Computational Intelligence in Nuclear Engineering

The overall performance of America’s fleet of 103 nuclear power plants has improved dramatically in the past decade through the use of increased and more effective training and a significant increase in the use of on-line maintenance. As a result, the availability of plants is asymptotically approaching the theoretical maximum in which the refueling activities define the length of the outage. Yet, there are major issues that should be of concern to the nuclear industry and the Nuclear Regulatory Commission (NRC) that bear directly on the continued safe and efficient operation of these plants. Most of these concerns arise from a combination of “stresses” on the reactor core being introduced by three trends: 1) longer fuel cycles, 2) increases in thermal power, and 3) proposed increases in the maximum allowable burnup in the fuel. There are also concerns about unforeseen issues that may arise as plants continue to operate for up to 60 years. The first line of defense against such problems is continuous, on-line surveillance and, when possible, concurrent diagnosis of problems when they occur. The role of Computational Intelligence (CI) in the nuclear power industry is in constant transition due to plant operating objectives, future needs, and regulatory requirements. For example, as plants move towards longer plant licenses, improved maintenance practices become vital. Past practices of corrective maintenance is not practical with one-day outages costing up to a million dollars. Current periodic or predictive maintenance practices may not be optimal when precursors to degradation or failure may be inferred. Many top performing plants are moving towards condition-based maintenance practices when technology permits. This allows a plant to optimize their maintenance by performing maintenance only when the condition requires it. These techniques require robust and reliable estimates of the plant condition, that in many cases requires the use of CI to process the plant data to infer condition. Many of the techniques developed over the past thirty years such as reactor noise analysis are now reaching maturity and paying dividends. Other techniques, such as on-line sensor calibration monitoring, are nearing the maturity in their development; however, just beginning implementation. Still others, such as on-line efficiency optimization and on-line transient identification, still have not yet proven their worth. And lastly, several new techniques, such as autonomous control and multi-intelligent agents are still in their formative years. Approaches to several recent issues in the operation of nuclear power plants using computational intelligence are discussed. These issues include 1) noise analysis techniques, 2) on-line monitoring and sensor validation, 3) regularization of ill-posed surveillance and diagnostic measurements, 4) transient identification, 5) artificial intelligence-based core monitoring and diagnostic system, 6) continuous efficiency improvement of nuclear power plants, and 7) autonomous anticipatory control and intelligent-agents. Several changes to the focus of Computational Intelligence in Nuclear Engineering have occurred in the past few years. With earlier activities focusing on the development of condition monitoring and diagnostic techniques for current nuclear power plants, recent activities have focused on the implementation of those methods and the development of methods for next generation plants and space reactors. These advanced techniques are expected to become increasingly important as current generation nuclear power plants have their licenses extended to 60 years and next generation reactors are being designed to operate for extended fuel cycles (up to 25 years), with less operator oversight, and especially for nuclear plants operating in severe environments such as space or ice-bound locations.