A merging of two technologies is expected to reduce costs and increase the accuracy of inspecting millions of wooden crossarms that hold up power lines worldwide.
Susceptible to rot, wooden crossarms must be inspected and replaced periodically, or otherwise lines could collapse and cause outages. Such inspections have traditionally required workers to climb poles, hit the crossarm with a hammer and judge its condition by listening to the resulting ring. But these manual inspections are imprecise, time-consuming, costly and hazardous.
Now, researchers at the Georgia Institute of Technology have merged laser vibrometry and neural networks to create a remote inspection system that analyzes crossarms from the air.
Work began two years ago when principal researcher Paul Springer of the School of Electrical and Computer Engineering responded to a request from Entergy Services Corp., which funded the project. A global energy provider headquartered in New Orleans, Entergy maintains more than 100,000 crossarms. Program manager of mechanical systems at Georgia Tech's National Electric Energy Testing, Research and Application Center (NEETRAC), Springer specializes in dynamics -- how objects respond when struck or shaken. Drawing upon dynamics, as well as his musical background, Springer began to think about crossarms as a sounding board. "If there's good structural integrity, it's going to have a more pleasing tone when struck," he said.
Springer reasoned that a laser Doppler vibrometer mounted on a helicopter could measure crossarm response to the helicopter's engine and rotors, thus eliminating the need to make direct contact with the structure.
"It's similar to using a laser gun to detect the speed of a car," Springer explained. "Except instead of measuring speed, the vibrometer detects the vibration of a surface by bouncing a laser beam off it." Damaged wood should give a different vibration spectrum than healthy wood.
There was just one hitch: The laser vibrometer could get measurements, but it wouldn't interpret the results of those measurements. So Springer turned to Dr. James Mahaffey in the Georgia Tech Research Institute's Information Technology and Telecommunications Laboratory.
"This is a classic application of neural network technology," Mahaffey said. "There are no set rules about whether the vibrations indicate a good crossarm or a bad one. You need an advanced signal processing method."
Neural networks are a type of computer artificial intelligence that attempts to imitate the way a human brain works. By harnessing the power of computers to sort massive amounts of data for patterns, neural networks create an algorithm of sorts. After "learning" from training data, they eventually can predict a reliable outcome from data never seen before.
Springer and Mahaffey began initial testing with 15 crossarms provided by Entergy. In a lab setting, an accelerometer measured crossarm vibrations while a gasoline engine operated nearby to simulate helicopter noise. After researchers recorded vibration spectra, they broke the crossarms to determine their condition -- a healthy crossarm requires much more force to break than one that's rotting.
Researchers then used data from seven of the 15 crossarms to train the neural network. They divided the vibration frequency spectrum from each crossarm into 200 pieces (ranging from 25 Hz to 520Hz) and then fed the information into the neural network. Researchers used the remaining crossarms to determine whether the neural network had learned to recognize crossarm condition based on vibration spectra.
"We already knew what breaking strength was, so we compared actual value against the estimated value," explained Dr. Ronald Harley, a professor in Georgia Tech's School of Electrical and Computing Engineering.
Lab tests indicated a strong correlation between vibration signals and crossarm strength. The next step was to procure a laser vibrometer and test its ability to make remote measurements.
Air2, a Miami-based flight company, provided a helicopter that enabled NEETRAC research engineer Janeen McReynolds to conduct a field test in January.
Researchers had two goals for the field test: (1) Determine whether the laser vibrometer would work from the air and (2) obtain a larger sampling of crossarms to fine-tune the neural network. "The wider the sample of crossarms, the more accurately you can train the neural network," Harley explained. "It can only learn from those things you show it."
With data collected from 92 crossarms, the field test "far exceeded the most optimistic expectations," Springer said. Researchers plan to test the technology on inspections of up to 10,000 crossarms later this summer. Meanwhile, Springer, Mahaffey and the Georgia Tech Research Corporation have filed a provisional patent application, and now the technology is near commercialization. There are still a few considerations to address, Springer said: "There are no commercially available vibrometers with enough power to get a vibration signal from a long distance. Fortunately, a long-range vibrometer will soon be available."
Less expensive vibrometers require helicopters to fly in at closer range, he explained, which makes inspections more time-consuming and cuts into the economic feasibility.
Yet even by conservative estimates, savings are impressive. Average costs for manual inspections are $50 per crossarm. Using a laser vibrometer and neural network, remote inspections would slice that to about $5 per structure -- one-tenth of current costs.
"Utilities are responding to deregulation by searching for ways to reduce costs and improve the reliability of electric transmission systems," Springer said. "This project will lower inspection costs and improve accuracy. Utilities will be able to use wood components longer because strength, rather than appearance, is the inspection criterion. Components with hidden flaws are detected, while structurally sound parts with superficial defects can remain in service."
For technical information, contact:
1. Paul Springer, 404 675-1815
paul.springer@ee.gatech.edu
2. Ronald Harley, 404-894-1876
ronald.harley@ee.gatech.edu
3. James Mahaffey, 404-894-8967
jim.mahaffey@gtri.gatech.edu
NOTE: A color slide is available from the media contacts listed above. Or you may download it from http://www.gtri.gatech.edu/res-news/latest.html after June 28, 2000.