Is this your challenge?

In reactor operation and design, we have seen huge improvements over the past few years with tools developed as part of Industry 4.0. Digital twins, IIOT’s and Virtual Reality are all helping towards better designs. Unfortunately, there are still some areas where old fashioned methods are still largely in place. One of them is fuel reloading.

One of them is fuel reloading. Typically, in PWR’s and BWR’s, a fuel cycle lasts approximately 24 months. Once a plant is under an outage, only a third of the reactor core is changed and replaced with fresh fuel. Therefore, a bundle of fuel from its initial loading to unloading stays in a core for three fuel cycles (i.e., 6 years).

So, every 24 months, the planning team must account for a fresh array of fuel, with specific characteristics such as the Uranium distribution, enrichment but also numbers and positions of Gadolinium rods (absorbers). All these elements being considered to ensure the reactor will deliver the planned energy yet operate within set limits whilst protecting the equipment exposed to the neutron flux (fluence) and other physical/chemical phenomenon.

The cost of fuel for one cycle is around $150 million but might goes over that value because of deviations between planning and operation. All to say, navigating between technical considerations, economical elements but also regulatory requirements is a challenge. Any tool able to optimise core loading will allow to reduce costs in the range of millions of dollars whilst having long term benefits on the equipment within the reactor. Efficiently solving hurdles, new AI tools specifically setup for the nuclear industry are now available and have proven to offer ways to drastically improve core loading, optimise planning and unlock savings.

The main reload design problems

For PWR and BWR, protecting the integrity of the fuel is essential. Deviation between measured performance and design predictions can lead to operational challenges, such as unplanned derated conditions, premature shutdown, or increased fuel costs by loading more fuel than required for targeted energy production.  There are three main issues:

• Inability to predict moisture carryover (MCO): In a BWR, steam generated from the reactor core goes through a moisture separator to avoid damaging the turbines and other plant components. However, the separators are not 100% efficient and droplets are seen after the separators. If the MCO exceeds certain design specifications, it would lead to an accelerated rate of erosion of the main turbines, and the main steam isolation valve would see as well an accelerated erosion, leading to safety concerns. Moreover, excess moisture can also carry ⁶⁰Co into the plant making clean-up more expensive. As MCO has been nearly impossible to predict with conventional methods, the only option for the plant operator is to derate when MCO is too high, leading to revenue losses as there is a lower output.

• Unpredictability of K_eff (Eigenvalue): K_eff gives you the criticality of your reactor. It is an essential element when planning a core loading as it tells you the energy capability of your new load. If you do not know it exactly, you might burn more fuel (or less) than expected – hence impacting the economics. Historically, the real K_eff is extremely hard to precisely estimate.

• On-line/Off-Line Thermal Limits Bias Uncertainty: Accurate predictions of core-wide and local behaviour are crucial to assuring that targeted margins to the operating limits are maintained. Thermal Limits ensure the integrity of the fuel cladding. Historically, inability to accurately predict online thermal limits from offline methods has challenged core design and cycle management. Between core design and core monitoring, there is a non-negligeable bias as one is using “off-line” nuclear methods and the other one is using “on-line” methods through feedback from the in-core nuclear instrumentation. Because of the uncertainty in conventional methods, excess margin is engineered in leading to increased fuel costs. Accurately predicting the thermal limits in a BWR is challenging as this technology works in a complex two-phase flow.

inTechBrew’s insight

Blue Wave AI Labs is not just “another AI company from the Silicon Valley”. First, they are not from the Valley. Second, they decided to take a different route compared to other AI companies. Whereas most AI companies offer a one-size-fits-all platform that requires clients to sort through their own data to come up with a general solution to a general problem, Blue Wave AI Labs are AI-savvy scientists and engineers tailoring solutions to answer complex industrial problems for each of their clients.

Since 2016, Blue Wave AI Labs has been developing the next generation of nuclear power analytics, working to solve challenges for nuclear core loading, developing different tools to enhance diagnostic and predictive capabilities when designing a new core. Their blending of AI models with reactor physics expertise has already saved over $120 million dollars for their customers around the globe in just a little under two fuel cycles.

For each challenge, they work closely with the nuclear plant operators to completely characterize the problem, manage existing data and successfully incorporate Artificial Intelligence and Machine Learning (AI/ML) to properly evaluate and mitigate risks, drive more efficient operations, and solve complex challenges, and most importantly, provide predictive visibility into quantities that heretofore were unpredictable. Blue Wave’s solutions are more reactive than current methods and require less human interaction, which leads to improved efficiencies, a lower risk of incident or accidents, and mitigates potential risks to personnel and the environment. Blue Wave likes to point out that not only do their methods save the expense of unneeded fresh fuel but also the downstream costs to store those extra assemblies as “spent fuel.”

Here at inTechBrew, we believe Blue Wave AI Labs is truly at the forefront of AI tools applied to solve high-value, complex problems from the nuclear industry. Their solutions integrate customer’s knowledge and data into concrete solutions which offer the potential to save hundreds of millions in reload fuel costs, eliminate operational challenges, predict component failure, or optimize scheduled downtime.

Their powerful AI-centric products can integrate seamlessly into established processes, whether it be for reload core design or cycle management applications. Transforming data into game-changing solutions, Blue Wave has proven to be the trusted leader in AI solutions for nuclear power generation, already serving over half the U.S. domestic fleet of boiling water reactors and making a significant difference in their operational efficiency.

Blue Wave’s powerful predictive tools already support:

• Optimizing fuel loading by forecasting MCO and eigenvalue at the design stage,
• Monitoring MCO and eigenvalue evolution throughout the operational cycle,
• Prediction of remaining useful life and predictive modeling of in-core neutron flux measuring instruments
• Forecasting Thermal Limits to help reduce design margin.

User case: Core Loading

Constellation, United States-wide, since 2017

Constellation operates the largest fleet of nuclear power plants in the United States. With 14 BWR’s and 7 PWR’s, they have been working with Blue Wave AI Labs to leverage Machine Learning (ML) to optimise planning across their fleet of BWR’s.

Simply put, ML is a set of algorithms analysing past data to learn and understand the underlying behaviour and predict how it will behave in the future. Actual historical data is used to train the ML predictive model. Usually, the sets of past data are extremely large (e.g., for image recognition, we are talking about millions of training samples) but in this case, Constellation only had between six and eight fuel cycles worth of data (tens of thousands of datapoints in total) across most of its BWR’s. Blue Wave AI Labs managed to establish workarounds to get a ML algorithm able to predict well the MCO K_eff and Thermal Limits.

MCO.ai

Up until recently, there has been no reliable method to forecast future MCO levels prior to or during a new fuel cycle. Excess moisture in the steam is problematic for many reasons, most importantly due to its ability to carry impurities dissolved in the water throughout the entire plant. MCO can increase erosion of the internal surfaces of the main steam isolation valves and at the turbine, potentially causing costly repairs. Perhaps even more troublesome, soluble ⁶⁰Co is carried over with the steam which increases plant dose rates and the collective radiation exposure of plant personnel. Beyond this, a small reduction in electrical output occurs with high MCO. Consequently, the primary method to mitigate high MCO is to design the core with a larger-than-required reload batch size, thereby introducing potentially unnecessarily high reload fuel costs.

MCO.ai provides unparalleled accuracy for MCO forecasting in both reload core design and cycle management engineering applications. The predictive capability of MCO.ai is illustrated below for 2 BWR units. The diagram below depicts how Blue Wave’s predictions lead to millions of dollars in savings per fuel cycle with more efficient fuel arrangements, reduced risks of derating due to exceeding limits, increased safety and protection of downstream assets.

Eigenvalue.ai

K-effective (aka, the eigenvalue) is one of the most fundamental parameters in nuclear engineering and has been notoriously difficult to predict accurately in BWRs.

Conventionally, eigenvalue predictions rely on estimates made by core designers looking at past eigenvalue behaviour and the characteristics of the reload core being designed. This approach has its limitations, especially when new fuel or core designs are introduced, and on average has been sufficient to achieve a deviation of Δ ∼ ±0.002 between the design and online eigenvalue. The possibility exists to reduce this deviation more than 4-fold, thereby leading, potentially, to millions in annual savings.

Eigenvalue.ai delivers here unparalleled accuracy for eigenvalue forecasting in both reload core design and cycle management engineering applications. The predictive capability of eigenvalue.ai is illustrated below for a 24-month BWR fuel cycle. The diagram below depicts the delta for design targets and for actual eigenvalue readings, as well as Blue Wave’s daily model predictions. The eigenvalue model performance demonstrates greater than a 4-fold reduction in prediction uncertainty when compared against the current state of practice (conventional design targets), with an average error less than ±0.0005.

The data presented in this use case are from two reactors, but it is worth noting that eigenvalue.ai and MCO.ai have now been deployed across the entire Constellation’s fleet with similar results. Comparable levels of accuracy have been obtained at multiple other BWRs that have adopted this enabling technology:

• Over the past three years, the average prediction error for the MCO is +/- 0.018% for this 2-reactors station;
• The limitation on precision is now only due by the resolution imposed from the MCO measurement uncertainty;
• The uncertainty for K_eff has been divided by 4 against previous analytical methods with an average error less than +/- 0.0005;
• The predictive models still work with new fuel types that are introduced in the core (e.g., Accident Tolerant Fuels);
• In the graph for the MCO of Unit B shown below, the MCO spikes above 0.4% due to a fuel defect leading to the insertion of two suppression rods. MCO.ai accurately predicted this spike and was used to derive a strategy to limit the MCO. Prior to using MCO.ai, the operator was wondering if a mid-cycle shutdown was necessary. If the operator would not have used MCO.ai, but rather performed a mid-cycle shutdown, the expected additional cost would have been above $6 million is lost generation revenue.

ThermalLimits.ai

Historically, inability to accurately predict online thermal limits from offline methods has challenged core design and cycle management. While actual operations may at times depart from cycle design basis projections, there exists an inherent bias between offline and online methods that stems from the nature of the two systems. Both methodologies rely on a three-dimensional neutronics simulator model to calculate the reactor’s power, moderator, void, and flow distributions—from which margin to thermal limits can be determined. However, these calculations are approximations, and the offline quantities determined from them are inexact estimates that lead to uncertainty in thermal limits.

Online methods, on the other hand, employ an adaptive process through feedback directly from in-core nuclear instrumentation while the reactor is online. Up until recently, there has been no reliable method to bridge the gap between online and offline methods leading to inaccurate and inconsistent predictions of online thermal limits.

ThermalLimits.ai is a robust application for the nuclear power industry that provides unmatched accuracy for online thermal limit forecasting in both reload core design and cycle management engineering applications.

The predictive capability of ThermalLimits.ai is illustrated for a typical test cycle for a large BWR. Here you can see individual models for each of the MFLPD (maximum fraction of limiting power density), MAPRAT (maximum ratio of average planar linear heat generation rate), and MFLCPR (maximum fraction of limit for critical power ratio) distributions demonstrate an average reduction in the observed bias by 73% (3.64x) for MFLPD, 46% (1.82x) for MFLCPR, and 67% (3x) for MAPRAT.

Moreover, across all fuel cycles independently tested, the maximum bias between online values and model predictions never exceeds 3.9% (for MAPRAT and MFLPD) and 1.5% for MFLCPR. Utilities estimate that one fuel assembly’s worth of energy can be reclaimed for each 1% of excess Thermal Limit margin reduction.

The Technology: BWnuclear.ai

Blue Wave AI Labs proprietary AI and Machine Learning tools are utilised to evaluate past and real-time data to create fast-running models that will analyse data sets, predict and prevent catastrophic operational delays, and drive better protocols for regulation compliance, efficiency, and safety, saving millions of dollars in the process.

More reactive than current methods, Blue Wave’s solutions lead to improved efficiencies, a lower risk of incident or accidents, and mitigates potential risks to personnel and the environment. Powering nuclear fleet with artificial intelligence, Blue Wave is already a key player in improving nuclear plant efficiency and costs with their AI-toolbox: BWnuclear.ai which is a robust state-of-the-art cloud application for the nuclear power industry including the Blue Wave forecasting tools.

How does it work?

Blue Wave AI Labs combines the insight of exceptional scientific technical talent with the latest advancements in AI and Machine Learning to transform data into solutions for key challenges in the nuclear industry.

MCO.ai, eigenvalue.ai and ThermalLimits.ai are all cloud-based platforms, available from any desktop and are fully integrated in a reload process. Used as iterative tools, the planning teams can run hundreds of scenarios, changing the batch size, bundle specifications, the K_eff strategy and many other parameters.

To be able to successfully develop these algorithms, Blue Wave AI Labs defined what they call the reactor statepoints. A statepoint is a collection of physical parameters / states used as inputs for the ML algorithm. The sought-after information (K_eff, MCO, etc.) are estimated by regression from these statepoints.

Then, the algorithms use these statepoints to train themselves and produce a 3D image of the reactor core, allowing the user to apply “filters” of exposure, void, power, etc. Through this, Blue Wave AI Labs exploit a convolutional neutral network (CNN) architecture to derive dozens of global reactor variables to develop high fidelity models. Below are 3D images of a BWR core with the different “filters” applied.

Enhanced capabilities

Nuclear power plays a vital role in satisfying the ever-growing global demand for low-carbon energy production. However, there are persistent economic pressures that have real potential to challenge the long-term economic viability for domestic nuclear power generation. For instance, Operations and Maintenance currently comprise 60-70% of the overall generating costs of nuclear plants. These costs are expected to rise as the longer-term operation of the existing fleet will require increased monitoring and maintenance capabilities to address the ageing of key systems, structures, and components (SSCs).

The benefits of enhanced diagnostic and prognostic methods include the elimination of unnecessary maintenance, reduction of unplanned outages due to equipment failure, intelligent maintenance scheduling and allocation of resources, and increased awareness of ageing-relating degradation that may threaten plant safety.

It is undeniable that the Blue Wave’s solutions are amongst the capabilities every site will own in the future. The tools developed here are more reactive than current methods and require less human interaction, which leads to improved efficiencies, a lower risk of incident or accidents, and mitigates potential risks to personnel and the environment. But what we have seen here is only the beginning of AI in the nuclear sector. The potential is much broader than just fuel reloading and other tools could certainly join the rank of Blue Wave AI Labs platforms:

• Remaining Useful Life (RUL) of components,
• Ageing Management,
• Virtual Calibrations,
• Virtual Sensors, Substitute measurement for offline monitors,
• Fuel Cycle & Reload Core Optimisation.

Any questions ? Interested in another artificial intelligence system ? Do not hesitate to contact us directly, we will help you find a fit-for-purpose, cost-efficient artificial intelligence system solution to your challenge.