Using Distributed Sensing for Generator Condition Monitoring

A new technology based on extended cavity sensors and optical fiber could be used to provide distributed sensing of vibration in the stator of hydroelectric generators.

By Peter Kung, Zengguang Qin, Wenhai Li and Xiaoyi Bao

There are two failure mechanisms for large power generators: partial discharge (PD) and vibration sparking (VS). PD involves gradual breakdown of the insulation material of the electrical windings and might take five to 10 years to develop into actual faults. VS involves an aggressive failure mechanism that might occur even in new generators in the first five years of operation. The erosion current is limited by the semiconducting paint and would be only milli-amperes. Both problems are caused by vibration and can occur in hydro generators during operation.

Condition monitoring of hydro generators is challenging because of the number of windings in the stator, with some large machines having as many as 600. Discrete sensors would not fit inside the narrow gap of 7 to 10 mm between the stator and rotor. Even if thinner packages were to become available, they would obstruct air cooling and would be too expensive (potentially running into $500,000) and time-consuming to install on a large machine. Furthermore, generators present an extremely hostile environment of high voltages and a strong electromagnetic field, making cheaper electrical sensors unsuitable.

The underlying driving forces behind a condition monitoring solution that could extend the life of generators are economics and technology. Technology that uses Raman scattering can detect temperature rise as an increase in scattered light reflected inside a multimode optical fiber and works well for distributed sensing of temperature, but it lacks the required resolution, with resolution of about +5 degrees Celsius, sufficient for tunnel fire detection. Technology using Brillouin optical time domain reflectometry (OTDR) involves two beams of counter-propagating laser light with very closely matched center wavelengths. They beat together, and the beat frequency varies linearly with temperature and offers better resolution of + 1 C. Unfortunately, it costs close to $100,000 and is too slow to measure vibrations. The lowest cost solution for monitoring the condition of a hydro generator measuring vibration and temperature has become a popular research topic around the world, especially among generator manufacturers.

The more severe problem, VS, is related to corona formation due to high voltage buildup, followed by intermittent contact disrupting a ground current loop causing the corona plasma to implode, generating an erosive spark. However, this current is only milli-amperes and would take many erosion cycles to destroy the insulation layer. This process is well-described as electric discharge machining. The corona starts because strong vibrations inside the machine disrupt the connection of the winding insulation to the grounded core. The corona gives off a strong two time line frequency. For example, in North America where the line frequency is 60 Hertz (Hz), these machines would generate vibrations at 120 Hz. The amplitudes of these vibrations can be related to the looseness inside the machine. Signal generated during VS will involve a signal with much larger amplitude, plus an acoustic signal with some higher-frequency components. PD signals in the generator would be of an even higher frequency range of 30 kHz to 200 kHz.

In early 2013, with the help of Siemens, the authors performed a simulated arcing experiment on a short winding sample and found a sparking signal of around 6 kHz to 19 kHz forming its unique signature and becoming the basis of detecting the onset of the VS problem, giving owners an opportunity to rewedge the slot and make things tight, giving the machine another life extension. Unfortunately, Raman scattering and Brillouin OTDR can only capture frequencies of up to 30 Hz.

The authors searched for two possible solutions to the lack of a proper technology that can provide on-line monitoring of this VS signature. The first, discussed here, is a fiber-based technology called phase OTDR.

The other solution, not discussed in this article, involves arrays of fiber Bragg gratings that might give better speed and sensitivity performance, utilizing a combined platform of wavelength multiplexing and frequency multiplexing.

There are other challenges for the condition monitoring of hydro generators. Some of these problems relate to a new mode of operations. In the past, hydro generators have been used as base load machines. Increasingly, some are being used for pumped storage, working in conjunction with large wind farms. In this case, cold starts do not give sufficient time for the stator to warm up and expand enough to allow a reasonable gap between the rotor and stator. As a result, rubbing might occur, damaging the machine.

VS can be influenced by temperature

In a hydro generator, the voltage drops gradually across all the windings. It can be difficult to identify which winding has the highest induced voltage, and VS becomes the main problem. Normally, the windings in the slots are kept tight by side-mounted semiconducting ripples that press the semiconducting-paint-coated windings tightly against the grounded core. When vibration amplitudes become large, coupled with the aging of these side ripples, the contact to ground becomes intermittent. High-voltage arcing then occurs, forming corona and a plasma of ionized air. If the winding temperature becomes higher than usual (hot spots), the paint becomes very conductive, allowing a large induced current to flow through the core between two isolated points. This current is magnetically induced but cannot be sustained because the same vibration continues to cause intermittent contacts.

When the current is interrupted, the plasma will implode and cause erosion of the epoxy mica beneath the conducting paint on the windings (sparking). We cannot predict the location of these current loops or where contacts will be lost. The erosion might occur anywhere within the generator and affect several slots in the windings, causing the generator to be shut down. This failure mechanism is hard to predict and could occur in relatively new generators with only a short history of operation.

For VS to occur, large-amplitude vibration is needed. Unfortunately, vibration is always present in rotating machines because the rotor is never perfectly circular or centered within the stator structure. The slow turning speed of a machine with a four-pole design generates a two times line frequency (100/120 Hz). This is the key signal to monitor. Its amplitude will grow whenever there is looseness inside the stator structure. The challenge is identifying and locating these loose structures. Hence, the need for distributed vibration sensing.

Corona causes this key signal to increase momentarily from 20 to 50 mV all the way to 1 V peak-to-peak. In addition, excessively conducting carbon paint will only occur with elevated winding temperature. When the generator is at full load and beyond, the current might be so high that even the smallest contact resistance will introduce large Ohmic heating. Unfortunately, resistance temperature detecting sensors are usually mounted at the back of the core, far from the windings slots. There is no temperature measurement inside the slot. The gap is too small to fit packaged sensors, and electrical sensors would not work. Thus the need for both distributed temperature and vibration measurement inside the whole stator. This article focuses on distributed vibration sensing.

Phase OTDR

The phase OTDR sensor has been demonstrated for distributed vibration measurement.1,2,3,4

First, the team packaged a complete test setup and performed a series of lab tests on this phase OTDR technology at QPS. The test setup consisted of a pulse generation circuit used to drive an electro-optical modulator, which turned the narrow line laser source into optical pulses. Two fiber optics amplifiers were included to compensate for the high-loss components. A high-speed receiver was used to collect all time domain signals that could be observed with an oscilloscope.

A laptop computer was used to perform wavelet analysis using Labview, a graphic user interface program developed by National Instrument. The wavelet analysis removes noise related to polarization and the environment. Without wavelet analysis, it is difficult to locate the multiple-vibration points. To emulate the hydro generator, we constructed a multi-point vibration source. At the end of each test, a review was conducted and suggestions were made to improve performance of the technology and system architecture. This article covers the results of the first two test sessions conducted.

Inside the apparatus, each vibration point was separated from the next by a fiber length of 0.5 m to simulate the typical distance between neighboring slots in the generator windings. We will note the maximum number of vibration points detected by the phase OTDR technique. We need to identify the vibration frequency and amplitude at each winding. Signal processing will be developed, distinguishing various multiple vibration points. We will then refine the sensing (fiber) packaging to improve the detection sensitivity of the system and simplify the installation method.

Information about the vibration source

The vibration source built to test the phase OTDR technology contains 10 speakers that can generate 10 independent frequencies and amplitudes. Initially, each vibrating point was separated by an 85 m loop of optical fiber, manually coiled to form diameters of 35 mm. The range of frequencies varied from 5 Hz to 200 Hz. We learned that the hydro generator will show a built-in vibration of two times line frequency (100 Hz to 120 Hz depending on whether it is located in Europe or North America). The lower frequency of 5 Hz is to allow for the detection of shaking in the whole generator due to poor foundation. The coils are spliced together into a chain that terminates with index matching gel to prevent reflection.

In a hydro generator, the vibration signals inside the slot have the same frequency (namely two times lines frequency). What is important is to detect amplitude changes because amplitude increase is a sign for structural looseness. Once located, maintenance staff can install additional wedges and restore the generator to normal condition.

Results of the lab tests

The tests were performed using the sensing system built by Ottawa University on a vibration platform developed by QPS, which was taken as a simulator of in-line, multiple-point vibrations inside a hydro generator. Each fiber coil was contacted by the surface of a speaker diaphragm and mechanically deformed during vibration. The mechanical and periodical deformation of fiber coil, caused by vibration, induces repetitive local changes of the physical parameters of the fiber – such as refractive index, birefringence, or loss – which eventually affects the intensity, state of polarization and optical phase of the pulse as it passes through the fiber coils.

The deformation magnitude of each fiber coil is controllable through manually adjusting the driving voltage of each speaker. The oscillating frequency of each speaker was preset to ranging from 5 Hz to 200 Hz. During the test, we avoided sharp turns where curling diameter becomes less than 30 mm. In these experiment, the speakers were turned on or off according to the measurement requirements. During the test, only nine of the 10 speakers were responding. We suspect the sampling rate is still not high enough. To achieve higher-speed signal processing, we may have to use hardware solutions such as programmable logic arrays instead of software tools like Labview.

When several speakers are simultaneously creating a stream of mixed signals, the long-gauge Vibrofibre sensor from QPS can also detect vibration from 5 Hz to 100 Hz. However, the Vibrofibre system was not able to differentiate the position of the vibration sources. With the phase OTDR, only three simultaneous sources of vibration had been identified during the first test.

Direct detection was used in the second test, while the first test was conducted with coherent detection.

A single vibration event was first measured by the phase OTDR system using coherent detection. Five hundred traces were recorded using an oscilloscope with a 500 MHz sampling rate. The pulse repetition rate was set to be 10 kHz. About 500 denoised Rayleigh backscattering time-domain traces were collected. To reduce the amplitude fluctuation in the Rayleigh signal traces due to phase noise of the laser, partial interferometric problem (random polarization), and electrical noises, a wavelet denoising method was introduced to remove random noises.2 This signal denoising method is based on the idea of threshold wavelet coefficients of the noisy signal using the discrete wavelet transform to remove the noise. An obvious change of amplitude around 100 m was observed.

In the first trial, only three simultaneous events and their location and frequencies were detected (see Figure 1 on page 60). It was also difficult to differentiate the amplitudes of the vibrating events. The reason for this problem is Rayleigh signal fading induced by the polarization state mismatch between the sensing arm and reference arm in the coherent detection scheme. For the single vibration case, the polarization state of the Rayleigh signal at the position under the vibration can be adjusted to be matched with the reference arm. However, for the multiple events case, the polarization states of the Rayleigh signal at different vibration positions cannot be adjusted to be matched with the reference arm, leading to the signal fading.

three events detected in first trial

However, the locations of those three events detected were correctly identified using the preset frequencies. After a review, we decided to increase the pulse rate by 100 kHz and simplify the test circuit from balanced detection to single end detection. The second test took place one month afterwards, with the change to direct detection to reduce the polarization fading-induced fluctuation.

Figure 2 on page 60 showed a clear improvement. Increasing the sampling rate also increased the size of the data. Because the wavelet denoising algorithm was written in the Labview interface, it tended to be quite slow, limiting the frequency response of the system. We had difficulty in reading the time domain signal at elevated frequencies, and we could not accurately measure the amplitude response. This is a very important aspect of the target application. For frequency, our main target signal would be two times line frequency.

nine events detected in second trial

Conclusions

The direct detection methods showed improved capability to detect and distinguish multiple-vibration events. It is not yet able to read the amplitude differences. Figure 3 on page 62 shows the direct detection method is able to correctly identify nine of the 10 vibration events. We are improving the method further by increasing the sampling rate, which means the wavelet signal processing algorithm might have to be implemented by hardware and only provide the calculated result via Labview for display. This will make the task more challenging and might require more development time.

results of direct detection method

Testing of the multi-point vibration source with a QPS long-gauge vibration sensor clearly shows well-defined sinusoidal waveform. This compares with the waveform displayed by the direct detection phase OTDR method. The waveform at low frequency looked distorted, a sign showing the sampling rate might be inadequate.

Phase ODTR looks promising. It is a technology requiring speed enhancement to get up to a suitable sampling rate and high speed data processing in real time. Such improvement might be called for before performing a third lab test. That will involve the development of hardware signal processing to implement the wave analysis currently done in Labview and also further simplifying the system configuration to reduce the system selling price to about $30,000.

Notes

1Zengguang, Qin, Liang Chen, and Xiaoyi Bao, “Continuous Wavelet Transform for Non-stationary Vibration Detection with Phase-OTDR,” Optics Express, Volume 20, No. 18, 2012, pages 20459-20465.

2Zengguang, Qin, Liang Chen, and Xiaoyi Bao, “Wavelet Denoising Method for Improving Detection Performance of Distributed Vibration Sensor,” IEEE Photonics Technology Letters, Volume 24, No. 7, 2012, pages 542-544.

3Zengguang, Qin, Tao Zhu, and Xiaoyi Bao, “High Frequency Response Distributed Vibration Sensor Based on All Polarization-Maintaining Configurations of Phase-OTDR,” IEEE Photonics Technology Letters, Volume 23, No. 15, 2011, pages 1091-1093.

4Yuelan, Lu, Tao Zhu, Liang Chen, and Xiaoyi Bao, “Distributed Vibration Sensor Based on Coherent Detection of Phase-OTDR,” IEEE Journal Lightwave Technology, Volume 28, No. 22, 2010, pages 3243-3249.


Peter Kung is president of QPS Photronics Inc. Zengguang Qin is a PhD candidate, Wenhai Li is a post-doctoral fellow and Xiaoyi Bao is a professor with the Department of Physics at the University of Ottawa, Ontario, Canada.

This article has been evaluated and edited in accordance with reviews conducted by two or more professionals who have relevant expertise. These peer reviewers judge manuscripts for technical accuracy, usefulness, and overall importance within the hydroelectric industry.

Previous articleRehabs Squeeze More Power from Existing Plants
Next articleSticky Wickets: Repairing Failed Stator Core Bolts at Thompson Falls

No posts to display