By Laura Buenning

Eavesdropping on the huge mechanical systems in nuclear power plants may not invoke intrigue the way phone taps and hidden cameras do, but it fascinates University of Tennessee nuclear engineering professors Wesley Hines and Belle Upadhyaya.

Working independently and as a team, Hines and Upadhyaya have devised automated surveillance systems to track real-time conditions within a nuclear power plant. Now their challenge is forecasting conditions to predict failures and prolong the life of delicate sensors and the moving parts of mechanisms.

If It Ain’t Broke

Nuclear energy is relatively clean, but its dual role in weapons technology and the issues posed by spent fuel and potential reactor malfunction make it difficult to predict nuclear power’s future in this country.

“It all comes down to safety and economic competitiveness,” Hines says.

One way to reduce costs is to keep the nuclear plant running with as few interruptions as possible—a challenge when you’re dealing with fans, pumps, and scores of delicate sensors and switches, all working in harmony to safely extract energy from radioactive fuel.

“You have literally thousands of signals coming from different parts of the system,” Upadhyaya says.

When the plant shuts down to refuel, the critical safety sensors are recalibrated for accuracy, whether they need it or not.

“Think about changing the oil in your car. Does it really need to be changed every 3,500 miles?” Hines says. “We don’t know because we have no way to monitor it.”

The same is true for the sensors.

“Calibrating one sensor might cost $2,000, and there are hundreds of these sensors in the plant,” Hines says. “We want our surveillance systems to make it possible to perform only required maintenance.”

Hines’s models recognize normal patterns within data, whether he’s looking at a nuclear power plant, a hydraulic unit steering a deep-water oil drill, or the electronic signals coming from an aircraft engine. The techniques also detect minute changes in those patterns—­abnormalities caused by early degradation.

“You can find and flag very small changes, or drifts, in the sensors,” he says. “Then when you go in at the end of a fuel cycle, you recalibrate only the sensors that need it, considerably reducing outage time.” And reduced outage time translates into big savings, says Hines, as much as $1 million a day.

Chain of Events

“Good models,” Upadhyaya says, “come from good data.”

But, says Hines, there’s “no free lunch” when it comes to choosing a modeling tool, so both he and Upadhyaya work from a tool box of models to get the best fit for the situation.

“Some models can be developed from the physics of the system,” Upadhyaya says. Empirical data-driven models built from digital information collected from nuclear power plants take over in the absence of known cause-and-effect relationships.

“But you may also want to include a rule-based expert system,” Upadhyaya says, based on rules like, if A and B both occur, then C is the expected outcome. Hines’s tool kit, for example, includes rule-based fuzzy-logic systems designed to take uncertainty or vagueness in the data into consideration.

Once the model matches the process well enough to track a system, they move on to discovering how to use it to predict defects and failures, Upadhyaya says.

“Predicting failure [prognostics] is a challenge,” Hines says. “For that we need more data and more simulations exploring how physical changes and faults appear, as well as the factors that actually predict an all-out failure.”

Recycle, Recycle, Recycle

Larry Miller leans back in his chair, fingers laced behind his head, and talks about the serious business of managing spent nuclear fuel, his dry humor punctuating salient points throughout the conversation. Better stewardship, rather than scarcity of resources, drives his interest in recycling spent fuel.

“We’re spending a lot of money, time, and energy for an excellent facility—Yucca Mountain—that is very suitable for disposing of spent fuel,” says the UT nuclear engineering professor. Yucca Mountain, Nevada, is the nation’s planned repository for spent nuclear fuel, due to come online in 2017. It could be filled to capacity, however, with a few years’ worth of accumulated spent fuel, says Miller, who contends that we have to recycle to avoid the need for many more Yucca Mountains.

Curious about how various recycling and fuel-cycle scenarios might reduce the amount and storage time of materials destined for long-term repositories, Miller and his students designed a theoretical model to look at the effectiveness of different options.

“Our calculations show that if we would selectively separate out valuable isotopes in the spent fuel and send them back to the reactor, we would need only one-fiftieth of the storage space in Yucca Mountain. It’s a huge savings.”

No fuel cycles in current practice work as well as those assumed in Miller’s theoretical calculations. To get those results, he says, you have to incorporate different types of reactors and recycling into the overall system, which of course brings up the policy constraints surrounding separating out weapons-grade plutonium.

“You can use higher burn-up fuel, you can change the composition of the fuel, you can recycle some plutonium back into the reactor and still satisfy the requirements for disposition of spent fuel,” Miller says. But that will, at best, cut storage needs by only one-half.

“Some of the fuel-management strategies under consideration are talking [reductions by a factor of] 8 or 10, which can probably be achieved based on implementation of technology currently available.

“Theoretically we can do even better,” he says, but this will require investments in research and development considerably higher than our current policy supports.

Tags: Belle Upadhyaya • Energy • Engineering • Environment • Nuclear • ORNL • Technology • Wes Hines