On Receipt 12/29/94
NEW METHODS TO DETECT MALFUNCTIONS AT NUCLEAR PLANTS DEVELOPED
COLUMBUS, Ohio -- Researchers here have developed new computer-based methods to detect slow-developing malfunctions at nuclear power plants that could cause accidents and temporary plant shutdowns.
After eight years of study, researchers at Ohio State University have found that a system which combines different software designs could detect slow-forming malfunctions -- called faults -- in nuclear plants. By catching these problems early, researchers believe plant operators can reduce plant shutdowns and prevent accidents. Some of these findings were published in a recent issue of the journal Reliability Engineering and System Safety .
The research team, led by Don Miller, chair of nuclear engineering at Ohio State, successfully tested the integrated software approach in the training simulator at the Perry Nuclear Plant, about 40 miles east of Cleveland. The Ohio State team was one of the first in the country to test fault-detecting software at a nuclear plant simulator, Miller said.
"Integrated computer systems will enhance the reliability of
a plant by allowing operators to intervene before a fault causes a problem that would trigger the plant's safety systems, automatically shutting a plant down in case of an emergency," Miller said.
Nuclear plants are made of up to 80 mechanical and nuclear systems, all monitored by human operators. Monitoring of individual and safety systems alert employees to problems in the plant. However, because a plant's 80 systems work independently and interactively, the operators sometimes misdiagnose problems in systems that interact, or fail to detect small faults as they are developing, according to Miller.
For example, a mechanical failure in a valve that is circulating water through a boiling water reactor may be difficult for an operator to detect, Miller said. Although this problem may not pose an immediate threat, he added, it could develop into a more serious problem.
If the fault reaches a critical stage, it will trigger an automatic safety system that will shut down the plant. Plant shutdowns can cost up to $1 million a day in lost energy production, Miller said.
"An integrated computer-based system would alert an operator of hard-to-notice problems such as this and give the worker the opportunity to react to it, thus preventing a plant shutdown," he said.
When Miller and his colleagues began this line of research in 1986, they didn't believe that a computer-based system was capable of safely running a nuclear plant. When the study began, the fail-safe software needed to successfully monitor a plant's intricate systems had yet to be developed, Miller said.
Over the last eight years, however, developments in computer science gave Miller's team new methods to work with, including improved techniques in knowledge-based systems, modern control theory and neural networks. By combining these techniques, Miller said it is possible for a computer to operate or reliably monitor portions of a nuclear plant with a high degree of safety, although some degree of human oversight will remain necessary.
"A computer-based system to detect faults in a nuclear plant has to be structured differently for each of the 80 types of systems in the plant," Miller said. An integrated system would be tailored to each plant system's characteristics, he added.
Modern control theory is a technique that allows for multiple data input into a computer system, allowing operators to program the computer to anticipate a variety of different scenarios. Software using this technique is also capable of responding in a variety of different ways and can be tailored to meet a plant system's characteristics.
Knowledge-based systems use data about the plant supplied by operators and by plant design data to locate faults in plant systems. This type of fault diagnosis works well with individual plant systems, but may be unable to diagnose problems in plant systems that interact.
However, Miller said, interacting systems are monitored well by combining plant models (which detail how plant systems operate) and neural networks, which are designed to react to actions occurring within a plant and to identify problems in specific plant systems.
J. Wesley Hines, a member of the research team and recent doctoral graduate from Ohio State, found in his doctoral work that combining neural networks with plant models, which may include knowledge-based systems, will give better coverage of a nuclear plant's interacting systems.
The integrated software developed by Miller's team could be used in the 105 nuclear plants in operation in the United States, which supply about 23 percent of the nation's electrical energy.
Miller said plant personnel will need to learn more about the integrated computer systems before they can be put to use in nuclear plants, adding that plants that decide to use this technology could begin implementation in the next few years.
The group's research has been sponsored by the National Science Foundation and the Department of Energy.
Back to Archive
Go to Current Month News Research Stories
Go to Current Month Newsfeature Stories
Go to Current Month News Cancer Report Stories
Contact: Don W. Miller, (614) 292-7979
Written by Kelli Whitlock, (614) 292-9475