WO2024114948A1

WO2024114948A1 - Method and system to classify partial discharge severity

Info

Publication number: WO2024114948A1
Application number: PCT/EP2023/050768
Authority: WO
Inventors: Brian OBERER; Raul Febres AGUIRRE; Padhraig RYAN; James Mark OLSON; Suchandra MANDAL
Original assignee: Eaton Intelligent Power Limited
Priority date: 2022-12-02
Filing date: 2023-01-13
Publication date: 2024-06-06

Abstract

In this disclosure there is provided a partial discharge data classification method and system. The method comprises: acquiring partial discharge data from an electrical device; extracting phase-resolved data of the partial discharge data to obtain phase resolved partial discharge data; analysing the magnitude and intensity values of the phase resolved partial discharge data for noise; and analysing fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve to determine the partial discharge data classification.

Description

METHOD AND SYSTEM TO CLASSIFY PARTIAL DISCHARGE SEVERITY

Field of Disclosure

The field of this disclosure relates to methods and systems for classifying partial discharge severity. In particular, this disclosure relates to methods and systems for classifying partial discharge severity based on phased resolved data.

Background

Insulation breakdown, i.e. damage to insulation, is a leading cause of failure for mediumvoltage equipment. Example forms of medium-voltage equipment are motors, switchgears, cables, or other forms of medium-voltage equipment with ranges from 3.8kV to 38kV.

Although, insulation breakdown is often associated with medium-voltage equipment, equipment operating at other voltage ranges can also experience such failures. Thus, the principles described herein can also apply to equipment operating at other voltage ratings.

Partial discharge can serve as an indicator of insulation breakdown in medium-voltage equipment. Partial discharge is an electrical discharge that does not completely bridge the space between the positive and negative electrodes of an electrical device. Usually this is a spark or arc that discharges in a void within an insulating material located in the space between the electrodes. Thus, it can be a good indicator of insulation breakdown. However, accurately detecting the severity of the partial discharge and the damage to insulation in such equipment can be difficult as the data from the sensor signal can often be masked by noise related to factors such as environmental humidity and temperature.

The relationship of partial discharge with temperature and partial discharge with humidity varies in magnitude and direction across equipment types. Thus, accurately identifying the severity of the partial discharge and cause of the insulation breakdown is difficult and currently requires extensive monitoring by skilled operators and tedious data analysis.

These experts inspect phase resolved partial discharge (PRPD) patterns and partial discharge intensity (PDI) to infer which patterns reflect a significant partial discharge (PD) needing immediate attention, actual PD that is of low intensity and/or noise. Real PD, minor PD and very low severity PD all exhibit sinusoidal behaviour in PRPD data, so further analysis of the PRPD patterns and PDI is required to distinguish between the different PD type. For example, factors such as the repetition of pulses or intensity of the partial discharge trend or the like are inspected and analysed by specialist operators to determine the different PD types. However, assessing the noisy PD data in order to determine the PD type can be complicated as variations exist across different motors, switchgears and other medium voltage equipment.

Due to the extensive analysis required by specialist operators in determining the differences between real PD, minor PD, noise, etc., significant time and effort from specialist operators are needed to identify instances of concern such as insulation breakdown. In addition, there is time and effort required in fixing the insulation of the medium voltage device. As such, it is preferable to detect partial discharge activity as soon as possible to limit the insulation breakdown in medium voltage equipment. Thus, there is a need to accurately identify the various types or classes of partial discharge and the severity of such partial discharge to minimise the accounts of insulation breakdown in electrical devices.

Summary

In this disclosure, there is provided a novel automated method and system for classifying instances in electrical devices of real partial discharge, minor partial discharge, very low partial discharge and noise. Each of these classifications aid in identifying the severity of the partial discharge in the monitored electrical device(s), and indicates which devices require action, attendance or additional monitoring.

In an aspect of the invention there is provided a partial discharge data classification method, the method comprising: acquiring partial discharge data from an electrical device; extracting phase-resolved data of the partial discharge data to obtain phase resolved partial discharge data; analysing the magnitude and intensity values of the phase resolved partial discharge data for noise; and analysing fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve to determine the partial discharge data classification.

The method may further comprise performing pulse repetition analysis on the phase resolved partial discharge data.

The method may further comprise classifying the features of the phase resolved partial discharge data using a rule based classifier. The partial discharge data may be acquired as a time series such that the electrical device being monitored can be monitored over a period of time, with any changes in the partial discharge data more easily detectable.

The partial discharge data may be stored in a database.

The method may further comprise performing a matrix transformation on the partial discharge data. Performing such matrix transformation corrects the phase starting position and allows the fitting of the data to the sinusoidal curve to be simplified.

The matrix transformation may comprise: calculating the correct phase order for the partial discharge matrix; and reconstructing the matrix.

The method may further comprise: selecting the maximum magnitudes per phase of the partial discharge data; selecting all magnitudes per phase of the partial discharge data; or selecting a subset of magnitudes of the partial discharge data.

The method may further comprise checking predetermined peak and predetermined trough values (high - low states) of the sine curve of the phase resolved partial discharge data for noise. This check allows for a quick elimination of easily detectable noise data from the phase resolved partial discharge data, meaning less data is required in fitting the sinusoidal curve.

Checking the high - low states of the phased resolved partial discharge data may comprise: calculating the number of peaks and troughs; comparing the number of peaks and troughs to a predetermined threshold; and determining if the phase resolved partial discharge data is noise or partial discharge.

The method may further comprise checking active regions of the sinusoidal fit. The active regions check provides additional support to the sinusoidal fit findings, i.e. confirms outcomes of the sine fit check, and thus provides a more accurate result.

Checking the active regions of the sinusoidal fit may comprise: finding the maximum magnitude in the first quadrant of phase and the third quadrant of phase of the phase resolved partial discharge data; checking if the maximum magnitude active regions fall within the phases of the fitted sinusoidal curve; and flagging the data if the peaks and maximum magnitudes of the phase resolved partial discharge data follow the sine phase pattern; or flagging the data if the peaks and maximum magnitudes of the phase resolved partial discharge data do not follow the sine phase pattern. The pulse repetition analysis may comprise applying colour significance to the phase resolved partial discharge data for the repeated pulses having the same bandwidth. Applying colour significance allows for ease of observation of the data, if needed by an operator or specialist, and ease of analysis.

Applying colour significance to phase resolved partial discharge data may comprise: encoding pulse repetition ranges in colour; colour coding the pulse repetitions for all magnitudes across a phase; calculating the colour count for all colour coded pulse repetitions; and generating one or more flags of colour significance based on the colour count and the partial discharge intensity for all pulse repetitions of the phase resolved partial discharge data.

Classifying the features of the phase resolved partial discharge data using a rule based classifier may comprise classifying the data into real partial discharge; minor partial discharge; very low partial discharge or noise. Classifying the data into these categories provides information on whether or not action needs to be taken on one or more of the electrical devices being monitored.

Classifying the features of the phase resolved partial discharge data may further comprise determining if the phase resolved partial discharge data may be noise by establishing if: the transformation matrix values are below a pre-defined threshold; the number of cycles are more than the number set to be sampled; and/or the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is below a pre-defined threshold fit.

Classifying the features of the phase resolved partial discharge data may further comprise determining if the phase resolved partial discharge data may be real partial discharge by establishing if: the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is above a pre-defined threshold fit; and the partial discharge intensity is above a pre-defined threshold.

Classifying the features of the phase resolved partial discharge data may further comprise determining if the phase resolved partial discharge data may be minor partial discharge by establishing if: the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is above a pre-defined threshold fit: and one or more flags of colour significance have been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold; or no flags of colour significance has been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold. Classifying the features of the phase resolved partial discharge data may further comprise determining if the phase resolved partial discharge data may be very low partial discharge by establishing if: the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is above a pre-defined threshold fit; and one or more flags of colour significance have been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold; or no flags of colour significance has been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold.

In an aspect of the invention there is provided a partial discharge classification system, the system comprising: one or more electrical devices; an acquisition module; a computational device, wherein the computational device comprises one or more processors which are configured to perform a partial discharge data classification method, the method comprising: acquiring partial discharge data from an electrical device; extracting phase-resolved data of the partial discharge data to obtain phase resolved partial discharge data; analysing the magnitude and intensity values of the phase resolved partial discharge data for noise; and analysing fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve to determine the partial discharge data classification.

Brief Description of Drawings

The disclosure will be described, by way of example only, with reference to the accompanying drawings, in which:

Fig. 1 is a schematic diagram of components of a partial discharge monitoring system; and

Figs. 2A and 2B illustrate a flow diagram of the method steps for classifying the severity of the partial discharge of an electrical device.

Detailed Description

Fig. 1 is a schematic diagram of components of a partial discharge monitoring system 100. The system comprises one or more electrical devices 102 which are connected to different measurement sensors 104 such that the partial discharge can be measured and monitored. Each measurement sensor 104 measures a single electrical device 102. The measurement sensor 104 acquires the voltage data as a time series of each of the electrical devices 102 of the system 100. From this voltage data the partial discharge data can be monitored. Similarly, the measurement sensor 104 may acquire charge data, current data, capacitance data, power data, or the like. As such, the system may also comprise a coupling capacitor, although other sensors may be incorporated such as a resistive temperature device.

The partial discharge system 100, as shown in Fig. 1 , also comprises an application interface 106 and a computing device 108, wherein the application interface 106 aids communication and transmission of data between the measurement sensor 104 and the computing device 108. The computing device 108 comprises one or more processors such that the methods disclosed within this document can be executed by the system 100. Such methods will be discussed in relation to Figs. 2A and 2B. The computing device 108 may be a remote device to allow remote monitoring of the partial discharge of the one or more electrical devices 102 of the system 100. The methods disclosed may also be implemented from a remote or cloud-based server or the like. The computing device 108 comprises a database 110 for storing the data, including the measured data, analysed data, processed data, threshold data, alarm threshold data, etc. Again the database 110 may be a local database of the computing device 102 or stored locally on an ‘edge’ device, for example a microcontroller. The data may also be transmitted to a remote database located on a remote server or on a cloud-based server. The data may be transmitted to the cloud in batch format.

As shown in the schematic diagram of Fig.1 , the system 100 comprises an alarm 112 connected, or in communication, with the computing device 108. The alarm 112 may be a visual or audio alarm, or some other type of alert based system. The alarm 112 may also be a notification or text alert or the like, which is sent to a user when an action is needed, for example when insulation breakdown has been detected. The alarm 112 can also be an indicator of a change in partial discharge, for example when the monitored partial discharge changes to a different classification, as determined by the disclosed method. Again the alarm 112 can be remote to the computing device 108 or any of the other components of the partial discharge monitoring system 100. The alarm 112 can also be located on or at the site of the one or more electrical devices 102. It will be realised that the system 100 may comprise additional electrical devices 102 or other elements which have not been discussed here. The system 100 may also be integrated as part of an overall power and device management system, such as a building management system.

Figs. 2A and 2B illustrate a flow diagram 200 of the method steps for classifying the severity of the partial discharge of an electrical device 102. In this embodiment the partial discharge data of the electrical device 102 is identified and classified into either real partial discharge, minor partial discharge, very low partial discharge or noise. It will be realised that these classifications are not limited and that more or fewer classification categories can be established depending on the application or the types of electrical devices 102 being monitored. Likewise, the thresholds used to define each of the classification categories may also be altered depending on use case scenario.

In this instance the real partial discharge, i.e. real PD, is defined by the partial discharge data following a sine curve pattern closely, while displaying high partial discharge intensity with either high bandwidth of pulse repetitions or low bandwidth of pulse repetitions. Due to the significance of the activity and high partial discharge intensity, the partial discharges can be harmful to the electrical device 102 being monitored. Thus, it may be advised that when real PD is determined additional testing or inspections are performed. Early detection of this type of partial discharge will allow more time for people on site to carry out inspections or repairs.

The minor partial discharge or minor PD is defined by the partial discharge data following a sine curve pattern closely, while displaying medium partial discharge intensity with either high bandwidth of pulse repetitions or low bandwidth of pulse repetitions. The minor PD highlights that there may be a problematic partial discharge issue but unlike the real PD does not require urgent inspection of the monitored electrical device 102. Thus, it may be advised to monitor the PD activity of this electrical device 102 more closely or arrange an inspection in the near future.

The very low partial discharge or very low PD is defined by the partial discharge pattern following a sine curve closely, while displaying low partial discharge intensity with either high bandwidth of pulse repetitions or low bandwidth of pulse repetitions, but more often low bandwidth repetitions of pulses. The very low PD indicates that a small amount of partial discharge is present in the electrical device 102 but there is no urgency to inspect the device so can be inspected at a later date. Thus, it may be advised to continue to observe the partial discharge of the electrical device 102 without increasing the monitoring activity as there is a higher probability that the equipment is in no danger soon.

Noise is defined by the phase resolved partial discharge not following a sine curve or having a sinusoidal nature with high or low partial discharge intensity, and either high or low bandwidth of pulse repetitions. Filtering out the noise data automatically allows the specialists to focus on determining the severity of the actionable partial discharge data (if the partial discharge is not classified by the system) and allow any actions needed on repairs to occur faster. Noise within the acquired data can be difficult to determine when the partial discharge intensity is very high or there is a very high bandwidth of pulse repetitions which trigger the alarm thresholds of intensity. The method, as shown step 202 of Fig. 2A, commences with initialising the monitoring system 100 and acquiring partial discharge data from the one or more monitored electrical devices 102 using the monitoring system 100. The partial discharge data is collected as a time series and includes data on partial discharge intensity, phase start of pulse repetitions, phase resolved partial discharge matrices per channel for every electrical device 102 under inspection. It will be understood that other data may be collected by the measurement sensors 104, such as current, voltage, capacitance, etc. The collected partial discharge data (and other data) is stored in a database 110. As the partial discharge data is acquired by the measurement sensor 104 a temporal dataset is attained and stored in the database 110. The data may also be stored on an edge device, for example a microcontroller, or may be transmitted remotely to the cloud. Data may be stored for some time on an edge device, and then later transmitted to the cloud or another device in batch format. Data may be acquired at regular intervals such as every day or every few days. Other schedules are also possible, for example, 2, 4 or 8 times per day are potential data acquisition schedules that could be deployed depending on the severity of the partial discharge found in the monitored electrical device 102. The acquisition schedule may be altered depending on if the severity of the partial discharge detected has changed for a particular electrical device 102. For example, if the measured data has been identified as going from very low PD to minor PD for one or more electrical devices 102, the schedule may be altered from acquiring once a day to twice a day.

The stored data is then used in a matrix transformation step 204, wherein the stored matrix or matrices are transformed to correct for phase order. For example, the phase resolved partial discharge data may be in a form of 504 values of pulse repetition rates and used to produce a two-dimensional matrix of information. It will be realised that a higher or low pulse repetition rate values, i.e. higher or lower resolution, may be attained to help in better gauging the partial discharge activity. The example of 504 values provides sufficient information such that the partial discharge severity can be classified. More pulse repetition rate values provides a higher resolution of data in the matrix and thus makes classifying the severity of the partial discharge data easier. If too few pulse repetition rate values are attained the data may not be able to be classified or will be very difficult to provide an accurate classification. The engineered two-dimensional matrix may comprise information such as phase information, pulse repetitions and magnitude or intensity of the partial discharge. In order to calculate the correct phase order of the matrix, as in step 204, the pulse repetitions are measured in specific magnitude windows. For example, there may be 24 windows starting from 0 to 6894 and phase from 0 to 360 degrees. In order to determine which phases record which pulse activity, the correct phase starting position is calculated based on the phase shift and phase connection. The calculation is performed separately for each channel that records data on partial discharge activity. The matrix is then reconstructed with the correct phase order and the maximum magnitudes are recorded with pulse activity across every phase.

Steps 206, 208, 210 and 212 can be executed sequentially or in parallel across multiple cores. There is no interdependency between these steps. Step 206 relates to checking for high low states of the phase resolved partial discharge data, step 208 relates to a sine fit check, i.e. checking the fit of the partial discharge data to a sinusoidal curve, step 210 relates to checking the most active regions within the partial discharge data, and step 212 relates to pulse repetition analysis.

As illustrated in the flow diagram 200 of Fig. 2A, step 206 checks for high low states of the phase resolved partial discharge to determine if the acquired data is noise or actual partial discharge. Checking the high low states allows for a quick check to classify the easily detected noise without the need for in depth analysis. Certain forms of noise are easily detectable. For example, the presence of constant weak pulse repetitions with low partial discharge intensity in the phase resolved partial discharge data can be evidence of noise. In contrast, the opposite may occur where higher bands of pulse repetitions along with high partial discharge intensity is present in the phase resolved partial discharge data, which can also be evidence of noise. A more accurate analysis to distinguish noise from an actual partial discharge pattern is the fit of the partial discharge magnitude data to a sinusoidal curve (this will be discussed in relation to step 208). However, step 206 provides a quick determination of the presence of noise in the partial discharge data without the need to perform a detailed sine fit analysis.

A noise pattern can be detected by analytics that capture the number of peaks and troughs that appear when plotting the maximum partial discharge magnitude across phases. This is illustrated in the method flow diagram 200 of Fig. 2A at step 206 by calculating the number of peaks and troughs. If the number of peaks and troughs is more than the number sampled then it may be partial discharge, but if the number of peaks and troughs is the same as the number sampled then the data can be flagged as noise. For example, if one cycle of wave activity of an electrical device 102 is captured or sampled (by one of the measurement sensors 104) then one would expect no more than two peaks and two troughs, i.e. one cycle of a sine wave. The high low state threshold, i.e. the number of peaks and troughs, may be updated based on the number of cycles to be sampled, thus filtering out noise for all wave cycles. For patterns with a higher number of peaks and troughs compared to the number expected for the defined cycle, it implies a high probability that the wave pattern is noise related.

In this high low state analysis 206 of the phase resolved partial discharge data a simple binary categorical flag is generated to quickly distinguish between noise and actual partial discharge data, i.e. without categorising the severity of the partial discharge. As mentioned above, the analysis counts the peaks and troughs and compares them to the threshold number expected based on the number of cycles and flags the data either noise of partial discharge. It will be realised that other flags may be generated based on the high low state (peaks and troughs) analysis. Each of the flags created based on various features can have either binary or continuous values. Flags such as chi-squared statistic, p-values are continuous whereas high low state flags have a discrete number of values they can assume.

The partial discharge detected from the high low state analysis in step 206 can be used in a sine fit check as shown in step 208 of the flow diagram 200 in Fig. 2A. As discussed previously the sine fit check step 208 may be executed in parallel or without the high low state analysis step. However, this would require the sine fit check to be performed with all data, including noise. The sine fit check is best performed when some or all of the noise has been removed as it provides a clearer partial discharge pattern to be assessed against the sine wave and thus is more likely to be genuine partial discharge.

Genuine partial discharge patterns captured at sufficiently high resolutions display a sine pattern. A sine pattern is a sinusoidal wave between 0 to 360 degree phase with peaks at 90 and 270 degrees and close to 0 magnitudes at 0, 180 and 360 degrees. When measuring partial discharge activity from one or more electrical devices 102 it is common to see the sine curve of the measured partial discharge shift from the idealised (expected) sine wave pattern by a number of degrees. To fit the measured partial discharge pattern to the expected sine wave a sine function is used. The sine function comprises initial calculation guesses for frequency, amplitude and offset. These values get updated over time to fit the best sine wave possible to the data of maximum magnitudes. The negative sine wave values are truncated to zero and the goodness of the sine fit is calculated using mean squared error, chi-squared statistic and p-values. It will be realised that other statistical methods may be used for determining the best fit.

Following the sine fit check 208, a check of the most active regions is performed as in step 210 of the method 200. Again, as previously discussed this step may be performed in parallel to steps 206, 208 and 212. The active regions check 210 is a further check performed to support the sine fit findings and determine how closely the sine wave fit follows a contiguous sine wave pattern. For example, does the sine fit follow and increasing magnitude across phase cycles 0-90 degrees and 180-270 degrees with a decrease thereafter, i.e. decreasing across phase cycles 90-180 degrees and 270-360 degrees. The sine fit check step 208 also involves checks on where the peaks in magnitude are detected. Based on these checks a Boolean binary flag is generated, wherein the flag returns “true” if the maximum magnitudes across the phases closely follow a sine pattern correctly shifted or “false” if the phases do not closely follow a sine pattern that has been shifted.

From a combination of the above checks on the measured partial discharge data it is determined that the resulting data is actual partial discharge data. However, more steps are required to determine the severity of such determined partial discharge data. Pulse repetition analysis is performed with colour significance to determine the instances of real partial discharge (real PD), minor partial discharge (minor PD) and very low partial discharge (very low PD) as shown in step 212 of the method steps 200 of Fig. 2B. The discriminant between an instance of real PD, minor PD and very low PD is based on the phase resolved partial discharge pattern matrix bandwidths of pulse repetitions and partial discharge intensity.

Firstly, in the pulse repetition analysis, the bandwidth of pulse repetitions seen across every phase are counted. For example, there may be 24 windows of magnitude measurements for a phase of 60 degrees, and the measured phase resolved partial discharge data shows higher bandwidth pulses repeated across most of the measurement windows of magnitude. The bandwidth range of pulse repetitions are then colour coded for ease of observing and ease of analysis. For example, low bandwidth pulse repetitions of 0 - 0.425 may be colour coded as black, 0.426 - 0.87 as blue and >= 6.26 as yellow. The encoding of the pulse repetition ranges in colour is performed for all magnitudes across a phase. All the colour counts are then calculated except for noise, i.e. the black colour coded counts in this case. Based on the number of high pulse repetition bandwidths observed, a multi categorical flag is generated. The multi categorical flag may have options such as colour significance plus high intensity. For example, a category flag may be when the threshold set (i.e. the threshold for high bandwidth pulses is >2) is exceeded for a count of high repetition pulses and a high partial discharge intensity. Another categorical flag may have a high colour significance but with low intensity, i.e. when the count of high bandwidth pulses is high but the partial discharge intensity is low. When there is little or no colour significance and a lot less partial discharge intensity it may be categorised by another categorical flag. And, finally, a categorical flag may be generated for noise. It will be realised that the value of the ranges of bandwidth pulse repetitions, the number of ranges of bandwidth pulse repetitions and the threshold value for high bandwidth pulses may be altered depending on the type of electrical device 102 being monitored. The method steps 200 from step 202 to 212 output various variable flags and outputs which are either categorical in nature or continuous values of fit metrics such as p-values, chi- squared statistic of fit, sparsity of matrix and mean squared errors, as discussed above. These categorical flags and metrics are then used as features in a rules based classifier to classify the severity of the partial discharge of the monitored electrical device 102. The method steps of the rules based classifier in accordance with the invention is shown in step 214 of Fig. 2B.

The rules based classifier combines various metrics and features to determine whether to mark a phase resolved partial discharge pattern as real, minor, very low PD or just noise. The classifier firstly checks the sparsity of the transformation matrix generated after step 202. If the matrix values are null beyond a defined threshold, i.e. sparse, there it too little data to be used for analysis and the classifier cannot predict the severity of the partial discharge. In this instance, the phase resolved partial discharge data is marked as having sparse data. A typical sparsity threshold used is 2%, and thus matrices with less than 2% data are not analysed and marked as sparse data. For the example disclosed in relation to step 204, the 504 values of pulse repetition rates used to produce the two-dimensional matrix would mean anything less than approximately 10 values would be marked as sparse data. This sparsity threshold may be higher or lower than 2% depending on type of electrical device 102 or application of this partial discharge severity classification method.

If the matrix is not sparse, the next step in the rules based classifier steps 214 is to check if more than the expected number of cycles are determined according to the settings of cycles to be sampled. When more cycles are seen, it indicates there are random pulse repetitions across higher and lower magnitudes, which is more indicative of noise since partial discharge activity clumps in a sine wave pattern. Next the classifier considers the sine fit and if the result of the fit is below a pre-defined threshold fit the phase resolved partial discharge data is classified as noise. If the fit is above a pre-defined threshold fit the classifier uses the continuous metric values of mean squared error, p-value, chi-squared statistic, partial discharge intensity values of the sine fit and categorical flags such as colour significance to classify between instances of real PD, minor PD and very low PD.

As illustrated in the flow diagram 200 at step 214, once a good sine fit is determined, i.e. above a pre-defined threshold fit, the classifier classifies the aforementioned severity states of the phase resolved partial discharge data. The classifier classifies the phase resolved partial discharge data as real PD if the partial discharge intensity is above a pre-defined threshold, regardless if a colour significance has been flagged within the data. If one or more flags of colour significance have been detected in the phase resolved partial discharge data, the partial discharge intensity is below a pre-defined threshold, and the partial discharge intensity is less than a given threshold where the threshold can be decided based on magnitude windows, resolution of phase resolved data, etc., then the phase resolved partial discharge data is classified as minor PD. Or if no flags of colour significance has been detected in the phase resolved partial discharge data, the partial discharge intensity is below a pre-defined threshold where the threshold again is decided based on factors such as magnitude windows, resolution of phase resolved data, etc., then the phase resolved partial discharge data is classified as very low PD.

Since the system generates various informative flags on colour significance, the number of peaks and troughs calculated, mean squared error etc., the method provides not only a classification on partial discharge severity or noise but also provides an explanation on why a pattern was classified into the class it was. Also, since there are no hardcoded thresholds, the thresholds are subject to a particular system and the initial settings of such system, and may be altered depending on the information gathered about said system. For example an electrical device 102, such as a motor, with a frequency of 60 Hz and one cycle (one peak and one trough) of observed phase resolved partial discharge patterns may be used to determine thresholds for initial guesses of cycle frequency of the sine wave, and also for filtering out more peaks and troughs seen in a phase resolved partial discharge pattern than one cycle.

It will be appreciated that the above described embodiments of the present invention are given by way of example only, and that various modifications may be made to the embodiments without departing from the scope of the invention as defined in the appended claims.

Claims

1. A partial discharge data classification method, the method comprising: acquiring partial discharge data from an electrical device; extracting phase-resolved data of the partial discharge data to obtain phase resolved partial discharge data; analysing the magnitude and intensity values of the phase resolved partial discharge data for noise; and analysing fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve to determine the partial discharge data classification.

2. The method of claim 1, further comprising performing pulse repetition analysis on the phase resolved partial discharge data.

3. The method of claim 1 or 2, further comprising classifying the features of the phase resolved partial discharge data using a rule based classifier.

4. The method of any preceding claim, wherein the partial discharge data is acquired as a time series.

5. The method of any preceding claim, wherein the partial discharge data is stored in a database.

6. The method of any preceding claim, further comprising performing a matrix transformation on the partial discharge data.

7. The method of claim 6, wherein the matrix transformation comprises: calculating the correct phase order for the partial discharge matrix; and reconstructing the matrix.

8. The method of claim 7, further comprising: selecting the maximum magnitudes per phase of the partial discharge data; selecting all magnitudes per phase of the partial discharge data; or selecting a subset of magnitudes of the partial discharge data.

9. The method of claim any preceding claim, further comprising checking predetermined peak and predetermined trough values (high - low states) of the sine curve of the phase resolved partial discharge data for noise.

10. The method of claim 9, wherein checking the high - low states of the phase resolved partial discharge data comprises: calculating the number of peaks and troughs; comparing the number of peaks and troughs to a predetermined threshold; and determining if the phase resolved partial discharge data is noise or partial discharge.

11 . The method of any of the preceding claims, further comprising checking active regions of the sinusoidal fit.

12. The method of claim 11 , wherein checking the active regions of the sinusoidal fit comprises: finding the maximum magnitude in the first quadrant of phase and the third quadrant of phase of the phase resolved partial discharge data; checking if the maximum magnitude active regions fall within the phases of the fitted sinusoidal curve; and flagging the data if the peaks and maximum magnitudes of the phase resolved partial discharge data follow the sine phase pattern; or flagging the data if the peaks and maximum magnitudes of the phase resolved partial discharge data do not follow the sine phase pattern.

13. The method of any preceding claim, wherein the pulse repetition analysis comprises applying colour significance to the phase resolved partial discharge data for the repeated pulses having the same bandwidth.

14. The method of claim 13, wherein applying colour significance to phase resolved partial discharge data comprises: encoding pulse repetition bandwidth ranges in colour; colour coding the pulse repetitions for all magnitudes across a phase; calculating the colour count for all colour coded pulse repetitions; and generating one or more flags of colour significance based on the colour count and the partial discharge intensity for all pulse repetitions of the phase resolved partial discharge data.

15. The method of any preceding claim, wherein classifying the features of the phase resolved partial discharge data using a rule based classifier comprises classifying the data into real partial discharge; minor partial discharge; very low partial discharge or noise.

16. The method of claim 15, wherein classifying the features of the phase resolved partial discharge data further comprises determining if the phase resolved partial discharge data is noise by establishing if: the transformation matrix values are below a pre-defined threshold; the number of cycles are more than the number set to be sampled; and/or the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is below a pre-defined threshold fit.

17. The method of claim 15, wherein classifying the features of the phase resolved partial discharge data further comprises determining if the phase resolved partial discharge data is real partial discharge by establishing if: the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is above a pre-defined threshold fit; and the partial discharge intensity is above a pre-defined threshold.

18. The method of claim 15, wherein classifying the features of the phase resolved partial discharge data further comprises determining if the phase resolved partial discharge data is minor partial discharge by establishing if: the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is above a pre-defined threshold fit; and one or more flags of colour significance have been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold no flags of colour significance has been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold

19. The method of claim 15, wherein classifying the features of the phase resolved partial discharge data further comprises determining if the phase resolved partial discharge data is very low partial discharge by establishing if: the fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve is above a pre-defined threshold fit; and one or more flags of colour significance have been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold; or no flags of colour significance has been detected in the phase resolved partial discharge data and the partial discharge intensity is below a pre-defined threshold.

20. A partial discharge data classification system, the system comprising: one or more electrical devices; an acquisition module; a computational device, wherein the computational device comprises one or more processors which are configured to perform a partial discharge data classification method, the method comprising: acquiring partial discharge data from an electrical device; extracting phase-resolved data of the partial discharge data to obtain phase resolved partial discharge data; analysing the magnitude and intensity values of the phase resolved partial discharge data for noise; and analysing fit of phase distribution of the phase resolved partial discharge data to a sinusoidal curve to determine the partial discharge data classification