Tektronix 3 Series MDO - Part 3: Partial Discharge Analysis

• • Introduction
  • Capturing PD Pulses from TGA-B
  • Modification of Simulated Pulses
  • Tektronix Arb Software
  • Generator Partial Discharge Measurements
  • 132kV Cable Discharge Measurements
  • Switchgear Partial Discharge Test
  • Conclusions
  • Addendum

Introduction

I plan to utilise the 3 Series MDO to collect partial discharge data from a generator in operation. Opportunity arose to also take readings on some 132 kV cabling from a transformer to a Grid Connection and then on a high potential test carried out on a 12 kV breaker. First though I will take a look at the TGA-B analyser used to collect the data.

Capturing PD Pulses from TGA-B

The TGA-B from Iris is a professional piece of test apparatus designed to capture partial discharge activity within high voltage motor and generator stator windings. Generally it connects to the high voltage circuit via capacitive couplers. It is a six-channel device with three channels for the machine and another three channels for the system, a set of algorithms based on time of flight between the machine and system channels then extracts the partial discharge activity and displays it as a number of plots and comparative measurements values.

This device has a test output that can be utilised to check the operation of the data collection channels on the TGA-B. I hooked up these test outputs to the 3 Series MDO to take a look at the signals it produces. Three sets of signals are looked at, the 50Hz reference signal, the partial discharge pulse stream for the machine and then for the system inputs.

The screen capture looks like this. Note the scope measurements have reported 'clipping possible' indicating that the scope thinks there is a signal present that is not displayed on the screen. This is likely to be the partial discharge pulses it simulates that will have a higher amplitude and faster frequency than the reference signal.

At present I cannot find any digital filters that can be applied to the inputs, so in the absence of an external filter, I set the bandwidth limit to 20MHz, which should be under the partial discharge signal frequency. Now I am just left with the 50Hz reference signal and the two measurements no longer show the alarms.

It is a pretty messy reference signal, lets see what an FFT of it looks like. It is definitely not the best looking sine wave, but it is only a reference signal for the apparatus, so I presume they didn't spend too much time creating it.

I will increase the bandwidth again so that we can take a closer look at the simulated partial discharge pulses. In the first screenshot, if you look closely, you can just see the faint spikes of the simulated partial discharge activity occurring every half cycle of the reference waveform.

Reducing the horizontal scale to display as near as possible only one cycle, shows that there are three pulse trains for each half cycle. This is exactly the same for the system pulse train except that there is no 50Hz reference waveform present.

Using the zoom function of the oscilloscope, it can be seen that the three pulses are further broken down into a set of four individual pulses within them, the timing between each burst of pulses was measured as around 570 us using the cursor function. The timing between each PD pulse was measured as 35.5 us by zooming further in. The final screenshot above shows what each of those four pulses look like.

Each simulated pulse has a positive width of 68.4 ns with a rise time of 4.466 ns. To record this, the timebase is down to 20 ns/div, so we haven't quite reached the 400 ps/div limit of the oscilloscope. The width and rise time measurements were made with the automatic measuring function of the oscilloscope.

The other measurement of interest is the time delay between the pulse on the machine channel to the pulse on the system channel. We can add in channel 2 to display the pulse on the system channel. Apart from the slight voltage offset of the system pulse , it looks almost identical to the machine pulse.

The delay between the two pulses can also be measured with one of the functions and comes in as 11.63 ns. Haven't added in the pulse width and rise time measurements for the system pulse on channel 2, as this would take one over the four measurement limit of the oscilloscope.

Just for good measure, I will turn the horizontal scale down to 4.00 ns/div to get a more accurate measurement of the delay.

This now comes in as 11.33 ns, so not a great deal of difference. I am curious about the measurement though, as the position of the automatic measurement cursors, seems to be about midway up the rising slope of the signals.

Modification of Simulated Pulses

The 3 Series MDO contains an arbitrary wave form generator that I had planned to utilise to play back modified partial discharge pulse patterns back into the TGA-B analyser. The waveform generator has 128 k memory depth and the ideal record length will be 20 ms, to cover a single 50 Hz cycle. This is because the individual pulses are set to occur at specific times across the 50 Hz waveform, so to get them in the correct place one full cycle over 20 ms seems to be the best approach.

Looking at the functionality of the arbitrary waveform generator, it will only load either 'isf' or 'csv' files. The 'isf' format is Tektronix's own file format for saving files on the oscilloscope and is compatible with their devices and their TekScope Anywhere software package. It doesn't give the opportunity to modify the file. This leaves me with the 'csv' format, which of course can be loaded into Excel and numerous other spreadsheet packages.

Signal manipulation within Excel proved to be very awkward. Due to the speed of the signal and the depth of memory utilised to capture the traces, the Excel files were quite sizeable. Excel will load a maximum of 1,048,576 rows. A file saved on the 3 Series MDO with a memory depth of 5 Mpts was therefore truncated and not of any use, so a memory depth of 1 Mpts was the maximum that could be used.

With the file loaded into Excel, it is a case of manually finding the pulses of interest along the waveform and carrying out the required modification. Amplitude modification is relatively simple, and I altered the pulses to be the same across the waveform. Once the pulse locations were found, hyperlinks were added to the worksheet to allow them to be easily found again.

The other modification made is to the pulse rate and location. Pulse rate was modified by adding extra pulses in between the existing pulses and the location was changed by copying pulses to a different part of the reference waveform.

Whilst this is a 'csv' file, it should be noted that the file requires the header data for the 3 Series MDO to interpret the data in the file. If this is missed out, then the error message seen in the last photo above is displayed. This means it is more awkward to take a 'csv' file from another source an load it starlight into the 3 Series MDO.

The following screenshot compares the original Excel file played by the waveform generator back into the oscilloscope in yellow, against the captured trace from the TGA-B beneath it. The 50Hz reference waveform has come out a little cleaner. But looking closely at the PD pulses along the waveform, it can be seen that both definition and amplitude have been lost in the waveform from the waveform generator. This is likely to be due to the limited memory depth of the waveform generator, in comparison to the capture capabilities of the oscilloscope.

Looking more closely at the actual PD Pulse from the AWG waveform shows that it has lost its shape and is now much longer. This may cause problems when played back in to the TGA-B analyser as the algorithm may not recognise the extended pulse as simulated partial discharge activity.

The setup to play the signals back into the TGA-B is very similar to the collection of the reference signal from the TGA-B. The TGA-B requires a 50 Hz reference signal on the machine input or it will not collect data, therefore when applying the test pulses to the system inputs, I utilised the internal reference signal from the TGA-B on the machine inputs as the AWG only has one output channel. Unfortunately, as I seem to be utilising the AWG output a lot, its placement recessed into the back of the unit has started to become a little frustrating. I guess if the instrument was desk bound you could permanently leave cables connected to the rear connections so it wouldn't be much of an issue.

To put the test pulses I have made into context, below is a capture of the built in pulses of the TGA-B. This type of plot is known as Phase Resolved Partial Discharge and is one of the ways of displaying partial discharge activity. The graph is displaying the location of the activity in relation to the 50 Hz reference in relation to the magnitude of the pulse in mV. The colour of the pulse defines the number of pulses per second at that location and mV level.

This is the standard plot that I have always used to verify correct operation of the TGA-B, but it is only one plot, so I aim to utilise the combination of the oscilloscope and AWG functions of the 3 Series MDO to produce some other plots and see how the TGA-B responds.

This first plot is the original signal from the TGA-B, captured on the scope and then played back into the TGA-B from the waveform generator. When this was analysed on the oscilloscope, the PD pulses were seen to have lost amplitude and increased in duration, so I expected the plot to be different from the original signal.

The reduce amplitude has required the mV range to be reduced from 850 mV down to 170 mV, but the pulses are now spread over a wider area due to the increased pulse width within the waveform. It looks like the software has translated this increased pulse width into more pulses across the reference waveform as opposed to wider individual pulses. Lets see what happens when the amplitude of the pulses is increased to 800 mV peak to peak. The screenshot below shows what the scope captures.

The Phase Resolved Plot below shows the increased amplitude of the pulses captured by the software. The pulse location has remained consistent with the previous plot at the lower amplitude.

I reconfigured both of the test files to have a higher number of negative pulses, but left the number of positive pulses the same. This is what the scope captured at the higher pulse amplitude. You have to look closely, but a higher number of negative pulses can just be made out.

At the higher mV level, the negative cluster of pulses appears to be more dense, but the software reports the same Qm values as before.

At the lower amplitude, the negative pulse activity is again more dense, but is showing a lower mV level. This did happen on all three phases, so appears to be consistent, but is not what was expected. The expectation is that the Qm- value would have increased. This will require further investigation and testing.

I then tried this same signal on the system input to the TGA-B. Although the plot obtained was relatively consistent across all three phases, it was completely different to the plot obtained from the same signal applied to the machine input. This is again another area that will need further investigation.

One final test was to create extra pulses across the waveform to see if the TGA-B would capture them. The pulses were the same magnitude and width as the initial pulses. The oscilloscope screenshot shows the extra pulses across the waveform.

As can be seen below, the TGA-B recorded the extra pulses across the waveform which was the expected outcome.

This aspect of the testing has worked out relatively well. Some of the results were not as expected and will need further investigation, which will require the creation of different test waveforms, so will move towards a long-term goal.

Tektronix Arb Software

Tektronix offer two software packages for their arbitrary waveform generators and both are free to download once you are registered on their website. SourceXpress is the first package I found and is for use with their AWG5200 and AWG70000 units. It will load the 'isf' files created by the 3 Series MDO, but will only save files in its native format that the 3 Series will not read.

Waveform manipulation within the SourceXpress package seemed to be via a datable of the points in a similar manner to the Excel file manipulation. I must confess though that I did not investigate this much as soon as I learned that I could not output the waveform in a format suitable for the 3 Series MDO.

The other package available from Tektronix is the ArbXpress software made for their range of arbitrary waveform and function generators. The screen capture below identifies which models it is compatible with and it can be seen that the 3 Series MDO does not appear to be amongst them, despite some of the MDO oscilloscopes having compatibility.

The software however, will load both 'isf' and 'csv' file formats and editing can be carried out in the waveform window by either copying and pasting existing waveform elements, 'joystick' style manipulation of waveform points or manually drawing a waveform.

File load options is well supported with 'isf' 'wfm' and 'csv' compatible with the Tektronix Oscilloscopes. It will only save to 'wfm' or 'csv' formats for import to an oscilloscope. The other formats I believe are for their waveform generators. the 3 Series does not support 'wfm', so I still have to stick with the 'csv' format.

Unfortunately, the 'csv' format produced by the ArbExpress software is not the same format as the 'csv' file created by the 3 Series MDO and they will not load directly into it. The 'csv' file from ArbExpress places the data column first and then seems to use clock data for formatting the waveform. The easiest way I found to overcome this was to edit the waveform and save as 'csv' and then copy the data column from this 'csv' file into an existing 'csv' file from the 3 Series MDO.

The oscilloscope will then read this file and load the waveform and as the last picture shows in the gallery, I can zoom into the PD pulse simulated in the ArbExpress software. It is a bit messy and time consuming, but ultimately I can achieve pulse modification and play it back into the TGA-B and observe the effects. For me it would have been nicer if Tektronix had given a little more thought into the software integration of the function generator within the 3 Series MOD and its compatibility with their other hardware and existing software.

Generator Partial Discharge Measurements

The next test was to take the 3 Series MDO on a trip to a power station to capture partial discharge data from a generator alongside the TGA-B analyser. This particular generator operates at a line voltage of 21kV, just over 12kV to earth. Connection is made to the busbars via 80 pF capacitive couplers as in the graphic from the Iris literature.

With this system, I can simply replace the TGA-B analyser with the oscilloscope and use the zoom and horizontal timebase to capture data of specific interest. The image below shows the scope connected to the coupler termination box at the top of the picture. To maximise the screen, only one phase was connected at a time.

The initial screenshot shows quite clearly the six pulses across each cycle that are created by the rectification of the supply to create the DC excitation voltage for the generator. This actually shows up better on the oscilloscope than it does on the TGA-B. The third pulse in each half cycle appears to be much smaller, this was again detail not picked up on the TGA-B analyser.

Away from site, the captured waveform can easily be loaded back into the oscilloscope and zoomed in to take a closer look at the pulses along the waveform.

This type of extended pulse with a long oscillating tail is more typical of noise than genuine partial discharge activity. The zoom facility of the scope allows individual pulses to be looked at more closely. The cursors are relatively easy to use and when they disappear off screen due to zooming can easily be recovered by bringing up the cursor menu or right clicking the mouse on a blank part of the screen to bring up a context menu.

The cursor menu has a function to bring both cursors onto screen and the 'right' click menu has the option to bring either cursor to that point on the screen.

This is a relatively friendly environment for the 3 Series MDO to be transported to. The transport bag works well and offers good protection whilst being transported in the car and carried around site. Whilst the scope can easily collect the data, I did find it easier to carry out analysis back in an office environment as it did become time consuming to manually look at the different pulses across the waveform.

132kV Cable Discharge Measurements

An extra job was carried out on this site in the form of partial discharge measurement on the outgoing cables of the step-up transformer to the 132kV switchyard. For this analysis a Radio Frequency Current Transformer (RFCT) was utilised, clamped around the earth connection to the cable sheath. This measurement has been carried out with an UltraTEV analyser using the same RFCT, but this is a relatively basic analyser that is very susceptible to noise. Monitoring the 132kV cables with this device, gives a high level reading, that puts the cable into an alarm category.

I was keen to utilise the 3 Series MDO to carry out some further analysis on the partial discharge activity on these cables. The same RFCT utilised with the UltraTEV was used with the oscilloscope.

The first picture shows the RFCT clamped around the earth cable to the sheath. The RFCT looks just like a split core current transformer, and it is, only having a much higher bandwidth than a standard power current transformer.

The second picture shows the overall set-up. Some would regard this as quite a harsh environment for this type of instrument, but it seemed to cope with it quite well. The ability for the handle to fold down and widen the base of the oscilloscope certainly helped with the stability on the pebbles covering the bund floor grating.

The third picture shows the first capture of data from the RFCT, pretty messy, but you can just make out a 50 Hz sine wave together with a serious amount of activity. The final picture shows the results from the UltraTEV locator using the RFCT. The reading was found to be around 7000 pico-Coulombs (pC) but had reached up to 9117 pC, this would be considered as a worrying level of activity if it is genuine partial discharge.

Not much analysis can be done out on site, but the initial impression from the screenshot on the scope is that this is just noise, there may be some partial discharge activity hidden away within all these spikes, but it will be negligible and extremely difficult to find without specialise software for the analysis.

With the waveforms captured on the oscilloscope in its native 'isf' waveform format, I could return to the office to take a closer look at the data. The 'isf' file can either be loaded back into one of the reference channels of the oscilloscope or canoe viewed on a computer using the TekScope Anywhere software. On this occasion, I stuck to carrying out the analysis using just the oscilloscope.

Called back into the reference channel, this is what one of the captures looked like. I am not sure why the grey bar has appeared on the screen. It seems to occur occasionally when the zoom function is utilised, although it appears to be part of the screen save message, as saving the screenshot gets rid of it.

There are a number of sharp spikes standing out across the waveform, that I would like to take a quick look at, although their regularity suggests they are also noise.

The first three screenshots are capturing individual pulses, to look at their shape and duration. Partial discharge pulses are characterised by their fast rise time, usually less than 6 ns and their short duration, less than 100 ns, this does vary based upon the type of detection and the medium the discharge pulse is travelling through. The scope clearly shoes that all of these three pulses are longer than expected for a discharge pulse, their general shape is also more representative of noise than a partial discharge pulse.

The last two screenshots are more generalised to show the level of noise across the waveform.

Throughout the process, the zoom function performed well and the touch screen allowed for easy manipulation of the waveform. Personally, I found that having a mouse plugged in to one of the USB ports aided the operation. I did find the '+' and '-' buttons on the zoom control were quite small and regularly missed them using finger control, but was easily achieved when utilising a mouse.

Switchgear Partial Discharge Test

Whilst working out on one site, the opportunity arose to carry out a high voltage test on a 12 kV breaker to verify the integrity of the vacuum within the contact bottles. This is carried out by monitoring the leakage current through the open contacts with 20 kV applied across them. As it is an over voltage test, this generally also creates partial discharge activity across the bottle surface. The test set-up can be seen below.

To carry out some monitoring I ran the return lead from the breaker to the test set through an RFCT and monitored the signal with the 3 Series MDO. For a comparison I clipped a Transient Earth Voltage (TEV) sensor onto the switchgear and monitored this with the UltraTEV. The background activity, with the test voltage removed, can be seen on the oscilloscope screen above and also on the UltraTEV below.

With the test set energised and 20kV applied, the reading on the UltraTEV increases significantly, from 11 dB up to 56 dB.

The increase in activity can also be seen on the oscilloscope in th following screen capture.

Instead of using the oscilloscope for signal analysis, I utilised the TekScope Anywhere software available from Tektronix. This is a licensed software package, but a 30 day trial can be downloaded for evaluation purposes.

I have to confess that I found it much easier and quicker to manipulate the waveform in the software rather than on the oscilloscope. It was a reasonably simple matter to determine which pulses were genuine partial discharge activity from those that are much more likely to be noise.

Conclusions

This aspect of the roadtest was expected to be more challenging. Whilst the oscilloscope aspect of the 3 Series MDO can capture partial discharge data, trying to utilise the waveform generator aspect to modify and playback test waveforms was not as easy to achieve. It can be done, but is not as easy as it could be if Tektronix had provided support for the 'wfm' file format within the oscilloscope or adopted the same 'csv' file format as their ArbExpress software.

Partial discharge simulation pulses were created and successfully played back into the TGA-B from the waveform generator built into the oscilloscope. The 128 kPTS limit of the waveform generator did ultimately affect the results of the tests. I suspect that I would be better off in utilising the 3 Series to collect the data and then look at their AFG31000 series function generator to create the waveforms. This would be interesting as they have a dual channel option that could be utilised with the TGA-B to test both machine and system channels at the same time.

Out on site the 3 Series MDO performed perfectly well. The option to use the handle as part of the stand for the oscilloscope gives it a solid platform even when working on the cobble stones of a transformer pen. It travels without any issues within its protective soft-case and was easily carried around site.

Data can easily be collected and stored for analysis at a later date. The size of the files created due to the memory depth utilised meant that data could only be saved to an external USB and not to internal memory, but this is no different to any other oscilloscope I have. Whilst the scope can be used for analysis, I found it easier to control with a mouse and this isn't always easy or possible out on site. In an office environment, the mouse can easily be used with the scope, but this also opens up the option to use the TekScope Anywhere software that complements the 3 Series very well. It was found to work in a similar manner to the oscilloscope and provided extra analysis options that aren't available on the oscilloscope.

Addendum

Up to this point, I have deliberately left out descriptive information on partial discharge in high voltage insulation systems, as it is a very complex subject and I didn't want it to detract from the use of the 3 Series MDO that this blog is really about.

It is important for insulation engineers to monitor the partial discharge as it effectively eats into the insulation around a high voltage component and thins out the insulation until the electrical stress over the remaining insulation is so great that it is punctured and an electrical flash over occurs. This is usually highly disruptive and can cause apparatus to be out of service for months whilst repairs are made, not to mention the risk to personnel.

Here is a picture of some partial discharge on a generator end winding basket.

You can see the areas marked in-between a stator coil and the support ties. It is also another nuisance on this kind of environment, in that as it eats into the insulation it causes a larger gap, that in turn increases the amount of partial discharge. It can also cause the tie to become loose, this will then add mechanical vibration to the failure mechanism that will increase the speed at which the insulation is rubbed away. In this picture the partial discharge has been caught early in its life and can therefore be repaired.

This picture from Doble Engineering, shows what happens if the partial discharge goes on for too long and causes a puncture in the insulation. Depending on how much copper damage there is, there are repair techniques to rebuild the insulation, or it may be that the whole bar has to be removed and a new one installed, which is likely to have the machine down for one to three months.

This is just a brief look at partial discharge and the issues it causes, to give the reader an insight into why the testing above is important to high voltage insulation engineers.