Science & Technology: SAFETY
RIPS IN THE NUCLEAR SAFETY NET?
How safe is your local nuclear power plant? Constantly assaulted by radiation and temperature variations, nuke plants' steam tubes can develop dangerous cracks. To find them before they grow, engineers apply electric currents to the metal and read the results on an oscilloscope. The Nuclear Regulatory Commission is highly confident of these tests--so much so that operators who produce good test results can seek licenses to run their plants for two decades past their originally intended 40-year life spans.
But now, this so-called nondestructive evaluation is under fire. Terry Oldberg, a 55-year-old veteran of the nuclear industry, argues that the statistical basis of verifying nondestructive evaluation is fundamentally flawed. The result, he says, is that there is no way of knowing whether nondestructive evaluation as currently practiced is reliable or not. Says Oldberg: "They've got an instrument; they can't tell how good it is. It doesn't produce information."
TOUGH CALL. If Oldberg is right, the ramifications are enormous. If nondestructive evaluation underestimates the damage in a nuclear plant, a catastrophic failure could occur unexpectedly. On the other hand, if the tests overstate damage, a plant could be shut down unnecessarily. Plant shutdowns related to metal damage are frequent. Maine Yankee Atomic Power Co., for instance, is spending $40 million to encase all 17,000 steam tubes in its Wiscasset (Maine) plant after tests indicated cracks. Oldberg argues that the NRC should conduct new tests of the accuracy of nondestructive evaluation or reanalyze the raw data from old tests using standard statistical methods. That, he says, would produce a solid understanding of how accurate nondestructive evaluation actually is.
NRC officials say their verification methods are logically consistent and reliable. As for Oldberg, "it's very difficult for us to figure out what he's saying because his arguments don't always make sense," says Lee R. Abramson, senior research statistician in the Probabilistic Analysis Branch of the NRC's Office of Nuclear Regulatory Research.
Nonetheless, Oldberg is starting to pick up support in a crusade he has been on since the late 1980s. A key ally is Ronald A. Christensen, president and co-founder of Entropy Ltd., a statistical research and consulting firm in Lincoln, Mass. Entropy's clients have ranged from drug companies to the NRC. Oldberg has also gotten support from Eric Perdijon, director of quality assurance in the department of thermo-hydraulics and physics at France's Commissariat de l'Electricite Atomique. Says Perdijon: "We should have clear laws" for how testing methods are verified.
Oldberg's controversial stance began to get more attention in January when he presented a peer-reviewed paper he co-wrote with Christensen at the annual engineering conference of the American Society of Mechanical Engineers (ASME) in Houston. The ASME writes codes for steam tubes, among other things. Donald E. Bray, a Texas A&M associate professor of mechanical engineering who served as chairman of Oldberg's panel, says he plans to put some of Oldberg's findings in the next edition of a textbook. But he also says he thinks nondestructive evaluation as practiced has helped make the nuclear industry "extraordinarily safe." Says Bray: "I don't know who the good guy and bad guy is, frankly."
The man who's stirring up this controversy is hardly an outsider to the nuclear industry. In the 1960s, Oldberg worked for the government's elite Lawrence Livermore National Laboratory, designing a nuclear reactor for outer space. Later, he worked on a fast-breeder reactor for General Electric Co. From 1974 to 1986, he worked for Electric Power Research Institute in Palo Alto, Calif., the R&D arm of U.S. electric utilities. His last job there was to oversee research on the safety of steam generators in pressurized-water reactors.
It was during that EPRI assignment, Oldberg says, that he discovered the statistical flaw underlying nondestructive evaluation research. He says a disagreement over the issue led to his departure from EPRI, but he won't comment further. After failing to find another job in the nuclear industry, he went back to school to study programming and has been trying to get a job while living off his wife's salary and firing off letters to the NRC and American Society of Mechanical Engineers.
Oldberg is careful to say he doesn't object to nondestructive evaluation itself, only to the statistical methods engineers use when they check the test's accuracy by occasionally cutting steam tubes to compare indications of flaws with actual flaws.
In medicine, verification is simple: The study population consists of every person in a trial. A "false positive" is calling a well person sick; a "false negative" is calling a sick person well. Oldberg favors applying the same approach to steam tubes. Each, say, two-foot length of tube would be a unit, analogous to a patient. The test would label a section positive (that is, sick) if the minimum thickness of its wall appeared to be less than, say, 60% of the original thickness. For verification, the nuclear "doctors" would cut the tubes and measure the actual minimum thickness of each length.
The NRC went wrong, Oldberg says, by defining a nonstandard population. Instead of the units being lengths of tube, the NRC's population units are the flaws themselves. The NRC doesn't tie the flaws to any given length mf tube. Instead, it counts as a "true positive" any flaw that falls within three inches of an indication--the device's margin of measurement error. The problem is, there's no way of knowing whether the flaws found in this "zone of uncertainty" are actually being indicated by the device. It could be that the device just happened to make an indication in the same area. As Christensen notes: "The less accurate your instrument, the higher your rate of detection."
This may sound confusing, but it's considerably clearer than the correspondence among Oldberg, the NRC, and the American Society of Mechanical Engineers. Indeed, miscommunication may be a big part of the problem. Texas A&M's Bray says what's needed is a well-advertised forum where experts from all fields can grapple with the issue and come to a decision once and for all. A decade is too long for such an important issue to kick around unresolved.
NRC If an actual flaw is found within three inches of an indication, the NRC deems it a "true positive." The NRC says its test is good because nearly every flaw is called a "true positive" and there are few "false positives"--that is, indications that don't have flaws within three inches of them.
OLDBERG The NRC's verification method is untrustworthy, Oldberg says, because given the uncertainty in position, there is no way to be sure that an indication is pointing at an actual flaw. It could be that the device did not see the actual flaw, but just happened to make an indication in that area.By Peter Coy in New York Science & Technology