Data-Driven Decisions Benefit the Nuclear Power Industry

EPRI research investigates how artificial intelligence and data science can improve plant performance and lower costs

By Chris Warren

Containment buildings are critical for nuclear power plants’ safe and reliable operation. Usually made of steel-reinforced concrete, these airtight buildings house everything from the nuclear reactor and its pressurizer to its cooling pumps, steam generator, and piping. The fundamental job of containment buildings is to prevent any radioactive materials from leaking into the atmosphere.

Regular and rigorous inspections of containment buildings are standard at nuclear power plants and are monitored by nuclear regulators. Traditionally, these inspections are a time-consuming, manual task involving plant personnel visually examining a containment building for potential defects, often from a large crane.

Recently, EPRI began investigating the possibility of replacing or augmenting the standard visual inspection of containment buildings with a combination of unmanned aerial systems (drones) and artificial intelligence (AI)-powered machine vision models. In simple terms, the idea is this: Use drones to collect an abundance of videos and still images of the containment building, then tap the power of AI to analyze those images to look for potential cracks, abrasions, corrosion, and other structural defects.

More specifically, in its initial research to pilot this new inspection technique, EPRI teamed up with a member utility to gather about 2,500 images of a containment building. Those images were then used to develop and train a model to pinpoint possible damage. This approach is still under development as part of a larger EPRI program called Data-Driven Decision Making (3DM), an effort to expand the use of AI and the value it can deliver to the nuclear power industry.

“3DM is providing tools and identifying problems AI might help with. These could include flexible power and other applications in nuclear plants, like operations maintenance and inspections,” said Robert Austin, an EPRI senior program manager who leads a range of initiatives focused on plant modernization.

Four Research Areas

The 3DM program encompasses a multitude of individual projects. They can be bucketed into four broad categories:

• Insights—Artificial intelligence, machine learning, and data science have the capacity to improve the performance and operations of nuclear power plants in multiple ways. One way is by combing through data collected through years of past operations to look for insights, lessons, and best practices. For example, EPRI collected a large number of maintenance work orders from nuclear power plants and analyzed the data using AI algorithms. “It would be hard for one person to look through all of that data and figure out trends,” said Thiago Seuaciuc-Osório, an EPRI senior technical leader who is supporting the 3DM initiative. “What AI allows us to do is to look at the entire body of data together and draw out insights and trends that a subject matter expert can then look at for guidance in their own decision making.” EPRI is also developing a unique power industry dictionary to guide the text analysis done by AI. Without this fundamental natural language processing work, it would be extremely difficult or impossible to gather valuable insights from past work orders or equipment manuals.

• Prognostics—Ultimately, using AI to examine the past is a way to improve future decisions and operations. But data science technologies also make it possible to evaluate real-time plant data to predict important events, including equipment failure and the end of an asset’s remaining useful life. Leveraging data and AI to predict future equipment failure gives nuclear power plant operators the information they need to prevent unexpected outages and strategically plan their inspections and maintenance downtime. To help turn data into useful insights, EPRI has launched an effort to make better use of all of the information that is currently being collected by sensors installed at nuclear power plants. EPRI is also applying machine learning techniques to improve flow-accelerated corrosion inspection programs. The ability to better predict failures could reduce unnecessary and costly inspections.

• Automation—The use of drones and machine vision models to inspect containment buildings is just one example of how AI can improve inspections at nuclear power plants. EPRI has several other projects to bolster automated nondestructive evaluation (NDE) tasks, including fuel assembly defect detection inside reactors and dry cask storage vents. If successful, these and other more automated inspection techniques could replace or supplement the costly and time-consuming work currently done by plant personnel. Leveraging automation is also an opportunity to eliminate otherwise inevitable errors. “Some activities performed by people are subject to human error,” said Seuaciuc-Osório. “A good example is inspecting images of pipe looking for cracks. Most of the data they look at will have nothing because defects are rare. But after seeing hundreds of images with no defects, they may miss something because they’re tired. That is where automation comes in for reliability, and there are lots of applications where data screened by automation highlights areas where inspectors should look closer.”

• Optimization—EPRI is also exploring ways that AI can improve processes, plans, and strategies at nuclear power plants. One example is supply chain efficiency. Nuclear power plants prepare for a wide range of contingencies, which means that they order and warehouse a very large amount of inventory, much of which is never used. This is problematic because it can be expensive to store large numbers of unused items. AI can evaluate past order history and whether or not inventory was used to help guide future purchases. “We can look back and see how many times we purchased a part and never used it. AI comes in and takes all the information together and helps make decisions about what you should buy and what you don’t need to stock,” said Seuaciuc-Osório. “There is also a risk associated with not having a component you need on hand. AI can give you information about the likelihood you will need it, and you can decide what your risk tolerance is for not having it.” Beyond driving supply chain efficiencies, EPRI is also investigating how AI can improve waste packaging, scheduling, and the costs of decommissioning activities.

How Artificial Intelligence Can Help Nuclear Plants Drive Decarbonization

While the U.S. nuclear fleet has a long track record of safe and efficient operations, AI provides a tool for improving its overall performance at a time when market pressures threaten the financial viability of some plants.

In particular, the influx of intermittent generation like solar and wind, along with low natural gas prices and flat electrical load growth, has contributed to changes in how nuclear power plants operate—changes that often have a negative impact on revenues.

Ensuring the continuing financial viability of America’s nuclear power plants can play an important role in achieving ambitious decarbonization goals, including the Biden administration’s target of a carbon-free electric grid by 2035. In fact, according to the U.S. Department of Energy (DOE), nuclear energy provided over 50% of the nation’s carbon-free electricity in 2020.

“Artificial intelligence solutions can help to reduce costs to keep plants financially competitive. When plants remain financially viable, it indirectly helps with decarbonization efforts,” said Seuaciuc-Osório.

EPRI’s Role as Data Collector

Several significant developments have made it possible to tap AI to benefit the nuclear power industry. One is the proliferation of low-cost sensors able to monitor the operations and performance of the equipment in power plants. “Digitization means that it’s no longer only about going out and looking at analog gauges,” said Austin. “With digital sensors, you can constantly measure vibration noise and electrical current and bring that information back to a central location, where you can look at trend lines and see if something is wrong.”

The capacity to deploy AI to analyze all of the data collected by sensors is possible today, thanks to dramatic drops in computing power and data storage costs.

But there are two fundamental truths about AI’s capacity to deliver insights and other benefits to the nuclear power industry. One is that AI is only as good as the data that informs it; the other is that more data translates into better AI.

Data is very sensitive in the utility industry, especially among nuclear power plant operators. This is why EPRI is both promoting collaboration and data sharing while also developing safeguards for protecting data. “We developed a data intake process that is more formal and efficient than what was in place in the past,” said Austin. “EPRI has worked to provide a process where we do this data intake with an infrastructure in place that includes governance processes that allow us to be good stewards of the data.”

EPRI’s future work will continue to explore ways to leverage the unique capabilities of AI to benefit the operations and performance of nuclear power plants. “AI is going to be an important part of how plants move ahead and become more efficient and valuable,” said Austin. “Our work will continue to push and facilitate the collaboration needed to achieve that potential.”

Banner artwork by McKibillo

---