Wind turbines are unmanned, remote power plants which, unlike conventional power stations are very much exposed to highly variable, harsh weather conditions, ranging from calm to severe winds and conditions ranging from tropical heat, lightning, arctic cold, hail and snow. In addition, because of these external variations, wind turbines undergo constantly changing loads, unlike conventional power plants.
As a result of these highly variable operational conditions, there is high mechanical stress on wind turbines unmatched in any other form of power generation, and they therefore demand a high degree of maintenance to provide a safe, cost effective and reliable power output with acceptable equipment life.
Maintenance approaches in all industries can be broadly classified into three major groups:
The wind industry currently uses only Reactive Maintenance (fix it when it breaks) and Preventive Maintenance (following the wind turbine manufacturer’s service manual), and is not yet well versed in the newer forms of maintenance collectively known as Predictive Maintenance, which uses high tech condition monitoring technologies. Predictive Maintenance techniques and strategies are well known in more mature industries, such as in the energy (oil & gas) and utilities (coal and nuclear) sectors, as well as in the aircraft, military and major processing sectors, and the purpose of this paper is to demonstrate the benefits which the wind industry can expect from adopting these more modern predictive techniques and strategies to maintenance (collectively known as “PdM”) from these other, more mature industries.
As a starting point to this discussion, many studies have been done across a broad range of industries and these have shown conclusively that the first approach (Reactive Maintenance) is the least effective and most costly approach in the long run.
In the electric power industry, detailed case studies and an associated “Plant Maintenance Cost Justification” have been conducted in the US by the Electric Power Research Institute (EPRI), and these show the following results:
As noted, these studies by EPRI show an overall maintenance cost reduction of 47% is generally obtained by using predictive maintenance techniques, in comparison to the basic reactive approach, currently favoured by the wind industry.
In addition, Pat March, a senior mechanical engineer at the Tennessee Valley Authority Engineering Laboratory (one of the largest operators of electric power generation facilities in the US), prepared the following summary based on the results of a 1988 multi-industry survey of firms that had used or were using predictive maintenance techniques:
US aircraft carrier maintenance planners have used information from the Machinery Condition Analysis (“MCA”) program to define availability work packages since 1970. DLI Engineering Corporation has been the US Navy’s contractor for MCA since 1975 and Ship’s Force Machinery Vibration Analysis (“SFMVA”) since 1988. Since that time, information on the specific condition of over 400 pieces of rotating machinery on each ship has been determined and tracked for use in deciding if the machine should be repaired or overhauled, or if scheduled time-based work can be deferred. This information has made the maintenance and repair planning process more accurate and cost effective, and has reduced the incidence of failures and emergent industrial work for these machines.
Total Machine tests: 5,659, average cost per machine test: $ 161.00
Preventive Maintenance:
Preventive, or time-based, maintenance activities should, at a minimum, be followed according to the turbine manufacturer’s manual, supplemented by additional time-based items prescribed as a result of the real-world experience of the maintenance contractor. These activities usually comprise time-based turbine visits set at 6 month intervals, and include mechanical activities such as fluid level checks, greasing, bolt torque checks, filter changes, inspection of blades, inspection of brake pads, as well as electrical activities such as inspection of cable connections, fuse checks, voltage level checks, battery inspections, trip tests and electrical cable inspections.
These activities are essential, and their effectiveness can be greatly enhanced by the feedback obtained from coordinated predictive maintenance activities. It is even possible that some time-based activities may be reduced in frequency and scheduled only when indicated by the test results from the predictive maintenance activities.
Predictive Maintenance:
Predictive, or condition-based, maintenance activities can be broadly summarized as comprising equipment tests based on the use of on-line and off-line sensors and tests. The goal from these predictive maintenance activities is to produce an equipment condition assessment. They allow the plant owner/operator to plan in advance for ordering parts, scheduling work and planning for the repair-refurbish activities. A well organized predictive maintenance strategy, using state of the art sensors and tests will minimize unplanned turbine outages and maximize turbine energy output and revenue generation.
Experience has shown that wind turbine components routinely fail, to varying degrees, significantly before their design life is reached, resulting in significant unplanned (unscheduled) and therefore costly repairs, which is a problem be-deviling the wind industry in relation to the modern MW-class of wind turbines. Predictive maintenance tools installed on critical components, like gearboxes and generators, dramatically reduce the threat of sudden and unexpected failures in these components.
Described below are actual real case examples of the application of such preventive and predictive approaches to the maintenance of wind turbines or of components which are found in wind turbines:
No PdM was in use and a gearbox developed a bearing fault in January 2004, requiring the complete refurbishment of the gearbox, which necessitated the replacement of the existing gearbox with a new one. The 300 ton capacity crane was ordered as well as a replacement gearbox. The crane arrived and bad weather set in. The work was on hold for the next 3 weeks due to the weather and unsafe work conditions.
Result:
It is important to note that much of the breakdown repair cost arose from the crane waiting time, and as a result, WindRisk recommends using an advanced integrated weather forecasting module that monitors the wind farm for the following conditions:
Using this example, the following alternative result would have occurred if the WindRisk’s recommended PdM approach had been used:
Savings realized in the context of the above wind farm gearbox failure example using the WindRisk predictive maintenance approach compared with the traditional “run to failure” approach would be as follows:
A severe weather bulletin was issued 24 hrs prior to Feb 6, 2006 by Environment Canada weather services for the affected area, including a severe freezing rain warning followed by a severe wind warning with winds exceeding 100 Kph.
The operators did not implement any operational changes in response to the weather warnings. The freezing rain coated the nacelle and turbine blades with ice, which rendered the turbine shut-off mechanism ineffective, which meant that during the high wind event which followed, turbine shut off failure was likely, resulting in turbine high speed runaway, causing extreme tower vibration and resulting in the total loss of one turbine.
Loss: approximately $750,000.00
Using a predictive maintenance strategy, with its integrated weather forecast module, would have prompted the issuance of an order for the operators to shut all turbines down before the severe weather event, thus preventing the total turbine loss.
Study courtesy of Nancy Ettele, Massachusetts Water Resources Authority, Deer Island Treatment Plant.
Deer Island Treatment Pump, Massachusetts Water Resource Authority
Over the last two years, the Deer Island Water Treatment Plant has established an effective condition-monitoring maintenance program. The maintenance staff have been performing vibration–monitoring as well as spectral analysis, oil analysis, acoustic ultrasound detection, ultrasonic thickness testing, laser alignment and infrared thermography tasks.
Reliability-centered maintenance analysis of the plant’s primary scum pumps recommended a preventive maintenance task using acoustic ultrasound detection to monitor the motor and pump bearings to provide advance warning of potential failures. Using an ultrasonic detector, condition-monitoring engineer, Dan Parry, found unacceptable noises and noise levels in 10 of the 14 primary scum pumps, indicating potential bearing problems. Maintenance Planner, Michael Costs, then collected lubricating oil samples from six of the pumps gearboxes. The oil analysis showed high viscosity, indicating that the wrong lubricating oil had been used. The oil in all of the pumps was changed to the correct lubricant and ultrasonic monitoring was repeated. This time, only two pumps were found with unacceptable noise levels. An alignment check was scheduled before considering replacing the bearings.
While checking one machine’s alignment, mechanic Bob Greatorex and Peter McGee immediately saw that the coupling was in bad shape and noted that the machine was badly out of alignment. A new coupling was installed and the machine was laser aligned. Repeated ultrasonic detection confirmed that the problem had been solved. No further maintenance was required.
Preventive maintenance expenditures of about $280.00 for ultrasound, oil analysis, coupling replacement and alignment resulted in cost savings that can be estimated in two ways:
Avoided costs for this maintenance event range from $ 5,320.00 for scenario No.1 to $ 3,120.00 for scenario No. 2 for each machine.
When applied to all 14 primary scum pumps, expending $ 3,920.00 in preventive maintenance provided cost savings of $ 43,680 to $ 74,480 while increasing equipment availability and reliability. Condition-monitoring techniques used at the Deer Island have provided many such examples of the benefits and return on investment of proactive maintenance.
The Massachusetts Water Authority (MWRA) is a large, unionized public utility serving people in Boston metropolitan area. In 2000, the MWRA embarked on a comprehensive, multiphased asset management initiative. Its program includes dedicated staff along with a diverse senior management steering committee that organized early, communicated often, and conducted research and cross industry benchmarking. This effort allowed for a timely implementation of best practices, resultant efficiencies and cost saving benefits. The MWRA program is considered to be a model in public sector asset management.
Mick Flanigan is a vibration engineer at Intel’s Northwest Regional Operations facility in Hillsboro, OR., and is responsible for the implementation a corporate strategic maintenance program for Intel that encompasses all of Intel Corporation’s 12 major manufacturing sites, and in the process, standardizing on technology and best practices that are proving capable of saving the company massive amounts of money.
“We are trying to change minds”, says Flanigan, referring to the old-school maintenance mentality which he encountered at the beginning of his program, when he was often met with resistance at the individual facilities. “I can put a vibration sensor on equipment to diagnose a problem, but then someone else will say they too can tell if a bearing is bad simply by putting a screwdriver on it and putting their ear up to it.” It’s no doubt true that the technicians at Intel’s manufacturing sites are mechanically savvy and sensitive to subtle variances in equipment, but while the empirical approach works for the most part, it doesn’t get the crux of the problem, says Flanigan, which is, “can you tell me which bearing, where it is, and if something else is causing the problem ?”
These are important questions because the answers will affect the company’s ability to produce microprocessors. If any part of the facility fails, such as power supply, HVAC, water or chemical treatment systems, production could come to a rapid-and costly-standstill. “The reality is that replacing a fan or pump motor is a fraction of the cost of having a fabrication line down for any given amount of time”, Flanigan says. For that reason, Intel’s Hillsboro OR site launched a predictive maintenance program in 1998, using condition monitoring tools from Rockwell Automation to keep track of about 3000 pieces of equipment.
As of 2002, the program has helped the company avoid estimated lost production costs of more than $ 1.4 Million.
Much more savings will be realized once all of the manufacturing sites are using the same condition-monitoring technology. Historically, each of the Intel sites bought their own vibration and diagnostic tools, which work, but generate disconnected islands of information. “When I was asked to take part in the vibration program, I got the idea to get everyone on the same standard, using the exact same database, the same equipment, the same hierarchy, and network it all together, Flanigan explains. “We have the same pumps and equipment at each site, so there’s no reason why we can’t have the same software for easier statistical analysis. It becomes a powerful tool “.
That realization spurred action. Flanigan co-chairs an internal group called the Vibration Analysis Working Group, which took on the challenge of finding a way to share information across different sites. They chose a single package, which was Rockwell Software’s Enterprise Asset Health software, to be used universally among different Intel facilities. The system serves as a central repository allowing engineers and technicians to access any site’s database to examine equipment life cycle trends and determine if there are common failures occurring on specific types or models of equipment. The information itself is gathered on a monthly or weekly basis handheld data collectors (this procedure is also in the process of being improved).
Collectively, Flanigan uses all of his technology information to help persuade Intel plant managers around the world that a unified condition-monitoring program must be considered an integral part of Intel’s business strategy. Of course a couple of metrics to validate the program’s results in Oregon and a customized training course based on the program vibration-analysis technology also helps to make his case. “With the work we did in Hillsboro, people across the maintenance organization were able to see real cost savings, as well as the long term strategic benefits”, says Flanigan.
Savings of $ 1.4 Million in estimated lost production costs for 2002
Gearmesh vibration frequencies are typically easy to recognize, but not easy to interpret. This is due to two reasons:
For the analysis of suspected gear problems a high resolution spectrum analyzer is required to enable a high frequency range spectrum to be taken without loss of sideband data. Sidebands are very important and in most cases enable the analyst to determine which of the two meshing gears are at fault.
The following case history shows how spectral analysis proved invaluable in helping identify a problem gear in a cooling tower fan gearbox.
Figure 1, Vibration spectrum indicating gear damage.
Vibration data was collected from the electric motor by means of a vibration spectrum analyzer and a magnetically mounted accelerometer. Due to inaccessibility of the fan gearbox, permanently mounted accelerometers are fitted to the input and output shaft bearing housings and cabled out to a safe location. A spectrum was collected from the gearbox output shaft as shown in figure 1.
The input bevel gear (18 teeth at 990 rpm) meshing frequency could be identified. Harmonic activity at the input shaft frequency of 16.5 Hz are evident at either side of the gear mesh frequency. From this information, and the technical data from the gearbox, it was concluded that the input bevel gear had sustained some degree of tooth damage.
Figure 2, Layout of gearbox internals.
Figure 3, Damaged input bevel gear teeth.
Examination of the gearbox proved the analysis to be accurate, with damage to several of the bevel gear teeth evident, as shown in figure 3.
It is estimated that a saving of $ 9,000.00 was made due to early fault detection, however, had this unit catastrophically failed then secondary damage could have run into many times this amount as the potential for damage to blades and structure is high (for example, in the case of a wind turbine gearbox, secondary damage from a source such as this could commonly result in the catastrophic failure of the entire gearbox, costing around $ 200,000.
In this case, a Bently Nevada on-line condition monitoring system was registering higher than previously experienced vibration levels on the steam turbine driving a Demag compressor.
Historical data recorded the monitoring database revealed that vibration levels prior to the process unit shut down were normal for all bearings cap readings and also the Bently Nevada displacement probe data. Bearing cap readings that day were typically below 2mm/sec rms and gave no cause of concern. However, during start up with the machine running at 7170 rpm, the Bently Nevada probes indicated the highest overall shaft vibration relative to the bearings of about 32 microns. When the running speed was increased to 8500 rpm the highest overall shaft displacement increased to over 43 microns, see figure 4 below.
Figure 4, trend plot indicating overall vibration level changes.
Examination of the vibration spectrum exhibited a dominant component at the running speed frequency with an amplitude of 38 microns pk-pk indicating a probable imbalance condition of the turbine rotating element. Further questioning revealed that a new rotating element had been fitted to the turbine during the January 1995 shutdown, strongly suggesting the turbine as the source of the vibration, as levels on the compressor exhibited little change from previous tests.
To confirm this, the turbine was uncoupled from the compressor and run up to 7250 Rpm with the coupling boss still in place. The running speed component was now less than 10 microns, clearly the turbine itself was not at fault and attention switched to the coupling. At this was the same coupling that was producing under 15 microns prior to the shutdown, suspicion turned to the build up of the coupling and its associated packing pieces/shims. During inspection of the coupling it was noticed that the spacers between the coupling bosses and the spool piece had been fitted in a different configuration to that prior to shutdown.
The coupling was refitted according to all the match-marks and the coupled unit was run up to 9000 rpm. Highest reading of around 10 microns clearly indicated that the source of the imbalance had been correctly located and rectified. See figures 5 and 6 below (before and after).
Figure 5, before coupling correction.
Figure 6, after coupling correction.
The process unit was brought on-stream 12 hours ahead of the planned schedule, generating approximately $ 100,000.00 of extra production. Due to the prompt action in determining and correcting the problem, thus preventing any secondary component damage such as a wiped turbine drive end bearing, it is estimated that the cost savings due to the condition monitoring system were as follows:
K6321 is a centrifugal compressor driven by a 450 Kw motor through an epicyclic gearbox increasing the compressor speed from 2970 rpm to 17890 rpm. The machine is responsible for compressing H2S (Hydrogen Sulphide) and fuel gas from pressure of 0.5 bar to 7.5 bar and forms part of the 5 week vibration monitoring schedule. Historically, vibration levels at the electric motor non-drive end bearing were typically around 0.8 mm/s rms. However overall vibration data collected during the test period at the same location indicated a steep change, taking it above the pre set alarm levels of 1 and 1.3 mm/s. A further set of data was collected on the following day, which also indicated a further increase, albeit slight.
A data set for the electric motor was designed with a collecting specification of 1000 Hz and 3200 lines of resolution. Based on the fact that the overall levels were still considered low, and being aware of this increase it was decided to continue to run the machine and re-test the motor in 5 days time during which provisional plans to remove the motor could be made. Then a further set of readings were taken, and again an increase in overall levels was evident, see figure 7.
Figure 7, Continuing increase in overall vibration levels.
Examination of the vibration spectrum exhibited a peak at 292.5 Hz and tied in to the calculated defect frequency for the motor non drive end bearing inner race, see figure 8.
Figure 8, High resolution spectrum indicating bearing damage.
Based on this information, a recommendation to remove the machine was made. As the machine could not be spared, the production had to shut down. The motor was removed and sent away for bearing renewal, the old bearings were returned to site for inspection into their condition carried out. The drive end bearing was found to be in good condition with a plentiful supply of lubrication grease evident. The non-drive bearing however, had as predicted from the vibration spectral analysis, a fatigued area of 10 mm by 8 mm on the inner race, see figure 9.
Figure 9, Photographic evidence of bearing inner race defect.
After the bearing renewal, the motor was re-installed and aligned with the gearbox, with a further set of vibration readings being taken during run-up, and again after 10 minutes on line, recorded overall vibration levels at the motor non drive end were seen to have returned below 1 mm/s rms.
Due to the accurate condition monitoring analysis and prompt action from the production team, catastrophic failure to the electric motor was prevented, with estimated mechanical cost savings in the region of $ 150,000.00.
The above case studies clearly show the direct economic benefits of using predictive maintenance strategies, and highlight the special importance of using such strategies in the context of wind turbines, where their unique operational considerations create additional abnormal stresses and strains.