Timothy P. Holtan, SmartSignal Corporation, Illinois, USA
New technology can provide an early warning of abnormal equipment performance in power plants. Such technology can help reduce derates and forced shutdowns by enabling plant personnel to fix small problems before they become large problems
Power generation operators are strategically adopting asset management to improve process efficiency and to increase return on assets (ROA). High value assets such as boilers, turbines, generators and auxiliary systems present an attractive target for asset management since they cause derates and forced outages when they fail.
A new technology, which ARC Advisory Group calls predictive condition monitoring, reduces forced outages and derates through actionable early warning of failure of critical power plant equipment.
Unlike preventative maintenance practices, which recommend maintenance based on failure statistics for a class of equipment problems over time, predictive condition monitoring provides equipment-specific, condition-based early warning. Predictive condition monitoring products, like SmartSignal’s Equipment Condition Monitoring (eCM), provide advanced, equipment-specific warning of deteriorating conditions leading to failure or poor performance of all equipment makes and types. In the power generation industry, Predictive Condition Monitoring provides early warning of failure of combustion turbines, steam turbines, boiler feedwater pumps, coal pulverizers, electrostatic precipitators and cooling water pumps.
This article first compares and contrasts Predictive Condition Monitoring with traditional technology, then explains the deployment process, and finally demonstrates the use of SmartSignal’s predictive technology to provide early warning of stator cooling water channel plugging in a
750 MW Westinghouse generator. In an analysis of historical data, this Predictive Condition Monitoring technology provided 400 hours of early warning.
Figure 1. SmartSignal eCM estimates values for each sensor in real-time based on the values of the correlated sensors
New versus old
Compared to traditional technologies, like first principle and neural network algorithms, SmartSignal eCM demonstrates significant advances such as monitoring multiple equipment operating modes like partial load conditions, analyzing multiple equipment and OEM types, and creating serial number specific models.
Comparing SmartSignal eCM to traditional threshold limit technology helps explain the predictive technology. In traditional monitoring, the manufacturer recommended upper and lower sensor threshold limits to initiate machine trips to avoid damage. Manufacturers set the thresholds based on deep first principle understanding of the equipment design parameters. In contrast, the SmartSignal eCM models sensor values of normal equipment performance instead of the design parameters for each serial number piece of monitored equipment. Doing so enables the software to quickly deploy enterprise wide fleet monitoring solutions compared to other technology such as neural network applications.
To start up, the SmartSignal eCM uses plant historical data to create a personalized, empirical model of the equipment’s normal operating range. Both ‘personalized’ and ’empirical’ are key distinctions: personalized because the model is for that equipment serial number and empirical because the model is built using only the actual operating data, no detailed engineering knowledge of the engine is needed at this stage. This empirical, serial number specific model of normal equipment operation creates an estimate for each sensor, in real-time, based on the values of correlated sensors. The software compares actual sensor values to estimated sensor values and detects subtle, but significant differences, called residuals. Residuals provide the basis for early warning of abnormal equipment conditions (Figure 1).
From a process standpoint, eCM starts by collecting a ‘snapshot’ of sensor values that make up an eCM model (see Figure 2). Next, eCM automatically creates an empirical model of normal performance of the asset using that statistical “snapshot.” Unlike other monitoring techniques, in real-time the eCM empirical model generates an estimated value for each sensor that would be characteristic of normal operation. As previously noted, each sensor estimate is based not only on that sensor’s history but also based on how that sensor interacted with every other sensor value.
The result is a data-driven empirical model of each asset. Then, in real–time, the software effectively removes the effect of normal operation by subtracting in estimated values from the actual values just collected to generate ‘residuals’. If the equipment is running normally, the resulting residuals should be small and evenly distributed around zero.
Equipment faults show up as spikes or trends in the residuals. eCM compares the residuals using a patented statistical technique. If significant deviations are found the equipment is running abnormally, and the eCM issues an ‘alert’. The ‘alerts’ are fed into the diagnostic rules engine, which analyses the pattern of alerts to see if this is a known pattern or if it meets the pre-established criteria for being promoted into an item on the WatchList and/or notifying an analyst. (These criteria are set-up during the SmartStart Installation Methodology phase.) Information about the incident is fed back into the eCM Database and all the information can be sent back to the control system or to a CMMS/EAM system.
Figure 2. The SmartSignal eCM development process
Main generator case history
The generator is a key piece of generating station equipment and the following analysis describes SmartSignal eCM monitoring of a 750 MW generator that provided nearly 400 hours of early warning of cooling channel plugging that resulted in a forced outage to clean the plugged cooling channels. Based on review of failure mode frequency, failure mode value, and sensor availability, engineers can develop a number of key process sensor groupings. Each model detects key failure modes that could result in forced outages or derates. Early warning of incipient failures provides generating stations with the opportunity to schedule a maintenance outage, to opportunistically make repairs during forced outages, or to devise means to address the problem on line.
In order to provide actionable and valuable information, SmartSignal typically breaks down the generator into four key models.
- Generator Stator Cooling Model: The stator cooling model normalizes generator heat removal for generator load, cooling water temperature, and cooling water flow. Deviations between actual and estimated stator cooling hose discharge temperature and cooling water pressure indicates a change in the heat balance of the generator. The most common failure mode associated with a change in the heat balance is cooling channel plugging. Over the long term, insulation degradation could also explain heat balance deviations.
- Generator Rotor Cooling Model: The rotor cooling model corrects rotor heat removal for changes in hydrogen flow, hydrogen temperature, and hydrogen pressure. Deviations between the model estimate and actual could indicate excessive heat generation, problems with the hydrogen circulation system, or problems with hydrogen water cooling system.
- Generator Mechanical Performance Model: The generator mechanical model relates bearing temperature and vibration and main lead return bend vibration with generator load. Deviations from the normal load relationship indicate possible key failure modes such as mechanical deterioration of rotating elements, lubrication problems, or bearing wear.
- Exciter Performance and Thermal Model: The exciter model normalizes exciter current and voltage and exciter temperatures for generator load. Changes in the estimated current and voltage could indicate mechanical degradation.
Each model will detect key failure modes that would result in forced outages or derates. Early warning of incipient failures provides generating stations with the opportunity to schedule a maintenance outage, to opportunistically make repairs during forced outages, or to devise means to address the problem on line.
Figure 3. Schematic of generator cooling system
This application note describes a situation where the generator stator cooling model detected signs of stator cooling channel plugging nearly 400 hours before the plant required a forced outage to remove the plugging material. For this case study, the early warning of failure would reduce maintenance expenditures by approximately $75 000 based on completing the maintenance on a planned basis. Furthermore, by scheduling the repair for off peak generation, the utility would have eliminated 36 000 lost peak MWh worth conservatively $180 000 at a peak-non peak differential of $5/MWh.
On a more general basis, discussions with generator equipment experts at a number of utilities have calculated the expected annual value of early warning of failure associated with these four models for maintenance expense reduction as approximately $76 000. Table 1 builds up the value case for generator modelling.
Description of findings
The SmartSignal eCM generator stator cooling model began noting statistically significant deviations in the stator cooling water pump suction and discharge pressure at constant cooling water flow rate. Negative residuals began to appear on the suction and discharge pressure indicating that even though adequate flow and cooling was maintained, soon the pump would run out of capacity. Figure 3 provides a schematic of the system.
Figure 4 illustrates the deviation between the actual and estimated value for the key pressures indicating the flow of cooling water through the stator cooling system. The first statistically significant deviations begin on May 15. The fouling finally limited heat transfer so severely that the plant needed to shutdown on July 5 to clean the cooling channel. The plant noticed the problem when the flow control valve ran to 100 per cent open. Even with maximum pump output, the fouling limited heat removal. The power plant took a forced outage during midweek in July to clean the system and restore the normal system pressure drop.
Response to early warning
If the generating station had used predictive technology during this time period they could have taken a number of actions to derive value from the early warning technology. First, they could have considered an off peak maintenance outage to clean the system. An off peak forced outage would have reduced the revenue impact of the problem. Another possibility would have been an online chemical cleaning. A number of vendors offer techniques to remove scale without a shutdown. Depending on the cost of the procedure and the value of incremental generation, an alternative maintenance procedure could have reduced the expense associated with this problem.
The financial impact of this forced outage amounted to 36 000 MWh of peak power. Shifting to non peak could have reduced the revenue impact by $75 000. Avoiding the shutdown altogether could have nearly eliminated the revenue impact with an increased maintenance expense.
Predictive condition monitoring, as exemplified by SmartSignal eCM, is a new technology for providing early warning of abnormal equipment performance. Actionable early warning of abnormal equipment performance can help reduce derates and forced shutdowns by enabling plant personnel to fix small problems before they become large problems. Some key benefits of this technology include the ability to monitor multiple operating modes, multiple OEMs and equipment types, balance of plant equipment (like ESPs and pulverizers) and difficult to detect failure modes, among others.
Lastly, SmartSignal effectively could have provided early warning of generator stator cooling system fouling – an excellent value to the utility. The 750 MW generator case study shows that a utility could save $75 000 through the use of predictive technology. In this case over 400 hours of early warning could have driven expense reduction and increased peak power production.