data lake hosted on a Microsoft Azure for Government cloud platform, which allows it to connect all other processes and software developed for the twin. The cloud platform also leverages the Azure Kubernetes system (AKS) to deploy and manage infrastructure.
“Unlike other data warehouses, DeepLynx users can use an ontology [a structured framework] to custom-define how their data will be represented,” INL notes. “DeepLynx enables users to store their data in a graph-like format, ensuring that connections between data can be easily seen and understood. This data lake also allows users to store tabular, or time-series, data such as the data coming in off of sensors and [internet of things].” However, INL adds, “It is important to note that while fast, this process is not considered ‘real-time’ but ‘near real-time’ due to the network latency between each process communicating over the web.”
One of the project’s key achievements is to have developed programs to enable visualization of the digital twin, which allows its operators to monitor and easily identify anomalies and issues in reactor operation. One beta desktop application, OperatorUI, is designed to provide insights into reactor function by incorporating the machine learning data ingested by DeepLynx and then displaying it in an easy-to-digest format. Visualization tools also leverage Microsoft HoloLens, an augmented reality (AR)/mixed reality (MR) headset with Microsoft positional tracking technology. Paired with Unity’s cross-platform game engine, the AR interactive view of the AGN-201 reactor allows operators to visualize real-time data flows in a fully immersive environment. The interface, notably, also integrates QR code scanning for precise alignment of digital data with physical components, essential for design reviews, ergonomics, and operational pathway testing.
Another core strength of the AGN-201 digital twin is its sophisticated machine learning and artificial intelligence (AI) detection programs, which pull live data from DeepLynx to enable real-time analysis and predictive insights. Tools like Papermill—an open-source utility for executing parameterized Jupyter Notebooks—streamline workflows by allowing automated, efficient processing without manual integration. Meanwhile, DuckDB, an in-process SQL OLAP database management system, serves as a temporary in-memory database to allow fast, columnar data storage optimized for analytics.
Overall, INL notes that two main constraints impacted the digital twin’s implementation. First, given data transmission limits over HTTPS and security firewalls, the digital twin operates in near real-time using batch processing rather than continuous data streaming from the AGN-201 reactor. Second, while INL chose to use Rust, a newer general-purpose programming language, given Rust’s fast runtime and secure memory handling, it required substantial debugging to establish reliable interoperability with Python for managing and analyzing reactor data in DeepLynx. The lessons “will continue to advance the understanding of how digital twins are implemented,” INL says.
Testing and Validation
To validate the digital twin’s (DT’s) security and system integrity, ISU operators have, over the past year, run the AGN-201 reactor under varied conditions while keeping the INL digital twin team in the dark. The approach has given INL’s digital twin team opportunities to independently evaluate the twin’s ability to detect off-normal operations.
The ISU and INL teams have also conducted a “Red versus Blue” test to evaluate the digital twin’s ability to detect and respond to potential misuse or nefarious activity. As part of the exercise, ISU reactor operators (as the red team) simulated potential intrusions by introducing hidden operational changes, like inserting a polyethylene rod or cadmium foil into the central irradiation facility to mimic reactor tampering. The INL digital twin team (the blue team) used the digital twin’s anomaly detection system to monitor these changes and respond accordingly.
 |
INL’s team says that to enhance the digital twin, it implemented an Isolation Forest machine learning algorithm during the test. “This algorithm is particularly good at spotting unusual patterns in data without the need for previously identified examples of such anomalies. The team first standardized the data, then fine-tuned the Isolation Forest model to the specific characteristics of the DT’s data,” it notes. The system successfully flagged each simulated anomaly and, in effect, showcased the digital twin’s reliability in identifying real-time security threats.
More work continues, however. Future development of the AGN-201 digital twin will focus on enhancing the reactor physics surrogate model using the Point Kinetics Equations-Surrogate Model (PKE-SM). That improvement should enable more precise tracking of power changes over time. Other efforts will address refining detection thresholds to identify even minor perturbations to ensure they stand out from normal operational variability, INL says.
For now, the digital twin has proven “a resounding success, despite the difficulties and hurdles in both operations and implementation,” INL says. “This work will lay the foundation for future digital twins in the Idaho National Laboratory’s sphere of influence and has proven that they have the capability to handle larger projects.”
—Sonal Patel is a POWER senior editor.
