ROGER DUNCAN, DAWN GIFFORD, and VEERA RAJENDRAN
As electrical generators age, they become increasingly prone to failure. The failure of a single generator will typically lower the output of a power plant by as much as 20% for as long as two months while the stator cores are evaluated and repaired. This decrease in capacity makes it difficult to meet demand, particularly during peak usage times. To prevent failure or at least mitigate rolling blackouts, it is increasingly important to provide real-time monitoring and diagnostics of power-generation equipment.
Perhaps the most common cause of failure in industrial electrical generators is electrical shorting due to insulation failure. Insulation tends to fail for one overriding reason: overheating. Overheating often results from operating machines outside of their specification envelope, intentionally and otherwise, to meet demand during peak usage times. It can also be the result of happenstance, such as ventilation blockage caused by excessively dirty environments.
To compensate for overtaxing their equipment during peak times and to extend operational life, power-plant operators often intentionally run their electrical generators at less-than-optimal efficiencies during nonpeak usage times. Neither failures nor operating at suboptimal efficiencies are desirable. Detailed temperature monitoring can alleviate both of these problems.
The power-generation industry typically uses thermocouples or resistance temperature devices (RTDs) to monitor temperature in electrical generators and motors. However, it has long been understood that these electrical-based sensors are at a disadvantage in a device that generates such excessive levels of electromagnetic interference (EMI). Thermocouples must be located away from the high-voltage regions of the machine, in the neutral phase, to avoid serious safety hazards.
Resistance temperature devices are typically installed not on the metal surface of the copper, where temperature measurements are most desirable, but in the insulating medium between the inner bar and the outer bar. A more realistic view of the copper temperature is then calculated by applying conduction models. Because of the labor-intensive nature of RTD installation, and because having a mass of electrical wires in an electrical generator is dangerous, only a few RTDs of specific lengths are used for an entire machine.
The temperature data received therefore represents an average of the total temperature picture. This averaged temperature map tends to mask any hot spots that might exist, making the accurate identification of likely failure points exceedingly difficult.
The failure of these conventional monitoring techniques to substantially improve operational efficiency has led engineers in the industry to examine alternatives. Fiberoptic sensors can offer numerous advantages over conventional electrical sensors, including a nonintrusive nature; immunity to most harsh environments; lightweight, intrinsic safety; and immunity to EMI. And unlike electrical sensors that are difficult to apply in significant number, distributed sensing techniques allow numerous sensors to be multiplexed along the length of a single optical fiber, effectively lowering per-sensor installation cost and increasing measurement density.
Fiberoptic-based distributed sensing techniques have recently gained market acceptance. Optical time-domain reflectometry (OTDR) is based on sending a pulse of laser light through the fiber and tracking the time at which the reflected signals are detected, thereby distinguishing the spatial location of the sensors. Although this technique works very well over great distances (kilometers), spatial resolution tends to be coarse and it rarely provides better than one data point per meter.
Perhaps even more popular is the wavelength-division-multiplexing (WDM) technique in which each sensor reflects a different wavelength of light. Though high spatial resolution can be achieved with this technique, the number of sensors tends to be limited to tens because of the finite bandwidth of broadband or swept-wavelength laser sources. Furthermore, WDM sensor fabrication is very labor intensive, and thus costly.
An alternative to these two distributed fiberoptic sensing techniques is optical frequency-domain reflectometry (OFDR). The OFDR technique was originally developed by researchers at NASA Langley Research Center (Hampton, VA) for testing on the X-33 and the shuttle, but has recently been commercialized for numerous monitoring applications. Optical frequency-domain reflectometry allows thousands of sensors with the same nominal reflected wavelength to be read with very high spatial resolution.
By applying OFDR to the electrical generator problem, it is possible to get real-time continuous temperature feedback. In addition, temperature mapping enabled by this technology allows for a substantial shift in the way power-generation engineers and technicians operate their equipment. Because hot spots can be readily identified, unexpected equipment failure caused by an electrical short resulting from insulation breakdown will no longer shut down a plant (see Fig. 1). Indeed, given the high sensor density, power-generation engineers can run their equipment at optimal efficiency without fear of overheating.
Theory of operation
The sensors measured with the OFDR technique are fiber Bragg gratings (FBGs). To fabricate an FBG, photosensitive fiber is exposed to a periodic pattern of pulsed ultraviolet light from an excimer laser, forming a periodic change in the refractive index of the fiber's core (see Laser Focus World, April 2000, p. 107). This pattern, or grating, reflects a very narrow band of light that is dependent upon the period of the refractive-index profile formed in the core.In its most basic operation as a sensor, a Bragg grating is either stretched or compressed by an external stimulus. This movement causes a change in the period of the grating, which in turn causes a shift in the wavelength reflected by the grating. By measuring the shift in reflected wavelength, one can determine the external stimulus applied (see Fig. 2).
When a Fourier transform is performed on the raw data, the signal from each grating then lies in a unique frequency window in the transform domain. A bandpass filter is used to window the signal from each individual grating. The data are then inverse-transformed and the center wavelength of each grating's reflected spectrum is determined. The shift in center wavelength from the nominal center wavelength can then be interpreted to determine the applied stimulus.
The principal advantage of the OFDR technique is that each of the FBGs measured can have identical nominal reflected spectra. Because of this and the high sensitivity of the method, this sensing technique offers three primary advantages over other FBG demodulation techniques. First, sensor fabrication can be greatly simplified by writing the gratings on a draw tower while the fiber is being fabricated, which reduces average sensor cost significantly by automating the process. Second, many more sensors—thousands instead of tens—can be multiplexed on a single fiber. Finally, high spatial resolution (as small as 0.25 in.) is easily achievable.
Recently, we have proven the validity and the value of this optical sensing technology. Sensing strings containing high-density arrays of FBGs (0.25-in. spatial resolution) were installed in a large-scale motor made by GE Industrial Systems. The motor is rated at 17,000 HP spinning at 327 RPM and operating at approximately 13,200 V. The sensing strings were installed in the copper windings in stator slots and covered the entire axial length of the stator bars.
As the FBGs are sensitive to both temperature and strain changes, the optical fibers, with a cladding diameter of approximately 125 µm and an additional buffer layer of approximately 25 µm of polyimide, a material known for its excellent thermal properties, were installed in a Teflon tube with an inner diameter of 300 µm and an outer diameter of 350 µm. This tubing isolates the fiber from strain by allowing the cable to be bonded to a structure for mounting purposes, yet does not transfer the strain to the sensors.
Excellent agreement between the fiber sensors and the RTDs was achieved with a root-mean-square value of the difference between the two sensor types of 0.5°C in the worst case. And, as expected, two hot spots were located with the fiberoptic sensing strings that were not detected by the RTDs. Given the results of these tests, we expect that fiberoptic sensing will be able to compete very favorably with electrical-based sensors in the power-generation industry, allowing for increased efficiency and fewer down times.
ROGER DUNCAN is the technical director of the distributed sensing group and DAWN GIFFORD is an optical scientist at Luna Innovations, 2851 Commerce St., Blacksburg, VA 24060; VEERA RAJENDRAN is a research scientist at General Electric Corporate Research and Development; e-mail: [email protected].