Professional Documents
Culture Documents
net/publication/289537160
PowerGen Gas Turbine Losses and Condition Monitoring: A Loss Data-Based Study
CITATION READS
1 869
1 author:
Bin Zhou
FM Global Research
17 PUBLICATIONS 174 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Bin Zhou on 08 January 2016.
95 using examples of the loss categories is included in this paper. Fig. 1 Percentages of total loss value and loss count by F1:1
96 While the true root cause of individual loss events may have been loss categories F1:2
97 ambiguous in a number of cases, the categorization described above
98 helps to identify CM variables for failure detection at the early
99 stages of the turbine loss.
100 The present study was based on PowerGen gas turbine losses of
101 insured fleet during a recent 10-year period. The study focused on
102 hundreds of gas turbine losses due to mechanical breakdown, which
103 was a dominant 96% of the total PowerGen gas turbine loss value.
104 The loss events spanned over all major turbine components. All loss
105 events were grouped into 13 categories as shown in Table 1, each
106 category representing a typical loss scenario or a group of similar
107 loss scenarios. Loss category no. 1, e.g., represents all loss events
108 with reported first known material failure of rotating blades and
109 subsequent domestic object damage (DOD) from released blade
110 mass. Typically, individual blade failure could be a result of several
111 potential material damage mechanisms, each of them forming the
112 basis of a unique failure scenario in this category. These failure
113 mechanisms may include blade vibration with consequent HCF
114 failure, blade thermomechanical damage, creep rupture or hot
115 corrosion, or cracking of the blade dovetail or attachment portion.
116 A given turbine loss event can be classified into loss category
117 no. 1, fitting either one of these loss scenarios. Loss categories Fig. 2 Normalized average loss value by loss categories F2:1
118 nos. 2 and 4 were created with an emphasis on the broad nature
119 of loss scenario rather than a particular family of failed components,
120 because these types of losses involve damage of multiple families
121 of components simultaneously; there was no clear indication unknown object damages (UODs) ranked right after. This indicates 145
122 of any incipient failure of a single family of components. It should that loss events as a result of impact damages are very common 146
123 also be noted that training and administrative-control-related in gas turbines. 147
124 issues would also contribute to mechanical hardware failures; how- The average loss value per event is compared in Fig. 2 for all loss 148
125 ever, they were not considered separately in the categorization categories. These values represent the severity of loss for each loss 149
126 process. category. Rubbing is shown as the most costly loss category, prob- 150
ably because it represents a type of system-level failure that involves 151
multiple families of component types including rotor, blades, bear- 152
127 2.2 Turbine Loss Data by Loss Categories. The reviewed
ings, vanes, seals, and casing. Average total loss value due to initial 153
128 gas turbine loss events were summarized in terms of total loss value
blade failure is the next highest on the list. 154
129 and numbers of loss events for each of the 13 loss categories out-
130 lined earlier. Figure 1 shows the percentages of total loss values and
131 loss counts for all loss categories, ranked in descending order by 2.3 Compressor Losses Versus Turbine Losses. Compres- 155
132 total loss value. The top four categories involve major blade dam- sor and turbine are the two main modules in the gas turbine, each 156
133 age, and the combined loss value represents more than 80% of the having a large number of rotating blades that are susceptible to vari- 157
134 total value of the losses reviewed. Loss events in these categories ous kinds of damage. The nature of mechanical failures in the two 158
135 also involve casing damage due to rubbing, vane damage as a result modules is, however, somewhat different. Understanding such a 159
136 of DOD or foreign object damage (FOD), and failures of attachment difference may help the appropriate application of CM methods 160
137 parts that are normally considerably smaller in value compared to in either module. Further investigation into the loss data was con- 161
138 blade damage. Loss dollar values due to blade damage further in- ducted to understand the characteristics and the composition of 162
139 crease because released material from damaged stator vanes or com- losses in compressor and turbine modules. In Table 2, loss events 163
140 bustors (loss categories nos. 5 and 9) may result in subsequent due to failures in the compressor and turbine are assembled sepa- 164
141 impact damage to the blades as well. The top four categories in total rately for relevant loss categories.2 165
142 loss value are also consistently the top four in terms of numbers of
143 losses. The blade-failure-initiated loss events occur most frequently. 2
On a few occasions, a single loss event was classified into two loss categories.
144 The numbers of losses due to attachments’ DODs and FODs or Some loss events incur both compressor and turbine failures.
247 instrumentation. With these limitations, turbine operators are left sources of clearance change may include casing contraction and 295
248 with only noncontact vibration monitoring systems for monitoring blade growth. 296
249 the rotating blades. Case-mounted noncontact blade vibration meas- Rotor lateral vibrations can be detected by measurement of dis- 297
250 urement (NBVM) sensors pick up a signal as each blade passes by, placement, velocity, and acceleration. This leads to common sensing 298
251 and the blade tip time of arrival is measured to infer blade vibration technologies including proximity probes, accelerometers, and veloc- 299
252 amplitude and frequency. Compared to strain gages, NBVM ity transducers. Rotor torsional vibration, on the other hand, is nor- 300
253 techniques have higher durability and also cover all blades in a mally measured by torque sensors. Proximity probes are often 301
254 given row. mounted on plain/fluid bearings to directly monitor the vibration 302
255 Resonance frequencies as well as vibration amplitude and phase of a rotating shaft, through the measurement of relative motion be- 303
256 of concerned blade vibratory modes are captured by the NBVM tween the rotor and bearing without contact. When dually used as 304
257 system. A significant percentage shift of resonant frequency of blade row tip clearance or blade vibration monitors, these probes 305
258 an individual blade may be an indicator of blade damage and/or are mounted on the casing directly above the blade rows, with con- 306
259 loose attachment [3]. The high vibration response amplitude of nected data acquisition and analysis systems adjusted accordingly. 307
260 all blades due to any type of abnormal vibration can be monitored Different from heavy industrial turbines, aeroderivative turbines 308
261 in real time for violations against alarm and trip limits to protect the or gas generators use rolling element bearings, where case-mounted 309
262 turbine from hardware damages automatically [3]. The blade static accelerometers or velocity sensors for vibration measurements 310
263 positions are observable to NBVM systems in addition to the vibra- are required, as no relative movement between the shaft and the 311
264 tion signatures. While the original blade static positions at a given rigid bearing would occur. Case- and frame-mounted sensors can 312
265 speed and power setting are dictated by as-manufactured geometry also be used to detect high vibration leading to HCF of these 313
266 and assembly, the unique positional signature can be altered at the structures. 314
267 same running condition by rubbing, surge, cracking, and impact
268 damage. The blade row static position can be continuously moni-
269 tored with imposed protection limits on deviation from the original 3.3 Temperature, Pressure, and Performance Monitoring. 315
270 profile [4]. Temperature measurements are widely applied to monitor condi- 316
tions for lube oil, fuel, air (exhaust spread, inlet, extraction and in- 317
jection, rotor stages, etc.), water or steam injection, and metal 318
271 3.2 Rotor/Case Vibration and Clearance Monitoring. (usually bearing). Typically, excessive temperature represents an 319
272 These condition variables are grouped together not only for their early sign of malfunction or failure. Temperature monitoring is 320
273 intrinsic cross relations but also because the monitoring technolo- normally done with temperature sensors including thermocouples 321
274 gies involved are the same or similar. and resistance temperature detectors, as well as noncontact infrared 322
275 Vibration of the rotor in a gas turbine is a system-level issue and thermometers when applicable. 323
276 can lead to damage of virtually any turbine component. Root causes Static- or steady-state pressure sensors measure pressure at a 324
277 of rotor vibration include, but are not limited to, damaged/worn steady-state for a given operating condition at a given location. 325
278 bearings, imbalance, rotor bow, misalignment, improper mounting, Measurements include stage pressures, fuel pressure, lube oil pres- 326
279 rubbing, undesired bearing and seal fluid conditions, and aerody- sure, inlet filter delta pressure, and exhaust backpressure. Abnormal 327
280 namic instabilities. These factors are interconnected and contribute measured pressure may be an indicator of leakage, blockage, or 328
281 to the change of the rotor’s natural vibratory modes or critical aerodynamic instability. The so-called “dynamic” pressure sensors 329
282 speeds, the mechanical, fluid-dynamic and aerodynamic loading, or unsteady pressure sensors measure pressures that change rapidly. 330
283 as well as the system damping. Vibration of the casing and frames For gas turbines with lean or dry low-NOx combustion, the pressure 331
284 is normally linked to rotor vibration through bearings, although pulsations in the combustion chamber are commonly monitored to 332
285 it may also come from other sources such as foundations. Rotor avoid the onset of combustion instability that would cause failure by 333
286 vibration and blade vibration can lead to each other. While blade HCF of combustor components. 334
287 damage and material loss will cause rotor imbalance and vibration, Performance monitoring uses measured temperature, pressure, 335
288 rotor vibration will result in blade tip rubbing, cracking, and further and/or flow measurement to calculate heat balance or turbine effi- 336
289 DOD damage downstream. ciencies, and power output online during turbine operation. Perfor- 337
290 Blade row tip clearance is an important parameter for turbine mance monitoring allows proactive responses to incoming failures, 338
291 operability and performance. Tight tip clearance helps to maintain understanding of long-term trends of deterioration, and perfor- 339
292 turbine performance but at the same time makes the tip prone to mance degradation so as to improve the scheduling of maintenance. 340
293 rubbing. The main source of rubbing is rotor eccentricity or run- Measurement of mass flow is typically inferred from measurements 341
294 out, which often times is associated with rotor vibrations. Other of pressure and temperature. 342
5 Conclusions 452
Loss events reported as PowerGen gas turbine mechanical 453
breakdown in a recent 10-year period were reviewed. Loss events 454
were grouped into 13 loss categories that supported the understand- 455
ing of typical gas turbine loss scenarios and associated failure mech- 456
anisms. Major groups of condition variables were subsequently 457
identified, so that continuous real-time monitoring of these variables 458
would lead to detection of failures and improve loss prevention. 459
Failure detection and monitoring technologies associated with these 460
variables also were reviewed. Lastly, identified groups of CM var- 461
iables were evaluated and prioritized based on the effectiveness of 462
F6:1 Fig. 6 Percentages of total loss value and loss counts failure detection for these known losses. The summary of findings is 463
F6:2 detectable by groups of CM variables presented below: 464
Acknowledgment 478
Support from Kumar Bhimavarapu, William Doerr, Erik 479
Verloop, and Franco Tamanini in reviewing and discussing the 480
contents of the manuscript is greatly appreciated. This work was 481
supported by FM Global internal funding. 482
F7:1 Fig. 7 Blade and rotor vibration monitoring
References 483
429 of total loss value, which can be detected by each group of CM [1] Simmons, H. R., Brun, K., and Cheruvu, S., 2006, “Aerodynamic Instability Ef- 484
fects on Compressor Blade Failure: A Root Cause Failure Analysis,” Proceedings 485
430 variables. Both blade vibration and rotor vibration monitoring of ASME Turbo Expo, Barcelona, Spain, Vol. 5, pp. 649–660. 3486
431 can effectively detect a dominant portion, i.e., 94% and 97%, re- [2] Latcovich, J. A., 2002, “Condition Monitoring and Its Effect on the Insurance of 487
432 spectively, of the total loss value. Detections of abnormal temper- New Advanced Gas Turbines,” Turbine Power Systems Conference and Condition 488
433 ature, pressure, quality of fluids, and material cracking or Monitoring Workshop, Galveston, TX, pp. 25–27. 489
[3] Roorda, B., 2009, “Prognostic Health Management (PHM) System for Monitoring 490
434 degradation on stator vanes and cases rank lower with 45%, Blade Vibrations in Turbo Machinery,” Propulsion—Safety and Affordable Read- 491
435 16%, 10%, and 6% of the total loss value of all the loss events. iness Forum, Myrtle Beach, SC. 492
436 A consistent trend or ranking can be observed on the percentage [4] Zielinski, M., and Ziller, G., 2005, “Noncontact Crack Detection on Compressor 493
437 of loss counts detectable by different monitoring variable groups. Rotor Blades to Prevent Further Damage after HCF-Failure,” NATO, RTO 494
(Research and Technology Organization)-MP (Meeting Proceedings)-AVT-121, 495
438 The loss events attributed to the top two groups were further Applied Vehicle Technology Panel Symposium, Granada, Spain. 496
439 investigated in order to differentiate the sequence and promptness [5] Sheng, S., 2011, “Investigation of Oil Conditioning, Real-Time Monitoring 497
440 of the failure detection. As shown in Fig. 7, 50% of the total loss and Oil Sample Analysis for Wind Turbine Gearboxes,” AWEA (American Wind 498
441 value detectable by blade monitoring is actually initiated by the Energy Association) Project Performance and Reliability Workshop, San Diego, 499
CA. 500
442 blade itself or FOD, with the other 44% attributed to secondary [6] Sheiretov, Y., Grundy, D., Zilberstein, V., Goldfine, N., and Maley, S., 2009, 501
443 blade damages following other component-failure-induced DOD, “MWM-Array Sensors for In Situ Monitoring of High-Temperature Components 502
444 rubbing, and other failures. On the other hand, only 29% of the total in Power Plants,” Sens. J., 9(11), pp. 1527–1536. 503
445 loss value detectable by rotor vibration monitoring is actually ini- [7] Momeni, S., Koduru, J. P., Gonzalez, M., and Godinez, V., 2013, “Online Acoustic 504
Emission Monitoring of Combustion Turbines for Compressor Stator Vane Crack 505
446 tiated by the rotor system itself, including typical rubbing, bearing Detection,” Proceedingsof SPIE, Industrial and Commercial Applications of Smart 506
447 vibration, and damage; the other 68% is due to rotor vibration Structures Technologies, Vol. 8690, pp. 86900B. 507