METHOD AND APPARATUS FOR LOCATING PARTIAL DISCHARGE IN ELECTRICAL TRANSFORMERS
This application claims priority from U.S. Provisional Patent Application Serial No. 60/009,264, filed December 27, 1995. BACKGROUND
Field of the Invention
The present invention relates to methods and apparatus for locating partial discharge sources in oil-filled electrical transformers.
Background of the Invention
Partial discharge (PD) refers to an electrical charge which is emitted inside the transformer due to a failure of its insulation properties This failure can occur due to the presence of a particle contaminant, an air bubble in the oil- soaked paper, or a physical breakdown of the insulating paper. Such a contami¬ nant or air bubble provides a conductive path which can sustain a spark, but not a complete arcing, hence the term partial discharge. PD is indicative of, or can lead to, serious insulation problems that can ultimately cause the transformer to mal¬ function or fail. Transformer manufacturers and owners routinely monitor for the occurrence of partial discharge by either electrical, acoustic or chemical means
Partial discharges can be detected electrically by measuring the voltage or the current off of the cap taps on the transformer. When partial discharges occur, pressure waves are generated in the surrounding oil which propagate in the transformer tank. These pressure waves can be measured as
acoustic signals. Additionally, the occurrence of PD releases chemicals into the oil due to the breakdown of the insulating paper. PD can be detected by measuring the quantities of these chemicals in the transformer oil. This last technique is called dissolved gas-in-oil analysis (DGA).
Once the occurrence of PD has been determined, locating the source is usually necessary for solving the problem. In the past, both electrical and acoustic measurements have been used to locate PD sources. When transformers are manufactured, or brought back mto a service center, electrical measurements such as radio influence voltage (RIV) and apparent charge are used to detect partial discharge, as discussed in "IEEE Trial-Use Guide for Partial Discharge Measurement in Liquid-Filled Power Transformers and Shunt Reactors," ANSI/IEEE C57.133 , Institute of Electrical and Electronics Engineers, Inc., New York, NY, January, 1988. A broad source location is then achieved by taking measurements on all high and low voltage cap taps, and noting where the electrical signal is strongest This technique can narrow the location down to a particular lead. To narrow the source location even further, a microphone is sometimes used to listen for the acoustic waves caused by the PD. The distance in the arrival time between the electrical and acoustic signals is used to determine the distance between the source and the microphone.
The field environment for operating transformers typically has too much electrical noise for accurate RIV or apparent charge measurements, and hence, some have turned to the use of multiple acoustic sensors to locate PD
sources as descπbed in, for example, the following references:
S.L. Jones, "The Detection of Partial Discharges m Power Trans¬ formers Using Computer Aided Acoustic Emission Techniques," Conference Record of the 1990 IEEE International Symposium on Electrical Insulation, Toronto, Canada, June 3-6, 1990, pp. 106-110 ("Jones");
T. Liang, W. Weilin, "Acoustic Location of Partial Discharge in On-Line Power Transformer," Sixth International Symposium on High Voltage Engineering," New Orleans, LA, paper 22-09, August 28 - September 1, 1989, pp. 1-4 ("Liang"), M. Thibault-Carballeira, et. al., "Fault Detection and Location in Transformers,"' International Conference on Large High Voltage Electric Systems, 12-01, September 1-9, 1982 ("Thibault-Carballeira"); E. Howells, E.T. Norton, "Location of Partial Discharge Sites in On-Line Transformers," IEEE Transactions on Power Apparatus and Systems, v. PAS-100, n. 1, January 1981, pp. 158-161 ("Howells"); and J.H. Carpenter, J.S. Kresge, CB. Musick, "Ultrasonic Corona Detection in Transformers," IEEE Transactions on Power Apparatus and Systems, v. 84, August 1965, pp. 647-651 ("Carpenter"). This approach involves determining the arrival time of the acoustic
PD signal at a three different sensors, and triangulating to back-figure the source location.
Alternatively, two sensors are used, as in Howells and Norton, to locate a plane in which the source lies. This plane is located by moving the two sensors around on a single side of the transformer until the arrival times at both sensors are the same. In this configuration the source is equidistant from both sensors and lies on the plane that crosses half way between the two sensor locations. This process is repeated to find another plane perpendicular to the first. The intersection of the two planes defines a line containing the source. The process is repeated another time to find the locations of the source along this line. This process, although mathematically simple, can be difficult to implement because it requires PD signals to be accurately measured at many locations on the tank. Often it is hard to find one spot on the tank that exhibits a clear PD signal, and relying on measurements m multiple locations can compound measurement error. Additional efforts have been made in clarifying the arrival time of the acoustic signal using signal averaging, as discussed in the Jones and Howells articles referenced above.
Still others have worked to develop transfer functions for electrical or acoustic signals in transformers, by calibrating the measured output to a known input, and then applying this transfer function to the measured response from a real PD source, as described in:
R.E. James. B.T. Phung and Q. Sum 'Application of Digital Filtering Techniques to the Determination of Partial Discharge Location in Transformers," IEEE Transactions on Electrical Insulation, Vol. 24,
No. 4, August 1989, pp. 657-668 ("James I"); R.E. James. F.E. Trick, B.T. Phung and P.A.A White. "A Micro¬ processor Based System for the Electrical Location of Partial Discharges, "Example of Application to a Large Transformer Winding, ' Cigre Symposium, Vienna, May, 1987, pp. 1-5 ("James II"); and in Thibault-Carballeira.
The prior art references discussed above assume that PD acoustic signals are made up of frequency components between 20 kHz and 500 kHz, with most of the energy concentrated around 150 kHz. For example, one PD acoustic monitoring system listens only for components at 150 kHz. However, such limited bandwidths result m false negatives, i.e., failures to detect actual PDs Higher frequency components at 10 MHZ and at 50 MHZ are discussed m James I
Distinctions have been made between the frequency content of partial discharge and the frequency content of other acoustic noises coming from an operational transformer Howells has identified noise coming from the core steel material centered around 40 kHz to 80 kHz. SUMMARY OF THE INVENTION
The partial discharge (PD) locator locates the PD source after the presence of PD has been established by electrical detection. A schematic of the overall system is shown in Figure 1 The main components of the PD Locator are the two or more ultrasonic acoustic sensors, signal conditioning hardware, data acquisition hardware and software, and signal processing software. The sensors
are sensitive to the ultrasonic range of acoustic frequencies. The data acquisition hardware is, for example, a PC-based acquisition board with additional processing and analysis capabilities such as filtering and signal averaging. The sensors are connected to the signal conditioning and data acquisition hardware through low noise shielded cables. The signal processing software controls the data acquisition board and executes the subsequent signal processing The hardware may also include a plug for headphones, so that a user can listen to PD components in the audible range
The presence of partial discharge is detected by radio influence voltage (RIV) or apparent charge measurements made independently by the user. Output from either of these electrical measurements is used as an external trigger to the PD Locator system. The electrical trigger signals the occurrence of a partial discharge. When the system is triggered, measurements are taken at each of the sensors to record the acoustic signals that propagate from the PD source through the oil (or other liquid, or, for a dry-type transformer, through air) to the tank wall. The sensors are coupled to the tank wall using a lubricating grease, e.g., petroleum jelly such as Vaseline™ or Moly E.P. multipurpose grease, which enhances transmission of out-of-plane waves in the tank wall, and minimizes transmission of in-plane waves. The waves which travel directly from the source to the receiver are out-of-plane waves Accordingly, the arrival time of out-of-plane waves can be used to calculate the distance from the source to the sensor.
The speed of acoustic waves (the "wave speed") in the oil (or other transformer liquid or. for a dry-type transformer, the wave speed in air) is entered by the user. The wave speed is roughly constant over a wide range of frequencies. The present invention includes charts of wave speed as a function of oil (or other fluid) temperature as a guide. The system calculates the time delay from the source to each sensor and converts it into a distance based on the wave speed in the oil (or other fluid).
A bandpass filter is applied to the acoustic signal to eliminate operating noise on the low frequency end, and electrical noise and aliasing on the high frequency end. The sampling rate and duration for the acoustic signal are fixed for the normal operating mode. The sampling rate is based on the 40 kHz to 300 kHz frequency band and the duration is based on the storage capabilities of the on-board memory of the data acquisition card. The PD locator also has a "long distance" mode which samples at a slower rate, thereby lengthening the duration to capture arrival times at sensors placed far apart.
Once started, the PD locator repeatedly samples and averages the signals to clarify the arrival time, i.e., to better define the arrival time by improving the signal-to-noise ratio. During the averaging process, the display is continuously updated to show the running average of the arrival time. The running average is stored in the PC memory, while the newly acquired signal to be averaged next is stored in the acquisition board. When the user executes the "Calculate Distance" command, the averaging stops and the PD locator positions a
cursor at the start of each acoustic signal and calculates the distances from the source. The user may override the PD locator's estimation of the arrival time by positioning the cursors manually and executing the "Calculate Distance" command.
The input gain is variable to avoid clipping. The input gain can be set manually, or automatically using the "Auto Gain option.
The PC memory stores two traces that can later be recalled to the screen for comparison The system also has a fast Fourier transform (FFT) function When the FFT is run, the time histories are saved in PC memory
These and other objects of the present invention are described m greater detail in the Detailed Description of the Invention and the appended drawings. BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 is a schematic of the PD locator mounted on the trans¬ former This figure shows a typical arrangement of the acoustic sensors placed on the transformer, and the electrical and acoustic inputs feeding into the data acquisition box
Figure 2 is a diagram of the data acquisition and signal processing display and control panel showing the functions available on the PD locator
Figure 3 is a flow chart depicting the acquisition of a single sample and the subsequent processing
Figure 4 is a drawing of the spring-loaded magnetic mount used to hold the acoustic sensors to the tank wall.
Figure 5 is a flow chart of the process for setting the auto gain.
Figure 6a is a schematic diagram of the signal conditioning board.
Figure 6b is a schematic diagram of the data acquisition board. DETAILED DESCRIPTION OF THE INVENTION
The present invention is a system and method for locating partial discharge in transformers at the service station and at test facilities. It improves upon existing systems for locating partial discharge by using both electrical and acoustic detection procedures in a data acquisition and signal processing system that automatically calculates the distances between the sensors and the source location.
The following description of the present invention applies specifically to PD location in oil-filled or other liquid-filled electrical transformers. However, the system and method may be applicable for use with other liquid-filled or dry- type electrical equipment. A. Layout
The main hardware components of the PD locator device are a PC-based signal conditioning, data acquisition, and signal processing system, two or more ultrasonic acoustic sensors, and two low-noise shielded cables. Figure 1 is a schematic of these main components, showing a typical arrangement for their use on the transformer. Two acoustic sensors 12 are placed on the tank wall to
measure the acoustic emission from the transformer. The two sensors have similar calibration curves, so that they produce similar amplitudes for the same input signal. The signal is carried to data acquisition box 11 through shielded, low noise cables 13, e.g., Belden 9223 low noise coaxial cable, 50 ohm RT. The external trigger 14 comes from the electrical PD monitoring device 15 and is plugged into the data acquisition box. B. Sensor Specifications
The ultrasonic acoustic sensors, e.g., piezoelectric sensors such as Dunnegan SE 900 sensors, used in the present invention are sensitive in the range from 40 kHz to 300 kHz . This frequency range extends lower than that typically used in the prior art, because the frequency content of partial discharge, particularly in new transformers, can be significant at frequencies between 40 kHz and 100 kHz. Therefore, concentrating the measurements at 150 kHz, as in some prior art systems, may not register PD signals generated by all types of PD sources The response spectra of the two sensors need not be completely flat in this frequency range, as long as they are reasonably sensitive throughout the range. It is also necessary that the spectra from the two sensors be similar, so that they produce similar responses to the same input "Similar" is defined as a difference of less than about 5dB, preferably less than about 3 dB in the calibration curves throughout the frequency range.
C Data Acquisition and Signal Processing Hardware
Figure 2 shows a diagram of the data acquisition and signal pro¬ cessing system with all of its components and functions. The viewing screen displays two channels of data 21 which correspond to the input signal coming from the two acoustic sensors The screen 22 also shows the sampling rate, the arrival time, and the distance to the source for each of two sensors 12 The two distances are obtained by processing the data as described below
Blocks 23 through 28 represent the data acquisition and processing functions The "Start" button 23 starts the signal averaging, acquiring a new signal with each electrical trigger. The "Calculate Distance" button 24 stops the averaging and calculates the arrival times and distances between the source and the sensors that are shown on screen 22 If the "Start" button is hit again, the averaging continues, incorporating new signals into the existing average. To start the averaging over again, the "Reset" button 25 is used to clear the buffers before the start function is executed The "FFT" button 26 calculates the fast Fourier transform of the two averaged signals on the screen. The FFT function displays the frequency spectra on the screen and saves the time histories to a buffer
The FFT function is used to examine the frequency content of the acoustic signals. Partial discharge signals tend to be fairly consistent in their frequency content, with most of the energy ranging from 80 kHz to 200 kHz. An experienced user can use the frequency content display to distinguish between signals due to partial discharge and those due to other phenomena.
A pair of signals on the screen can be saved to a buffer by using the "Save" button 27, and then recalled to the screen at a later time using the "Recall" button 28. Hitting the "Recall" button again brings back the screen display that was present at the time the saved signals were recalled When m FFT mode, the "Recall" button 28 is used to toggle the screen display between the spectra and the time histories
Blocks 29 through 32 represent the controls used to set the acquisition, display and calculation parameters. The input gam can be set manually, using dial 29, or automatically, using "Auto Gam Function" button 30 (which samples a number of triggers and adjusts the input gain to make full use of the dynamic range of the system, while avoiding clipping). The "Wave Speed Control" 31 is used to input the value of the speed of sound in the transformer oil, or other fluid As noted above, the instruction manual for the present invention includes charts of wave speed as a function of oil, or other fluid, temperature "Long Distance" switch 32 reduces the sampling rate to increase the data acquisition time The sampling rate is set to capture frequencies m the range from 40 kHz to 300 kHz. and to capture arrival times at both sensors when they are placed a maximum of 10 ft apart. The "Long Distance mode sacrifices some of the high frequency resolution, to detect arrival times at sensors placed farther apart This function is used to determine initial sensor placement, and is then turned off for more accurate distance calculations
Blocks 33 through 35 represent the input and output ports The
input from the sensors are plugged into the ports 33 labeled "Channel 1" and "Channel 2". The electrical trigger is plugged into the "Trigger" port 34 The "Audio Output" port 35 is used to listen to the audible components of the acoustic signals. Headsets are plugged into this port. The audio output is the unfiltered acoustic signal measured by the sensor, which would include audible frequencies. This feature is used to determine initial sensor placement However, if the signals are filtered at the sensors themselves the audible frequencies would be filtered out, and the "Audio Port" 35 would not be included
The data acquisition hardware consists of an IBM-compatible personal computer, with a storage drive, a color monitor for viewing the acquired data, and a keyboard and a mouse. The keyboard and the mouse can be used to change the system configurations. The microprocessor in the IBM-compatible personal computer is an Intel 486 or Pentium microprocessor, or equivalent. The signals received from the sensors are supplied to a signal conditioning board As shown in Figure 6a, the signals are amplified by amplifier 51. and filtered using low-pass anti-aliasing filters 52 at about 300 kHz. then high-pass filters 53 at about 40 kHz, using standard filters. The signal conditioning board also contains conditioning for the trigger signal (components 54, 55, 56 and 61) and the audio output ack 35. channel selector knob 60, and the audio gam adjuster knob 59 The two conditioned input signals and the trigger signals are then supplied to data acquisition board 70 (Figure 6b)
As shown in Figure 6b, on the data acquisition board 70 the signals
are first amplified by a programmable gain amplifier 71. The two input signals are then simultaneously sampled by a pair of sample-and-hold amplifiers 72. and then supplied to multiplexer 73. Alternatively, the multiplexer 73 could be replaced with two A/D converters (not shown) using methods known to those skilled in the art. The multiplexed signals are then digitized using high speed (approximately 10 megasamples per second) 8 or 12 bit A/D converter 74. The output from A/D converter 74 is then supplied to storage buffer 75. Storage buffer 75 allows sequential samples to be acquired and stored. Storage buffer 75 must have at least enough memory to store five thousand samples.
As new data are acquired, they are transferred to the IBM-compatible personal computer, for subsequent processing. The personal computer controls all functions and operating parameters of the data acquisition card, including signal gam. acquisition sample rate, sample count, and the timing of the data acquisition and data transfer.
D. Data Acquisition and Signal Processing Software and Algorithm
Figure 3 shows a flow chart of the acquisition of a sample and the subsequent processing. Steps 301 through 303 are preferably performed at the sensor to maximize the signal-to-noise ratio, although some of these steps may be performed by the data acquisition and signal processing apparatus, if necessary. The acoustic sensor measures a signal off of the tank wall (step 301). It is then sent through a pre-amplifier (step 302) which is driven by a power supply. The amplified signal is then sent through a bandpass filter (step 303) to filter out
operating noise of the transformer on the low end. and aliasing and electrical noise on the high frequency end The frequency bandwidth ranges from 40 kHz to 300 kHz.
When the "start" command is executed (step 304), the software waits for the external trigger and acquires a sample of data (step 305) from the acoustic sensor upon detecting a trigger If the buffer is not empty, the newly acquired signal is averaged in (step 306) and the new average is saved to the buffer (step 307). If the buffer is empty, the new signal is stored in the buffer This acquisition and averaging process is repeated until the "calculate distance" command is executed (step 308) At this point, the averaging stops and the software finds the arrival time of each of the two signals, (step 309) The arrival time is multiplied by the wave speed in the oil (step 310) to calculate the distance between each sensor and the source This calculation is then output and is displayed on the screen (step 311)
The auto gain function determines the optimal input gain This process is illustrated for a single channel in Figure 5 In step 501 of Figure 5 the input from the acoustic sensor is triggered randomly, with an initial setting of the input gain. In step 502. the percentage of the dynamic range used by the peak-to- peak value of the acquired signal is calculated. For example, assume the input gam is adjusted so that 10 mV spans the full dynamic range If the triggered signal has a maximum value of +2 5 mV and a minimum value of -2 5 mV, then the peak-to-peak value is 5 raV, which is 50% of the dynamic range. As shown in
steps 503 and 504, if the peak-to-peak value is within the dynamic range, the input gain is adjusted such that the peak-to-peak value is roughly 70% of the dynamic range As shown in step 505, if the peak-to-peak value is greater than the dynamic range, such that the signal is clipped, then the input gam is halved until the signal falls within the dynamic range
A signal is then triggered five more times to insure that all acquired samples are within the dynamic range (steps 506 through 510) If any sample is outside the dynamic range, then step 505 is repeated, to readjust the input gain Once five samples in a row have been acquired within the dynamic range, the input gain is set for that channel
Although the above disclosure discusses the invention as a stand-alone housing, the foregoing invention can be also be implemented as a custom-made board and related software that is run on a personal computer using methods known to those skilled in the art When the device is implemented on a PC the buttons and dials depicted in Figure 2 are all "virtual " I e created on the computer display The functionality of the invention remains the same, with the exception that the input ports 33 through 35 might be located on a separate junction box that is external to the PC
E Magnetic Mounts
The sensor is mounted to the transformer tank wall using the magnetic mount shown in Figure 4 The mount is spring loaded to provide constant, uniform coupling between the sensor and the tank wall A torqumg
screw 41 is provided for fine adjustments. The torquing screw controls a piston which pushes on the spring-loaded mount 42. The mount attaches to the tank wall by two magnets 43. The design of this mount allows for airflow around the sensor, to prevent heat build up. This design also permits the sensor to be easily dismounted and moved to a new location on the tank wall. F. Examples of Use
The following examples are provided to demonstrate how to practice the present invention. They are not to be construed as limiting the invention in any way.
Example 1: General Operating Procedure
The user begins by placing the sensors in various locations and using the headphones to determine the vicinity in which the noise is the loudest. The location of the loudest noise is not necessarily the location closest to the source of the noise. It is the location which has a clear unobstructed path between the source and the sensor. For this initial determination, the user would listen, for example, in the top and bottom halves on four sides of the transformer. The goal of this preliminary step is to define a region in which two sensors can be placed, preferably no more than 10 feet apart.
The user then sets the input gain manually, or by using the auto gain feature, and enters the wave speed parameter. Next, the user pushes the start button to begin the signal averaging and watches the averaging on the screen. When the arrival time looks clearlv defined, the user executes the "calculate
distance" command. The user then moves the sensor that is farther away from the source to a new location, leaving the closer sensor in place as a frame of reference. This process is repeated until moving the sensors no longer gets you any closer to the source At this point (if not before) the sequential measurements will have defined a reasonably small area in which the source is located. To confirm the location, the entire process can be repeated starting from a different section of the tank wall
Example 2 Using the Long Distance Mode
A variation on Example 1 is to use the long distance mode instead of the headphones to locate the general region where the acoustic emission is the loudest In this example, the user starts by placing each sensor on a different side of the tank. The device is put into the long distance mode, and then the signal averaging is performed as usual until the arrival time is clear enough to perform the distance calculation This process is repeated until the region in which the acoustic emission is loudest has been identified and the sensors can be placed within ten feet of each other in that region. Then the device is taken out of the
"long distance" mode, and the source location proceeds as described in Example 1
Example 3 Using the FFT to Differentiate Between PD and Other Noise Sources
Other sources of acoustic noise can sometimes mask the partial discharge signal. For example, the operation of the transformer can produce noise between 40 kHz and 60 kHz The user runs the "FFT" function to view the fre¬ quency spectra of the measured signals If the signal appears to be dominated by
frequency components at the low end of the bandwidth (around 40 kHz) or the high end of the bandwidth (around 300 kHz) then the sensors may be locating on a noise source that is not partial discharge. The user then moves the sensors around to compare the frequency spectra at different locations to see where the frequencies are concentrated away from the ends of the bandwidth
The foregoing disclosure of embodiments of the present invention has been presented for the purposes of illustration and description It is not intended to be exhaustive or to limit the invention to the precise forms disclosed Many variations and modifications of the embodiments described herein will be obvious to one of ordinary skill in the art in light of the above disclosure