O&M, Plant Optimization

Continuous, On-Line PD Monitoring for Generators

Issue 10 and Volume 121.

By Samuel Clemmons, James Hovious, Marco Tozzi and Enrico Savorelli

Breakdowns of the electrical insulation system have been documented to cause catastrophic failure of critical equipment. Partial Discharges (PD) are measured on rotating machines in order to prevent early failures occurring in stator insulation. However, only a few electrical generation companies have adopted permanent PD continuous monitoring solutions, while most use only periodic tests.

The Tennessee Valley Authority (TVA) has engaged periodic on-line PD tests for several years with mixed results regarding the agency confidence and understanding of PD system results. Sporadic acquisition of PD data is one of the barriers to widespread deployment of permanent monitoring systems, and the clear understanding of results is another drawback. In particular, TVA recognized the need for an easy to interpret ‘picture’ that can intuitively show PD results to power-plant operators and asset managers, without the need of an expert in PD theory to review and to provide data interpretation.

This article details real case studies using a new technology, designed specifically for generators, which shows continuous PD monitoring data in an intuitive and informative format. This information will help the plant to plan corrective actions, improve operating conditions, and defer the need for experts in data interpretation. These results are achieved by continuously monitoring PD activity and automatically correlating recorded PD with operating condition parameters such as active power and temperatures. Periodical comparison of data recorded at the same operating conditions provides reliable trending to trigger alarms. Visual maps provide immediate understanding of PD behaviour with respect to copper temperature and load. This comparison enables the machine owner to manage machine loading and cooling to extend insulation reliability. Additional diagrams provide a cumulative summary of PD activity at machine start, stop, low, medium and high load, all of which is of particular interest when generators operate at peak demand, cycling, or at system automatic loads.

Several stator failure mechanisms have reported a close correlation with PD activity. Some of the most common defects generating PD are described as follows:

  • Thermal deterioration: chemical ageing process increasing internal gas pressure and decreasing the adhesive strength of the epoxy-mica interface, resulting in voids, delaminations and PD;
  • Thermal cycling: thermomechanical stress due to different thermal expansion coefficients of the materials involved weakening and breaking the bond between the copper and insulation, generating delaminations within the insulation and PD;
  • Poor resin impregnation: leaving distributed air bubbles within insulation and generating PD;
  • Loose stator bars: due to the vibrations, the bar moves in the slot, damaging and abrading the slot conductive coating and generating PD (Slot PD);
  • Semicon coating: generating PD within the space between the stator and the coil (Slot PD) due to too excessive initial resistance or poor application of the conductive coating;
  • Semicon/Stress-grading junction: poor connection between the stress grading tape and the conductive coating, generating PD between the junction between the two materials;
  • Inadequate end-winding spacing: causing insufficient clearance between the bars and generating PD between the bars in the end-winding;
  • Contamination: causing surface tracking in th e end winding.

It is well known that the analysis of PD trends over the years establishes an effective method to assess the insulation degradation rate, since an increase of the defect size (volume of void, gap between coil and stator, surface tracking path, etc.) generally leads to an increase of PD intensity in terms of amplitude and/or number of discharges. Potentially, the trend analysis can be obtained by comparing periodic measurements (off-line and on-line) or using a continuous 24/7 monitoring system.

Off-line PD measurements (machine not running, external voltage source applied phase by phase) are generally inadequate to detect loose-bars defects (load is absent during the test), to confirm deterioration of the voltage stress coatings (temperature and humidity cannot be varied during the test) and to investigate discharges between phases.

On-line measurements represent the most effective method to detect all possible PD sources since the machine is tested at real operating conditions. However, periodic on-line tests have limitations since it is difficult to replicate the PD measurement at the same identical operating and environmental conditions, resulting in an uncertain and unreliable comparison. As a matter of fact, PD intensity can significantly vary hour after hour in normal conditions due to the load and temperature changes. For example, high discharges can occur at generator start-up, due to open gaps in the ground-wall insulation, and disappear in few hours due to the coil temperature increase, which causes the copper to expand and fill the gaps [5-6]. In other cases, high load or high temperature can result in a sudden increase of slot discharges which would disappear at cold temperature or low load. For this reason, PD trend evaluations must take into account environmental and operating conditions. The key point is to compare PD intensity at nearly the same voltage/load/temperature/humidity conditions.

The use of permanent monitoring systems, recording diagnostic (PD), operational (voltage, active power, reactive power, copper temperature, cooling temperatures, H2 pressure) and environmental (temperature, humidity) parameters represent the only way to carry out a meaningful correlation over the time at the same identical operating condition. The collected data shall be easily grouped basing on the chosen parameters to show a summary of the insulation status at cold or warm machine, at low or high load, etc. The importance of making the recorded data meaningful and easy-to-interpret is fundamental to allowing decision makers to compare the status of the generators, manage operational stresses in time, and devise a real condition-based maintenance plan.

The following paper describes a trial project, carried out by TVA and Camlin Power, which equipped four generators with a continuous PD monitoring system. The results from three of the four generators are reported due to two of the generators showing no PD. These results highlight the benefits of the applied technology within the TVA monitoring and maintenance program.


TVA Kingston Fossil Plant (KIF) has 9 generators, all of which were installed from 1954-1956. Four of them are rated 175 MW while the others are 200 MW. All machines are Hydrogen cooled and have VPI single-bar insulation, except one stator that is Resin Rich. A full rotor-out major outage is carried out every 10 to 12 calendar years. Wedge and endwinding tightness inspections, as well as routine electrical tests (winding resistance and 2500 volt megger/10 minute polarization index) are included in the outage scope. Minor outages are spaced between the major inspections and include routine electrical tests. Over the past decade, TVA Kingston has invested in partial discharge analysis on the generators. Most of the nine generators have bus couplers installed, and the units were tested annually. However, the data obtained from these tests did not prove to be useful. While the plant was supplied with Qm+ and Qm- levels from a testing vender, there was no indication of what was acceptable or unacceptable. In addition, the data seemed erratic. There would be a higher magnitude PD level one year, followed be a lower level the next. A PD expert had to be contacted to translate the data to be beneficial for onsite employees. Still, the information obtained did not contain operational recommendations that could be used to lower PD levels. As a result, partial discharge levels in Kingston generators were never fully valued and were only kept for trending purposes.


TVA has decided to engage a trial evaluating a permanent, continuous PD monitoring solution. The permanent monitoring system consists of a set of three capacitive couplers and an acquisition unit. Existing couplers already installed in the generator can be used, thus there is no need to replace the hardware installed in the past. For the TVA KIF trial, existing sets of 80 pF couplers were used. The acquisition system module includes an acquisition board, a module for external inputs, an embedded PC with integrated server, and modem. This module connects to the machine side bus couplers and records continuously. A de-noising logic algorithm uses the simultaneous signal acquisition from the three phases and automatically rejects what is considered noise. The resulted data is saved in an embedded database. A summary of the recorded activity is performed every 10 minutes. The output data includes the well-known parameters such as Qmax (Volts and pC), Repetition Rate (pulse-per-seconds), Qm+, Qm-, NqN+ and NqN-. Additionally, Qmax and Repetition Rate are combined into a non-dimensional parameter called PD Energy (PDE), which is evaluated for each phase of the machine. Operational and environmental parameters (megawatts, megavars, stator temperature, etc. are recorded simultaneously and continuously with the PD data. A total report is then produced with the correlated data (PD and machine parameters).

Using the report data, a 3-D PDE map is created. A dot is plotted every N minutes (N = data sample rate) with X, Y coordinates representing the operating conditions. The user can choose the parameter to be used in both axes depending on his needs, experience and parameter availability. Each dot is coded by color using a scale from green (low PDE) to red (high PDE), depending on the PDE level recorded.

As an example, Fig. 1 summarizes 4 months of data from a 300 MW turbo-generator with a sample rate of 10 minutes. The X and Y axis are active power and copper temperature respectively. The meaning of the shape of the cluster is straightforward: each PDE dot corresponds to a certain operating condition with a certain active power and copper temperature. As a result, four areas can be easily segmented in the plot:

  • SS: Start-Stop region, characterized by low power (below minimum-technical)
  • LT: low-temperature region, the machine operates at significantly low temperature, generally after the start
  • MT: mid-temperature region, it should represent the “normal” operating condition
  • HT: high temperature region, the machine operates at temperatures not far from the maximum from design

The thresholds for defining the four areas are configurable and is customizable to each machine.

As an example of interpretation, the PDE Map in Fig. 1 shows that the monitored machine is load-base (very few dots in the SS and LT area) with very low PDE activity at the start and low temperature conditions (mostly green dots). The machine is mainly operating at medium and high temperatures. The PDE is low in the MT region (still green predominance). However, the HT region, especially above 85°C, correlates to a higher PDE (red predominance).

In this particular example case, the utility previously used periodic PD testing (every 6 months). The owner was not aware of the correlation between PD and temperature. With the above results provided, the owner took action to increase cooling at higher loads. Thus, they were able to bring the machine to operate mainly in the MT area (even at high loads). The result generated a predominately green PDE map which depicts slower degradation.

In terms of trending PD, the monitoring system has the capability to extract a weekly summary indicating the PDE level at each of the designated operating temperatures. Figure 2 shows an example of the same machine above, monitored for a certain week where a few start-ups/shutdowns were also made. The weekly summary is in agreement with the PDE Map. It confirms that PDE increases with temperature and it is maximum at high temperature (HT), i.e. above 90 °C. The summary also shows PDE for each phase, highlighting the predominance of PD activities in C phase (red).

The values shown in Fig. 2 are then generated weekly. These reports create meaningful trends at similar operating conditions. Three different alarms can be sent to the owners/operators – one for each operating range (LT, MT, and HT). In addition to the three temperature ranges, the correlated PDE and operational data can separate start-up and shutdown periods. These operational conditions provide the potential for two additional alarms (start and stop). This feature is important for cycling units, which sometimes show higher PD levels during these periods due to voids and delaminations. Typically, these time periods go unmonitored when using the conventional periodic PD testing techniques.


The trial has taken place at the TVA Kingston Fossil Plant. Four generators where monitored for several months each. Table 1 shows the characteristics of the generators.

Figure 3 below shows PDE maps collected from Units 6, 7, and 8. Using the same PDE color scale (max PDE=90) and the same operational parameters on the X and Y axis, the maps are comparable between all three sister units. The comparison reveals that all three insulation systems are in different conditions. This fact alone stresses the importance of knowing which factors (temperature, loading, voltage, etc.) of PD are contributing to the resultant PDE levels. From these initial maps, each unit was analysed deeper.

Unit 6

The PDE map shows that this unit is mainly affected by PD at low temperatures. When temperatures exceed 60 °C, PDE levels are significantly attenuated. Note that when the machine operates at maximum power and temperature, PDs are not active (predominance of green dots). This could be due to presence of voids in the insulation when the copper is not fully expanded, producing PD when gaps are still present. As soon as stator temperature increases, copper expands and squeezes the voids reducing the PD activity. The analysis is confirmed by the weekly data aggregation for trending purposes shown in Fig. 4 (LT=55°C, HT=60 °C) which emphasises the PD decrease with temperature increase. This is an opposite behaviour with respect to the machine shown in the example in Fig. 1. – again stressing the importance of the correlation between PD and temperature.

After having noted this relationship, TVA took an action by raising hydrogen temperature from 90°F to 95°F which, in turn, raised the stator temperature of about 5 to 10 °C depending on the power. The mitigation effect on the PD activity was observed immediately. Figure 5 shows the PDE Map of about 3 weeks of data before and after the new “temperature setup”.

As highlighted in the picture above, there is a clear benefit of the temperature change when looking at the PDE at around 120 MW. Before the change, the temperature was ranging between 52 and 58°C with the color of the PDE dots mainly red. By increasing the temperature in the stator to 60-67°C at the same load, the PD activity is significantly reduced (predominately green dots). To confirm the results, Phase Resolved Partial Discharge (PRPD) patterns (for Phase A) are provided in Figure 6. Note that PD activity is considerably reduced; in turn, the degradation process will be lessened. At the same time, an increase from 58 to 67°C in the stator does not lead to any detrimental consequence since the machine is still operated at a relatively low temperature with respect to its insulation class.

Unit 7

Unit 7 shows red dots more or less in every operating condition, thus with no evident separation between low or high load and temperature. In this case the PDE Map is further analysed for each phase (depicted below in Figure 7). This figure allows the user to easily identify that there is just one phase particularly affected – phase B.

The Phase Resolved Partial Discharge (PRPD) pattern was analysed in this case, for phase B. The pattern, shown in Fig. 8, shows a predominant stress-grading activity [11] which suggests a visual inspection of the end-windings, in particular the corona protection tapes. A preliminary boroscope and bushing box inspection has been scheduled to identify potential issues in this area.

Unit 8

Unit 8 shows small PD activity. All of the PDE dots are green in each operating condition. This information allows TVA to defer any planned projects or inspections relative to the insulation condition of this machine. Also, this PDE map can serve as a baseline for weekly alarms. The recorded operational conditions will assist in troubleshooting and analysis if a step change in PDE should occur.


The results on Unit 6 have demonstrated that mitigation actions can be taken if meaningful information is provided to decision makers. Delivering periodic reports from on-line or off-line measurements, indicating just PD magnitude with no correlation to load or temperature, is insufficient to adequately describe and assess the insulation condition.

The system installed in TVA is the first monitoring equipment able to automatically perform an easy-to-interpret correlation with operating conditions. It can be used to suggest how to change operating conditions to mitigate PD effects and what offline tests can be planned or deferred. Data points are automatically aggregated at the same operating condition at the end of each week allowing reliable warnings and alarms to be set.

The PDE Map represents a powerful tool for O&M and Asset Managers to easily compare the condition of all generators by using just one image. It is simple to setup the same full-scale for the PDE level for each machine in order to quickly highlight overall condition. Useful, but complicated, tools such as PRPD patterns are still available. However, they are no longer used as a “first” level of information. Rather, they are used as an investigative mean only in case a particular PD behaviour is observed. A PD expert is no longer needed to determine that a machine is PD free, which frees up asset owners’ capital to invest in other areas of need. Not only can maintenance outages now reflect actual machine condition needs (as reflected in Unit 8), but asset owners can also avoid PD testing technician expenses and scheduling conflicts with a permanent continuous monitoring system.

Ultimately, a continuous PD monitoring system puts useful data in the hands of asset owners and helps identify corrective actions and maintenance recommendations. All asset owners look to extend material life and get the most payback from every maintenance investment. The system described in this report serves as an essential tool to accomplish these goals for generators.


Samuel Clemmons is a systems engineer for the Tennessee Valley Authority. James Hovious is a generator specialist at the Tennessee Valley Authority. Marco Tozzi is an electrical engineer for Camlin Power Ltd. Enrico Savorelli is product manager for Camlin Power Ltd.