US-based Blue Wave AI Labs has deployed machine learning (ML) tools at power utility Constellation’s Peach Bottom Atomic Power Station and Limerick Generating Station. The project was part of a $6m initiative supported by the US Department of Energy (DOE) to lower NPP operating costs using artificial intelligence (AI) and ML technologies. Argonne National Laboratory and Brookhaven National Laboratory contributed to the project. The effort also leveraged 158,000 core hours across the Nuclear Science User Facilities high-performance computing systems.
Blue Wave AI Labs signed an agreement with Constellation in February 2022 to supply its products in support of Constellation’s fuel reload design and cycle management processes for the company’s entire fleet of boiling water reactors (BWRs). The long-term agreement covered Blue Wave’s cloud-based Nuclear-AI Platform components: Eigenvalue and MCO.ai forecasting products. Constellation had worked with Blue Wave over the previous four years to investigate the capabilities of ML and AI. Testing of the technology at Constellation’s plants began later in 2022.
Reactor operators depend on sensors to measure power generation, fuel consumption, and the overall state of the reactor with respect to operating limits. Over time, these sensors can become out of calibration and lose accuracy. If enough sensors stop working correctly, the reactor will reduce power or shut down as a precautionary measure, costing millions of dollars a day in lost generation revenue.
All three of Blue Wave’s AI tools ingested vast amounts of historical plant data to analyse and improve sensor measurements within the reactor core. In 2023, Blue Wave identified sensors at Limerick unit 2 that were suspected to be out of calibration. These sensors were taken offline, allowing the plant to continue operating safely while staying in compliance with its operating licence. During the next sensor calibration cycle, the plant operators were able to verify that sensors that were taken offline were giving incorrect readings due to miscalibration, as was predicted by Blue Wave’s tool.
The AI algorithms also improved engineers’ ability to predict how much fuel must be purchased and how to configure the fuel to generate the greatest amount of power while preserving margin to operating limits — another time-consuming and expensive process. Blue Wave estimates that the AI tools combined have saved Constellation more than $1.6m each year per reactor by reducing fuel costs, minimising reactor downtime, and reducing the staff time spent on analysis and planning.
“Constellation's collaboration with Blue Wave AI Labs has allowed us to use powerful machine learning tools to complement traditional engineering practices when designing innovative nuclear fuel products for our operating fleet,” said Jason Murphy, Vice President for Nuclear Fuels at Constellation. “Widespread adoption of these new tools will benefit nuclear reliability and cost-effectiveness.”
Jonathan Nistor, Chief Operating Officer at Blue Wave thanked DOE “for their vision, support, and funding that enabled the early advancements leading to the creation of this suite of AI tools,”. He added: "Coupling modern machine learning methods with vast amounts of nuclear power plant data has resulted in exceptional improvements in fuel utilisation and operational efficiency."
The three-year Blue Wave project builds on previous AI/ML work funded by DOE to help lower the cost of operating BWRs. Operations and maintenance costs make up nearly 70% of NPP generating costs and reducing these costs can help make existing reactors more cost-competitive in certain markets.
With many plants extending operations through the 2050s to maintain this reliable source of clean energy, using the latest advancements in AI and ML can help lower the cost, save on fuel, and reduce the amount of waste that is generated over the lifetime of the reactor while maintaining the same high standards for safety and security.
Constellation now plans to expand AI applications to additional reactors in its BWR fleet. Blue Wave projects that their technology could be deployed across all 32 of the nation’s BWRs within three years, saving the nuclear industry nearly $80m. The company is working to adapt these AI algorithms to support the US pressurised water reactor fleet, which comprises the remaining two-thirds of America’s nuclear energy generation.
Image: Graph showing a drifting sensor reading (yellow) at Limerick Generating Station in 2023. Constellation used Blue Wave AI Labs’ technology to take the malfunctioning sensor offline, avoiding a costly and unnecessary shutdown (courtesy of Blue Wave AI Labs)
