Method for detecting partial discharges and system for diagnozing an electrical apparatus

(54) Title: METHOD FOR DETECTING PARTIAL DISCHARGES AND DIAGNOSTIC SYSTEM FOR ELECTRICAL APPLIANCE (30) Unionist Priority: 05/06/2003 fr 03 06808 (71) Depositor (s): Areva t & d sa (fr) (72) Inventor ( es): Raja Kuppuswamy (74) Attorney: Dannemann, Siemsen & Ipanema Moreira (57) Abstract: METHOD FOR DETECTING PARTIAL DISCHARGES AND DIAGNOSTIC SYSTEM FOR ELECTRICAL APPLIANCE: The present invention relates to a method for detecting partial discharges and a diagnostic system for electrical devices. The invention provides a method for detecting partial discharges in an isolated electrical device in a receptacle and a system for diagnosing the state of the device using this method. With the help of. an.ÜHF antenna placed on the device's receptacle, the method consists of analyzing 6 spectrum (RS1) of the electromagnetic signal captured by the antenna and identifying within the spectrum one or more frequencies of interest (B1., B2), for each urethra of which the signal has a greater amplitude than a predetermined limit value. To identify the frequency / frequencies of interest, the spectrum of the signal received by the antenna is compared with a reference spectrum (RS2).

<img file="BRPI0401902A_D0001.tif" />

·· ··· ·· · ·· · ·· ·· ·· ·· 9 · ······ · · ···· · · ··· ·· · ·· ··· ·· ·· ····

Descriptive Report of the Invention Patent for METHOD FOR DETECTING PARTIAL DISCHARGES AND DIAGNOSTIC SYSTEM FOR ELECTRICAL APPLIANCE.

The present invention relates to the detection of partial electrical discharges that occur in high voltage electrical appliances isolated in a receptacle, such as an isolated transformer in a dielectric liquid. The detection of partial electrical discharges allows the insulation failures in the electrical appliance to be diagnosed and its lifetime to be estimated.

In classic terms, the detection of the presence of partial electrical discharges in the receptacle of such an electrical device is based on the measurement of electromagnetic radiation inside the receptacle using a UHF antenna. This method allows a diagnosis of the electrical appliance without having to interrupt its operation.

However, the electromagnetic radiation captured by the UHF antenna includes a significant amount of electromagnetic noise that impairs the reliability of such measurements. This noise, which is due, among other things, to coronal discharges that occur in the environment of the electrical appliance, is difficult to assess, since it depends on the environment of the electrical appliance being tested, its load and other parameters. Electromagnetic noise varies greatly from one device to another, and this makes it difficult to isolate partial discharges.

US Patent Document No. 2002163344 demonstrates how to perform a spectral analysis of the electromagnetic signal captured by the UHF antenna by comparing the maximum signal amplitude at the limit values ​​for one or more predetermined frequencies. This method for detecting partial discharges in the receptacle of an electrical appliance tends to separate the noise component from the signal received by the antenna. However, as this noise is very different from one device to another, the detection accuracy remains inadequate. In this document, it is suggested that a comparison be made using a network of neurons. Since the variation in electromagnetic noise is important, resorting to a ······ ·· · ·· ··· ·· ···· ·· ·· • · · · · · 2 · neural network does not seem likely improve detection reliability.

Patent Number WO-0118554 describes a device that includes an ultrasound sensor and a flat capacitor plate for detecting, respectively, an ultrasound pulse and a radio frequency pulse generated by a partial discharge. These sensor signals are analyzed by signal processing to determine the appearance of discharges. When a discharge appears on a device, it produces an ultrasound pressure wave that moves according to the classical laws of physics and is captured by the ultrasound sensor 2. If a radio frequency pulse has been detected in a predetermined period of time before the ultrasound signal, a partial discharge is considered to have actually occurred. Nevertheless, this combination of ultrasound and radio frequency waves is not independent of noise. Several sources of noise can disturb the measurements, the most important of which comes from the magnetostriction of the transformer core that generates continuously ultrasound.

In another recognized measurement method, an RLC resonator circuit is connected to an insulating part of the electrical appliance, and a capacitor is placed in series with the resonator circuit to form a discharge loop. An impedance usually isolates the discharge loop from the high voltage source. The RLC circuit has a resonance range located between 30 kHz and 300 kHz and is configured to provide a response signal with a preset frequency that can be easily detected by a measuring device. This method is adapted for tests with low electromagnetic noise, such as in a laboratory. In fact, the resonance circuit provides the same response for all pulse signals, although it is not possible to distinguish the partial discharge signal from the noise.

In another recognized method, each partial discharge signal is registered by a device with a wide bandwidth, such as a power transformer for radio frequencies. The results of this type of measurement are recorded on a submicrosecond time scale, which corresponds to a frequency band that reaches

<img file="BRPI0401902A_D0002.tif" />

MHz verses.

The operating frequency band of several MHz does not exclude noise due to coronal discharges whose frequency band extends far beyond (up to approximately 200 MHz).

US Patent Number 6313640 describes a method consisting of two sensors to determine whether partial discharge signals originate from the apparatus receptacle or an external source. A first sensor is connected in the current supply line to the transformer. A second sensor is located at the output of the transformer. The signals from the two sensors are passed to a comparison unit that calculates the difference between these two signals so that the resulting signal represents the partial discharges that occur within the device. This method requires the device to be taken off line to calibrate the comparison unit before proceeding with the diagnosis, sending standard signals to the device and adjusting the comparison unit gain to obtain a zero signal. On the other hand, the sensors used are sensitive in a frequency band that extends up to 10 MHz, which includes the frequencies that correspond to high amplitude coronal discharges.

In another method, known from US Patent Document No. 6323655, antennas are installed in the apparatus receptacle to detect partial discharge signals in the UHF frequency range. The measurement system is sensitive in one of the two bands 300 - 600 MHz or 600 MHz -1.2 GHz, which corresponds to frequency bands that are not disturbed by coronal discharges. However, these two frequency bands cover signals emitted by television and signals from mobile phone networks, respectively. Consequently, these higher frequency bands allow coronal discharges to be eliminated but contain other sources of electromagnetic noise that significantly disrupt the measurement.

The purpose of the invention is to propose an improved method for detecting partial discharges.

To this end, the invention involves a method for detecting des ·· ·· ·· · ·· · ·· ··· ·· • · ···· · · ······ · · ······ ♦ ·· ·· ··· ·· · ·· 4 ** partial loads in an isolated electrical device in a receptacle, by using a UHF antenna located in the receptacle, and consists of analyzing the spectrum of the electromagnetic signal captured by the antenna and identifying in the spectrum one or more frequencies of interest for each of which the signal has an amplitude that is greater than a predetermined limit value, whereby, to identify the frequency / frequencies of interest, the spectrum of the signal received by the antenna is compared with a reference spectrum.

Furthermore, a method according to the invention can offer the following characteristics:

- a frequency of interest is identified when the difference in amplitude for this frequency between the spectrum of the signal received by the antenna and the reference spectrum is greater than a predetermined limit value;

- a frequency of interest is identified when the amplitude ratio for this frequency between the signal spectrum received by the antenna and the reference spectrum is greater than a predetermined limit value.

The purpose of the invention is also to provide a method for detecting partial discharges in an isolated electrical device in a receptacle, using a UHF antenna located in the receptacle, and consists of analyzing the spectrum of the electromagnetic signal captured by the antenna and identifying in the spectrum several more frequencies of interest for each of which the signal has an amplitude that is greater than a predetermined limit value, whereby, to identify the frequency / frequencies of interest, the spectrum of the signal received by the antenna is compared with a reference spectrum as follows:

- for several frequency ranges predefined in the two spectra, a characteristic value of the interval for each spectrum is calculated, and

- the characteristic values ​​of the two spectra are compared, interval by interval.

The characteristic value of the range of each spectrum is, for

<img file="BRPI0401902A_D0003.tif" />

example, an average value of the signal amplitude in the range or the signal energy in the range.

According to a first embodiment of the invention, the reference spectrum represents an electromagnetic spectrum captured by the antenna when the electrical appliance is turned off or when the electrical appliance is first turned on.

According to a second embodiment of the invention, the reference spectrum represents an electromagnetic signal captured by a second UHF antenna placed in a metal enclosure close to the electrical appliance.

The invention also relates to a system determined to diagnose the isolation state of one or more electrical devices, each of which is isolated in a receptacle, said system comprising a data processing and control unit, equipped to implement the above method for detecting partial discharges, and which comprises several UHF antennas that supply a number of electromagnetic signals. These antennas are connected to the processing and control unit via a parallel series converter that includes a high frequency electromagnetic multiplexer that converts parallel flows of electromagnetic signals into a single electromagnetic signal stream.

The invention will now be described in greater detail with reference to the accompanying diagrams which illustrate an implementation form as an example.

Figure 1 is a schematic illustration of the conversion of the signal captured by the antenna in the form of a spectrum;

Figure 2 is a schematic representation of a convert connected to the antenna;

Figure 3 shows the representative spectra of a partial discharge signal and an electromagnetic noise signal;

Figure 4 is an illustration of the identification procedure according to the invention;

<img file="BRPI0401902A_D0004.tif" />

Figure 5 is a block diagram representing the first implementation of the process according to the invention;

Figure 6 is a block diagram representing a second implementation of the process according to the invention;

Figure 7 is a block diagram representing a third implementation of the process according to the invention;

Figure 8 is a block diagram representing a fourth implementation of the process according to the invention;

Figure 9 illustrates in a schematic way the structure of the parallel series converter that comprises an electromagnetic multiplexer used for processing the signals from the antennas;

Figure 10 illustrates in a schematic mode the power signals sent to the electromagnetic multiplexer;

Figure 11 illustrates an algorithm to identify the frequencies of interest using the method according to the invention.

In Figure 1, a partial discharge signal SD is shown in its representation of time 1 with amplitude A as a function of time t. This SD signal is split into 2 into its various sinusoidal components such as SS1 and SS2, to be converted into a frequency spectrum according to a representation 3. As is well known, this spectral representation provides the amplitude of each sinusoidal component of the SD signal as a function of its frequency F.

In the invention, this conversion of an SD time signal into an RS spectral representation would, for example, be performed with a signal converter, such as that shown in Figure 2.

In Figure 2, a signal converter receives the signal from a UHF antenna 4 at its input and emits the spectrum of the input signal. Antenna 4 is placed inside the sealed receptacle 5 of an electrical appliance, which here is a transformer with a coil 6 immersed in a liquid dielectric such as oil. The electrical apparatus 6 could also be a circuit breaker immersed in a dielectric gas within a sealed receptacle. The signal converter 7 is represented by a block diagram in which the

<img file="BRPI0401902A_D0005.tif" />

input signal received by antenna 4 is attenuated by an adjustable resistance 7a, then filtered through a low-pass filter 7b to eliminate the higher harmonics, before being passed to a mixer 7c. The other input of mixer 7c receives the signal from an oscillator 7e, which is excited by a ramp generator 7d to generate a signal that progressively scans the frequencies of interest in the increasing direction. The signal obtained from the output of mixer 7c thus represents the amplitudes of the frequencies of interest of the signal received by the antenna 4. The signal provided by the mixer 7c is filtered and amplified in blocks 7f, 7g and 7h to direct the input to an oscilloscope 7i, a input x of this oscilloscope being directed by the pulse ramp 7e in order to form the spectrum of the input signal represented in block 7i. Such a converter could have good use in implementing the method of the invention, as detailed below.

In Figure 3, a graph shows two spectra RS1 and RS2 that represent, respectively, a partial discharge detected in the receptacle of an electrical device and an electrical noise that covers this electrical device for diagnosis. More specifically, the first RS1 spectrum represents a signal captured by antenna 4 during a partial discharge. The second RS2 spectrum is a reference spectrum that represents, for example, a signal captured by antenna 4 when the electrical appliance is not generating a partial discharge, for example when the electrical appliance is turned off. For this reason, the first RS1 spectrum represents partial discharge signals added to electromagnetic noise signals, while the second RS2 spectrum represents only electromagnetic noise signals. Figure 3 demonstrates that RS2 electromagnetic noise has an amplitude that varies enormously as a function of frequency. Specifically, electromagnetic noise is practically zero in two frequency bands denoted B1 and B2 and located between 300 and 350 MHz and between 800 and 850 MHz respectively. Electromagnetic noise is also zero for frequencies located above 1 GHz.

According to the invention, the RS1 spectrum of the signal received by antenna 4 is compared with the RS2 reference spectrum in order to ·· ·· ·· · ·· · ·· ··· ·· • · ····· ·· · · ···

<img file="BRPI0401902A_D0006.tif" />

identify one or more frequencies of interest for which the signal and noise satisfy a choice criterion. The criterion for identifying the frequencies of interest might, for example, consist of retaining those frequencies for which the amplitude ratio of RS1 and RS2 is greater than a predetermined value, which corresponds to the signal / noise ratio. If preferred, this criterion may also consist of calculating for a frequency the difference in amplitude between the RS1 spectrum and the RS2 spectrum in order to identify this frequency as a frequency of interest if this difference is greater than a predetermined threshold value, as illustrated in Figure 4.

The identification of frequencies of interest can also consist of identifying the frequency bands of interest. In Figure 4, the difference in spectra is always greater than -70 dB, corresponding to a context in which the noise is very weak and for which all frequencies could be identified as frequencies of interest. The choice of a limit value of -55 db leads to the identification of frequency bands B1 and B2 as frequency bands of interest, B1 and B2, corresponding respectively to frequencies located between 300 and 400 MHz, and between 450 and 550 MHz.

The RS2 reference spectrum can represent the signal received by antenna 4 when the electrical device is not generating a partial discharge, this spectrum resulting from reference measurements, for example, the first time the device is turned on or when it is turned off. Therefore, the reliability of the method is independent of the particular characteristics of the device and its environment, since these are taken into account when reference measurements are made.

This RS2 reference spectrum can also result from a signal captured by a second antenna located in a metal enclosure placed in the vicinity of the electrical device to be diagnosed. This second antenna is specially placed in a metallic enclosure located close to the electrical device to be diagnosed, so that the signal captured by the second antenna represents only the electromagnetic noise of the

<img file="BRPI0401902A_D0007.tif" />

electrical appliance to be diagnosed. In fact, any partial discharges that occur in the receptacle of the electrical appliance to be diagnosed, the electromagnetic radiation that they generate is confined within the receptacle of the electrical appliance, so that these discharges are not captured by the second antenna. Similarly, the reliability of the method is independent of the environmental characteristics of the electrical device to be diagnosed, as these are taken into account by the reference signal that is received by the second antenna.

As schematically shown in Figure 5, the method of the invention can be implemented by connecting the antenna 4 via a coaxial cable to the input of a signal converter 7, which is itself connected to a processing and control unit 8. The converter 7 is driven by the unit 8 to provide a spectral representation of the signal received by the antenna 4 in one or more frequency bands that are transmitted to it by the unit 8. Unit 8 here can be a portable computer, for example.

During the identification stage, converter 7 is activated by unit 8 to supply an RS1 spectrum of the signal received by antenna 4 over a wide frequency range, such as for example from 200 MHz to 1.5 GHz. This RS1 spectrum is registered in unit 8 to identify one or more frequency bands of interest by comparing it with an RS2 reference spectrum pre-registered in unit 8. The converter 7 is then activated by unit 8 to supply a spectrum of the signal received at antenna 4 for only the frequencies of interest, this spectrum being registered in unit 8 for the detection of partial discharges. The detection of partial discharges then consists of detecting peaks with an amplitude greater than a predetermined limit value, in the spectrum of the converter.

Furthermore, the comparison of the RS1 spectrum of the signal received by antenna 4 with the reference spectrum RS2 includes a split stage. This division consists of dividing the frequency range into several regular frequency intervals and calculating the average value of the spectrum for each

<img file="BRPI0401902A_D0008.tif" />

10 * range. This division is applied separately in the spectrum that represents the signal received by antenna 4 and in the reference spectrum. The identification of the frequency bands of interest then consists of comparing, for each frequency range, the average value for the antenna spectrum with the average value for the reference spectrum, to verify whether or not this frequency band meets the comparison chosen. Processing time can be reduced by increasing the size of the intervals and, conversely, the precision can be increased by reducing the size of these intervals. This comparison may also consist of comparing, for each frequency range, the signal energy for each spectrum. In fact, for a given interval, the signal energy in that interval is of a more significant size than the average signal amplitude value in that interval.

In the modality shown in Figure 6, a data processing and control unit 8 is connected to several antennas 4, 4 ', 4, in order to diagnose the presence of partial discharges simultaneously in several electrical devices. In this example, two electrical devices 5, 6 and 5 ', 6' are each equipped with UHF antennas, these antennas being connected by coaxial cables to a signal converter 7 via a parallel series converter 9. The parallel series converter 9 is driven by the unit 8, in order to select one of the three antennas to submit the signal it receives to the converter 7, as described later. For this reason, it is possible to identify frequencies of interest separately for each antenna in unit 8. After this identification, the unit 8 can drive in parallel the parallel series converter 9 and the signal converter 7 to select one antenna at a time, in order to monitor each antenna in the unit 8 in a cyclic mode.

In another modality shown in Figure 7, the method is used to permanently monitor an electrical appliance of a specialized unit permanently installed. Here, unit 8 is connected to an alarm device 10, such as an audible alarm that is activated when the partial discharge detection process crosses a certain

<img file="BRPI0401902A_D0009.tif" />

limit. More specifically, the parallel series converter 9, the signal converter 7, the unit 8 and the alarm 10 are grouped in a specialized unit 11, which is installed permanently or for a predetermined period on site, for example near the electrical appliance to be diagnosed. Unit 8 can be reduced to a microprocessor programmed to drive the parallel serial converter 9 and the signal converter 7, with the possibility of communicating with an external medium such as a computer 12 via a network connection for data exchange . Several specialized units 11 can be activated by this central computer by means of which an operator can obtain the data of each specialized unit 11 in a centralized mode, and parameterize the operation of each specialized unit remotely.

A similar specialized unit 11 can also be connected to several electrical devices 5 and 5 'under supervision, as shown schematically in Figure 8. Similarly, this specialized unit 11 is driven by a computer 12 which is programmed to supply the operating parameters for the parallel serial converter 9 and to signal converter 7 via central unit 8.

In Figures 6 to 8, the parallel series converter 9 combines the streams of parallel input signals produced by the antennas, such as 4, 4 'and 4, into a single stream by using a time signal or cut-off signal produced by unit 8, and defines the multiplexing time windows of the input signals, the number of time windows in a cycle corresponding to the number of antennas managed by unit 8.

Figures 9 and 10 illustrate an example of the implementation of the parallel series converter 9 which includes a high frequency electromagnetic multiplexer. Figure 9 shows three signals, SD4, SD4 'and SD4, produced by antennas 4, 4' and 4. The H signal is a time signal produced, for example, by an RS232 / RS422 / RS495 / USB serial port on the unit 8 and whose pulses define the multiplexing time windows. The H multiplexing signal is transformed by a generator 20 into a TTL-type power signal, indicated in Figure 9 by the HP reference, and then by the

12 * series / parallel 21 in a series of power signals C, C 'and C, which define all time windows shifted in time two by two and illustrated in Figure 10. These power signals C, C' and C serve as signals to control the relays of an electromagnetic multiplexer 22 receiving signals SD4, SD4 'and SD4. Since the power signals define the time-shifted windows, the input signals SD4, SD4 * and SD4 are divided and multiplexed into an SD output signal that is amplified in a high frequency broadband amplifier 23, having been sent to signal converter 7.

As indicated above, the process for detecting partial discharges can be monitored over time and normalized by experience in the form of data that represents the state of isolation of the device under research, and this allows a diagnosis to be performed on the device under research for scheduled maintenance.

More specifically, the trace of a signal after subtracting the noise is continuously recovered and its energy (surface area between the trace and the abscissa) is calculated and normalized; such normalization can correspond to the division of energy by a value determined by experience. The value that results from normalization can be used to index over time or classify the process to detect partial discharges and thereby the insulation status of the electrical device under study. This resulting value is compared with pre-established values ​​that correspond to pre-established levels of alarm activation, such levels of alarm activation indicating, respectively, the different isolation states of the device under study, for example, a normal state, an abnormal state, a critical state ..... If one of these activation levels is exceeded by the resulting value, a corresponding alarm signal is generated. The alarm can be audible, visible or otherwise, in order to send an appropriate signal to the personnel responsible for checking and supervising the electrical appliance. It can be placed close to the electrical device under investigation or at a distance from it.

Figure 11 illustrates an algorithm for detecting frequencies of interest taking into account the fc frequencies that are known to represent a known type of noise (radio telephone frequencies, television video signals, etc.). These known frequencies are stored in a database for the implementation of the algorithm.

Block 90 in Figure 11 represents an acquisition stage of a series i of RS2i noise spectra. Each RS2i noise spectrum is stored in an acquisition mode called maximum peak, according to which the spectrum amplitude at each frequency is kept constant until a new amplitude value that is more important than the previous one is detected. This stage 90 can be repeated in a cyclic manner over time. After stage 90, at 91 an average amplitude value and a maximum amplitude value are calculated for each frequency in the RS2i noise spectrum series, in order to obtain a reference range of average value and a maximum peak indicated by MRS2 and PRS2 respectively.

Block 92 represents the stage of acquisition of the signal spectrum received by the antenna 4. In this stage, a series j of RS1j spectra is acquired and memorized. This stage 92 can also be repeated cyclically over time. After stage 92, in stage 93 an average amplitude value and a maximum amplitude value are calculated for each frequency in the RS2j spectrum series, in order to obtain an average value spectrum and a maximum peak spectrum of the signal captured by the antenna. 4 indicated by MRS1 and PRS1 respectively.

It should be noted that in stages 90 and 92, the spectra are constructed with the same level of frequency resolution.

After stages 93, repeated processing for the detection of frequencies of interest follows, starting for example with an initial frequency of 200 MHz and ending for example with a frequency of 1500 MHz, with for example an incremental increase of 1 MHz.

In stage 94, it is determined whether the current frequency fi is a known frequency fc and, if so, the procedure continues to stage 95. If not, it moves to stage 96.

• · · · · · &lt;

• · · · · · • · ·· · * · · « • ·· · *14

In stage 95, the amplitude difference is calculated, at frequency fi, between the RS1 spectrum and the RS2 spectrum, for the mean values ​​and for the maximum peak values. This is represented by the relations:

S1 = MRS1 (fi) - MRS2 (fi)

S2 = PRS1 (fi) - PRS2 (fi).

If none of the values ​​S1 and S2 is greater than a limit value indicated by a, as indicated in block 97, the current frequency fi is retained as a frequency of interest, as represented in block 98. In the opposite case, repeated processing is executed for a new fi frequency returning to the stage indicated in block 94.

When in stage 96, that is, after detecting that the current frequency is not a known frequency of known noise, the amplitude distribution of the RS1 spectrum is calculated and the probability of Po distribution of this amplitude in the RS1j spectrum series is checked.

In the event that this probability of distribution follows a uniform or Gaussian distribution, repeated processing is performed on a new current frequency returned to stage 94. More specifically, it starts from the premise that, in general, the nature of electromagnetic noise is more deterministic than probabilistic. The techniques for acquiring the SD signals according to the invention involve the acquisition of UHF signals in time intervals of the order of a few seconds to a few minutes, or even a few hours, or continuously or distributed over time. If the signal captured at a given frequency corresponds to electromagnetic noise, it is possible that there is no correlation between the signals acquired at different times, that is, that these events are independent of each other. Such situations can be regularly detected using a Gaussian distribution. According to basic theory, random or independent events are best modeled by a Gaussian or normal distribution. The other possibility for the characterization of deterministic electromagnetic noise is a rectangular or uniform distribution. In other words, the amplitude of events remains constant over time within a limit ·· ··· ·· · ·· · ·· ·· ·· • · · ··· ♦ · ♦ · · ···· · ♦ • · · · · · ·· · · ••• 13 · · ····· ·· ·· ···· acceptable statistical variation. The signals that result from such events are dependent on the condition of isolation of the electrical device under study.

In stage 94, if the probability of distribution does not follow a normal or Gaussian distribution, successive processing at stages 95, 97 and 98 takes place, with the difference that in stage 97 the limit value α is adjusted as a function of rates confidence in the probability of identifying the distribution.

After stage 98, repeated processing is carried out for a new current frequency of block 94 until a full scan of the frequencies of the 200-1500 MHz frequency band is complete.

·· ··· ·· · "· · ·· ·· ·· ·· · · ··"··· · · ·"·· · · • · · · · Ç · ···· ··· · · ····· ·· ·· ····