You are on page 1of 53

Generic ageing characteristics of conventional power plants

Lessons learned from data-analysis a model for life extension planning


H.C. Wels Arnhem, 5 February 2007

911569/07.81244/C

Under the contract of EZ

author :

H.C. Wels

reviewed :

J.L. Brinkman

53 page(s) PPT/Wels/HGo 911569_07.81244_rapport.doc

approved :

A.J. Geutjes

NRG 2007
Subject to agreement with the client, the information contained in this report may not be disclosed to any third party and NRG is not liable for any damage arising out of the use of such information.

911569/07.81244

Contents
List of figures Summary 1 2 Introduction NERC data 2.1 Background of NERC data 2.2 Analysis method for NERC data 2.3 Statistical uncertainty in average values NERC results 3.1 Total plant 3.2 Results for components VGB data 4.1 Data for total plant 4.2 Dominance with regard to unavailability 4.3 Dominancy with regard to trips Application of failure data for Life Extension Studies Conclusions and recommendations 4 5 9 9 9 10 11 13 13 16 24 25 29 37 38 46 48 49

5 6

Literature appendix A Unavailability of individual VGB plants

911569/07.81244

List of figures
figure 1 figure 2 figure 3 figure 4 figure 5 figure 6 figure 7 figure 8 figure 9 figure 10 figure 11 figure 12 figure 13 figure 14 figure 15 figure 16 figure 17 figure 18 figure 19 figure 20 figure 21 figure 22 figure 23 figure 24 figure 25 figure 26 figure 27 figure 28 figure 29 figure 30 figure 31 figure 32 figure 33 figure 34 figure 35 figure 36 figure 37 uncertainty in failure data as a function of number of data NERC, unit EFOR as a function of age NERC, uncertainty in unit EFOR as a function of age NERC, failure rate as a function of age NERC, uncertainty in failure rate as a function of age NERC feedwsater system NERC, boiler control NERC, flue gas fans, ducts, dampers, air heater Ageing rate from NERC data Ageing rate from NERC data compared with TIRGALEX data main characteristics VGB plants operating time VGB plants Average VGB plant data Individual VGB plants Components dominant to forced unavailability, VGB data Pattern in time, HAD = evaporator Pattern in time, HNC = flue gas induced draft fan Pattern in time, LBA = steam piping Pattern in time, MAA = HP turbine, MAB = IP turbine Pattern in time, MK = generator, MKA = generator rotor/stator average no of damages per year, VGB data average repair time, VGB data Minimal repair Example of HLD = air heater failures Example of HNC = flue gas induced fan failures Example of LAD = preheater failures Example of MKA Generator stator en rotor failures Pattern for trip frequency Repair time for C&I failures, VGB data Example of a Reliability Block Diagram Use of a reference block diagram model Example of a Delphi exercise Finite number of system states in a Markov Decision model System states derivable from maintenance records Condition as a function of time by Markov modelling Inspection optimisation RAM data 12 14 14 15 15 17 18 19 21 23 25 25 26 28 29 30 30 31 31 32 33 33 34 35 35 36 36 37 38 39 40 41 42 43 44 44 45

911569/07.81244

Summary
Background Operation of electricity production units in the Netherlands over periods longer than 25 yrs was uncommon in the Netherlands. However, in a liberalised market, building of new units is irregular and existing units are used as long as possible. Life extension scenarios up to 40 yrs of operation or more are presently considered. Centrally in these scenarios is the question whether forced unavailability and/or frequency of failures of components will increase as a function of operation time within this new time frame, if for those components no life extension measures are taken. In the report observations are given from data analysis of conventional power plants between 1982 and 2006. Since in E-production units systems and components are present that are also common to other plants (industrial, nuclear) lessons learned can be applied to other sectors of industry and vice versa. Data sources To answer the question stated above, data have been used from NERC, VGB and a number of studies that NRG carried out for its clients. These datasets are however not equivalent. NERC data represents over 90 US power plants and can therefore well be used to establish statistical trends. With the data available to NRG, individual power plants as well as detailed background information on the failure could not be investigated. However in the VGB data and the data from NRG studies this information was available. Therefore the available information in the three data has to be combined to allow pinpointing generic trends. Results Analysis shows that only for a limited number of components in E-production units older than 25 years a generic increase in failure rate or a generic increase in average repair time can be expected. On average, these increases are compensated for by decreases (especially in failure rate) of other components. On average for the plant as a whole (when maintained properly), forced unavailability FOR does not increase. A type of roller coaster curve exists showing a reduction in FOR up to 20 years, a possible increase between 20 and 25 years and a constant average FOR up to 35 years of age. However, the uncertainty beyond 25 years is large, showing that problems might depend on the specific plant and/or component. If FOR increases, data analysis shows that this might be due to: certain problematic constructions (for instance cyclone burners); to minimal repair; to specific damages to a few components, or; to a change in operational mode (for instance daily cycling).

911569/07.81244

Components to investigate Components that should be investigated when the intention is to operate longer than 25 years (due to an increased failure rate and/or average repair time) are: Cold & hot reheat steam lines Boiler controls Burners Bottom ash system Desuperheaters Condenser (if other than titanic) Flue gas fans, air heater, duct expansion joints, dampers Generator exciter, rotor, stator Feedwater heaters LP and IP turbine Turbine controls Turbine valves Forced draft fans Given sufficient attention to ageing components, analysis of life usage factors for critical boiler and turbine components and normal preventive maintenance, the forced unavailability FOR as well as the trip frequency does not have to rise between 25 and 40 years of operation. A large amount of uncertainty is present with regard to the operation of older electrical machines such as generators and transformers. Overhaul strategy Normal preventive maintenance at power plants during overhauls is largely related to inspection. The normal practice in the electricity production sector is to carry out inspections on the basis of fixed intervals, which might have been lengthened since the start of operation. Given the uncertainty in behavior of old plant, it might however be optimal to inspect on a condition basis and carry out inspections more often when the expected failure time comes closer. Environmental NERC data seem to point to an increased number of outages that are related to environmental legislation. Application for life extension Power plant failures can be subdivided into normal failures without ageing, ageing failures and High Impact Low Probability HILP failures. Normal failures can adequately be modeled by using a Reliability Block Diagram for the unit and validation with plant specific data. Ageing failures either can be deduced from trend-analysis of plant specific records or generic databases such as presented in the report. Finally, HILP failures should be deduced taking the specific condition of components into account. This procedure was carried out and confirmed at life extension analysis for clients. Given the limited amount of failure data for some failures, expert judgment is necessary. This judgment can

911569/07.81244

be based on a simple Delphi estimate of the failure probability over a period of time for replacement decisions. It should be based on degradation probabilities in combination with a Markov model for optimization of inspection frequencies.

911569/07.81244

911569/07.81244

Introduction

Knowledge of the forced unavailability of power production units that operate longer than 25 years is very relevant in the Netherlands. The forced unavailability of plants younger than 25 years is very well known to NRG because of earlier data-gathering projects. However, costs per MWhr not delivered due to failures as well as the E-price have been rising in a commercial market with irregular newbuildings and units that are operated as long as technically and economically feasible. Therefore data from NERC, from VGB and from NRG clients have been investigated in order to answer 2 questions: Will the forced unavailability of components in power plants older than 25 years rise and, if so, for which components and to what extent? Is this a generic trend present at the majority of plants?

If so, then it would be wise to take this into account when life extension scenarios and associated activities are considered at power plants. As a first step, data in the NERC database purchased by NRG were investigated. Next, VGB data have been analysed. Also data from studies for NRG clients have been analysed in this respect. For confidentiality reasons, clients and names of power plants have to be protected. Conclusions from the investigations were compared with earlier NRG studies as well as literature research for confirmation purposes.

2
2.1

NERC data
Background of NERC data

The North American Electric Reliability Council NERC provides the GADS Generating Availability Data System. In this system, since 1982 failure data have been gathered from US power production units. Standard reports can be found on the Internet, however these do not allow making a selection over age, etc. Over 90 % of the total installed capacity in the US is contributing to the GADS system. For the NRG analysis, data up to the year 2000 have been used. Since so many power plants are present, the statistical uncertainty in average values is small. However, this database for the analysis under consideration has 3 important disadvantages: - The characteristics of specific power plants are in principle unknown. The names of power plants that contribute to the data in a specific NERC subset are known, however Internet investigations to find out more about these specific plants and the components in the plants were inefficient and therefore had to be stopped. Therefore, it is uncertain that NERC plants are sufficiently comparable to Dutch plants. - Event data were not available to third parties without contacting NERC. Therefore, it was not possible to check the background of any increase in the number of failures per hour or average repair time.

911569/07.81244

The Database Management System (Clarion TopSpeed) is slow, possibly due to the security characteristics. Therefore it was not possible to transfer the data to ACCESS or Excel in order to speed up the analysis. However, standardised reports could be transferred to Excel (so called CSV files). This facility was used for the majority of the analysis.
Analysis method for NERC data

2.2

In the total NERC dataset, subsets have been made for the category Conventional Steam. In these subsets therefore coal fired as well as oil- en gas fired units are present. It is expected that modern Combined Cycles are not present. Single cycle gasturbines (GTs) and Jet Engines (aero derivative GTs) are separately named in NERC and therefore are sure not to be present in the subsets1. Three main selections for the NERC datasets have been made: A Units that started operation between 1960 and 1965. B Units starting operation between 1965 and 1970. C Units starting operation between 1970 and 1975. The units in category A have an approximate age of 40 years at present. The units in category C are the most relevant for Dutch Life Extension studies. Average operating time for all categories is reasonably constant between 6000 and 7000 hours per year. However, the operating time can differ appreciably between units. As an example: units between 35 and 40 years old with data in the period 1995-2000 in the NERC system have an average operation time of 5665 hours per year with a standard deviation of 2459 hrs. The selections on start of operation implicitly are also a selection to technology applied and between 1960 and 1970 power production units have increased in size. Installed power for this period has risen from less than 100 MWe to over 500 MWe. Within each selection category, failure data were averaged over 5 years of operation. Please note that when for instance Forced Availability is plotted as a function of the age for a subset of units, this automatically implies that calendar years differ. If for instance at a particular year an environmental measure is enforced by authorities, the effect of this measure will show up at different points in time in the plots. The coding of components and subsystems in the NERC database is rather elaborate. Some 750 Cause Codes are a mixture of components and failure modes. As an example: code 265 = Primary air heater, code 266 = Primary air heater fouling. Since with increasing depth of

a short analysis of the operation of older GE Fr5 gasturbines has been carried out in a recent Failure Mode Effect Analysis.

10

911569/07.81244

coding the number of events per code becomes less and therefore the average value is statistically less certain, NERC codes have been summed using Excel. This has resulted in 95 codes that allow pinpointing the major components in E-production units. As an example: NERC codes 1400 to 1450 have been condensed to Air Supply fans & Ducting. The failures per code were added, divided by the number of E-production-unit-years and converted to operating hours rather than calendar hours. Since the number of components per codes is not precisely known, the failure rates thus calculated must be regarded as system failure rates, not component failure rates. Engineering judgement to convert to component rates is possible, however also in the Netherlands one forced draft ventilator may be present instead of the usual 2 ventilators in a power plant. Total repair time has also been added and allocated to the appropriate codes. Repair time is in equivalent hours, that is: weighted with the power loss involved. Dividing of total repair time by the number of failures results in the average failure duration. Multiplication of average failure duration with the derived failure rate should result in the FOR. However, then adding all FORs, it turns out that the sum is some 50 % larger than the FOR and EFOR of the unit in total. The most probable cause is that the NERC database allows for overlap of failures between components, contains (preventive) maintenance information and failures of redundant components (coal mills, feedwater pumps) may be present.
2.3 Statistical uncertainty in average values

Given a discrete number of failures, the calculated average failure rate is statistically uncertain. As shown in figure 1 (source: EPRI) for example the uncertainty in the mean time between failures = 1 / failure rate at 10 failures is - 35 % + 110 % (at 90 % confidence level). The figure can be also be calculated with a CHI-square distribution in Excel. In the present report, a standard 10%-90% uncertainty range was taken as standard confidence level. If one assumes a CHI-square distribution also for the average repair time, the uncertainty in repair time can be calculated also. In principle however, this is a crude approximation since very small repair times cannot be present. However, also the commonly used lognormal distribution has its drawbacks: some extreme repair times originating from this distribution cannot occur in practice due to replacement of the component or application of spare parts.

911569/07.81244

11

figure 1

uncertainty in failure data as a function of number of data

[1] Showed that the uncertainty in FOR over a period is a function of both the occurrence of failures and repair time. This can also be demonstrated using Monte Carlo analysis. The occurrence of failures over a certain period is defined by failure rate. If the failure rate is high, both failures and failure rate are relatively certain. If the failure rate is low (this is especially the case with High Impact Low Probability HILP failures), failures are uncertain. Repair times are described by a statistical distribution around the average repair/outage time. Please note that for specific components such as step-up transformers or generators it may be necessary to use 2

12

911569/07.81244

distributions in order to describe extreme repair times. Especially with regard to HILP failures, both outage times and the occurrence of failures are uncertain. FOR is therefore a function of 2 uncertain variables (failure rate and repair time) AND a function of the length of the period analysed. Uncertainty increases even beyond this when the failure rate and average repair time are uncertain themselves due to different manufactors. Given the above discussion, one can decide on ageing separately on the behaviour of failure rate and the behaviour of average repair time as a function of age.

3
3.1

NERC results
Total plant

Figure 2-4 show the number of failures per hr and Equivalent Forced Outage Rate (EFOR) for the total plant as well as their uncertainty, given by the standard deviation. The parameters are shown as a function of age. It is interesting that, depending on the starting years (which can be interpreted as state of technology) units on average are getting better in terms of lower failures / hr and EFOR. Improvement in EFOR stops beyond 25 years and uncertainty after this period rises again. There is only 1 data point present beyond 35 years, therefore the results after 35 years are inconclusive. Now, if the failure rate per hr is going down and EFOR stays equal, this might be due to repair times rising. Repair times may rise because spare part unavailability problems may occur OR the type of failure is different, having larger repair times. Since problem descriptions in NERC are not available to NRG, this could not be checked.

911569/07.81244

13

figure 2

NERC, unit EFOR as a function of age

figure 3

NERC, uncertainty in unit EFOR as a function of age

14

911569/07.81244

figure 4

NERC, failure rate as a function of age

figure 5

NERC, uncertainty in failure rate as a function of age

911569/07.81244

15

3.2

Results for components

In this paragraph, three examples are given with regard to ageing of components: NERCs Feedwater System, Boiler Control Systems, and Flue Gas Fans, Air heater, Ducts Exp Joints, Dampers. Per figure three parameters are given: - average number of failures per operating hr (failure frequency, failure rate) - average repair time - forced unavailability. Per component also the uncertainty range is given, using a CHI2-distribution for failure rate and repair time and Nelsons approximation [1] for forced unavailability. The uncertainty range can be used to check that, given enough data material, average values are relatively sure. In figure 6 the Feedwater System is shown. Independent on the category of the plant (operation starting in 60-ties to operation 70-ties), the failure frequency decreases. All categories seem to belong to 1 single line. The average repair time per category is not increasing as a function of age. The uncertainty in unavailability over a period of 5 years is such that no significant statistical differences are present. A significant statistical difference would implicate that differences cannot be contributed to chance. In a second example, Boiler Control Systems are shown in figure 7. It is clear that the average repair time increases as a function of age. This could be related to obsolescence, resulting in difficulties having enough spares. Trends like this also can be found at Generator Controls and Steam Turbine Controls. A third example is shown in figure 8 for the component Flue Gas Fans etc. In figure 8 one can observe an increase in the failure frequency as well as in the average repair time as a function of age. The failure frequency for units starting their operation in the 70-ties increases the most. The result is an increase in forced unavailability for this component, albeit probably statistically not significant.

16

911569/07.81244

NERC, failure frequency feedwater system

figure 6

NERC feedwater system

911569/07.81244

17

figure 7

NERC, boiler control

18

911569/07.81244

figure 8

NERC, flue gas fans, ducts, dampers, air heater

911569/07.81244

19

An interesting result is that from the NERC data analysis for the coding "continuous emission monitoring systems", "operating environmental limitations", "regulatory" en "safety" it can be concluded that for older units the authorities intervene more often, with results for unavailability of the plant. Analysis of NERC components defined above shows that sometimes the failures rate increases, but more often the failure rate decreases as a function of age. Especially for controls, the average repair time may increase. On a generic basis, those components should be investigated when assessing Life Extension scenarios. The costs for modification and/or life extension for these components should be weighed against the (increasing) costs for unbalance and replacement energy when the component fails. For the components where either the failure rate or repair time increased as a function of age, this increase was read from figures like 6 to 8 and tabulated. Also, if the starting value could be sensibly read from the figures, it was tabulated. In order to arrive at general conclusions, components were placed into categories (f.e. control, heat exchanger, etc.) The result is shown in figure 9.

20

911569/07.81244

figure 9

Ageing rate from NERC data

911569/07.81244

21

Figure 9 (ctnd) Ageing rate from NERC data

The category components can be compared with the TIRGALEX data, a set of data from NUREG expert judgements from early incorporations of ageing into PSA. The comparison is given in figure 10.

22

911569/07.81244

figure 10

Ageing rate from NERC data compared with TIRGALEX data

911569/07.81244

23

The comparison shows that the order of magnitude for the failure-ageing rate is the same for TIRGALEX and NERC. This gives confidence in the ageing rate derived from the datasets. In the category FAN the ageing rate is much higher, but this is only due to 1 data point. The additional number of failures over a period of 15 years is low, compared to other possible maintenance effects on failure data such as ineffective repair when not removing the root cause of the problem.

VGB data

The German institute Vereinigung Grosskraftwerk Betreiber VGB gathers already for a long time forced unavailability data of power plants of its members. These data can be made available to third parties under strict conditions of confidentiality. For an analysis of the effects of ageing on plants, data were made available to NRG. Names of power plants and companies are not recognisable from the data. Power of the unit, first year of operation, type of unit (so called monoblock or twinblock (2 boilers on 1 turbine), fuel (hard coal, lignite), conventional v.s. combined cycle, etc. were present in the VGB information. VGB made available all data of units between 1988 and 1997 with an age over 25 years. Therefore, although the amount of data is considerable (14.900 failure data), units cannot be followed from the first year of operation until the last year. However, generic trends beyond 25 years of age (if present) should be visible from this dataset. VGB essentially gathers 2 sets of data: - the so called Verfugbarkeits Statistik, with data for the plant as a whole (forced and planned unavailability, operating time, etc.). For conventional plants (with the exception of single cycle GTs), in this database failures during starts are not gathered. Similarly, no difference is made in trips, failures during operation and failures with the plant stopped. - the so called Nicht-Verfugbarkeits Statistik, with data for the failure of systems and components. Component failures can be deduced from these records if the failure results in forced unavailability of the plant. Therefore, the majority of failures of redundant components is not visible. With the aid of the so-called Ereignis Merkmal Schlussel (EMS, part-coding 1 & 4) one can analyse the background of the failure (damage? maintenance? operating error?) and the consequence (automatic trip? out of operation by operator? postponed to the weekend?). Important is an (although limited) amount of text describing the problem.

24

911569/07.81244

4.1

Data for total plant

In the VGB dataset data for 180 plants are present. The main characteristics of these plants are given in figure 11. The operation time as a function of fuel is shown in figure 12.

figure 11

main characteristics VGB plants

figure 12

operating time VGB plants

The operation time per year for lignite fired units and (to a lesser amount) coal fired units is very similar to that in the NERC data. This type of plants is operated in base load or with weekend stops. From some failure descriptions, it appears that some plants are also operated seasonally. The oil-fired units have low operating hours per year. Since the number of failures can be shown to be dependant on operating hours, plants with less than 100 operating hours per year were not taken into account. In total 9066 unit years with 753790 operating hours have been analysed in the VGB dataset. VGB defines failures as postponable and direct (in German disponibel and nichtdisponibel). Postponable is defined as an unavailability event that can be postponed over 12 hrs but not for more than 4 weeks. Direct is a direct failure or an event that can be postponed up to

911569/07.81244

25

12 hrs. Direct failures evidently are more important than postponable failures and have therefore been analysed. Per plant per year the equivalent forced unavailability EFOR was plotted. Average results are given in figure 13, individual plants are given in Appendix A. Figure 13 shows that the average Equivalent Forced Unavailability EFOR does not increase as a function of age. The average, the 10 % lower bound and the 50 % value rise somewhat between 30 and 40 years of age. It is the standard deviation between 30 and 35 years that is much larger than in the years before and after that period. From this increase it can be concluded that uncertainty in EFOR over that period is larger, although the number of data points beyond 35 years is small. When uncertainty is large, it means that some plants have problems and some dont. This roller-coaster curve was found earlier in NRG analysis of power plants.

figure 13

Average VGB plant data

It was tried to find the most meaningful way of presenting the large number of data, in such a way that individual power plants showed up. For each 20 units, an Excel overview was made with years and power units coloured whenever the EFOR was over 5 %. Two examples are given in figure 14. It appears that power units are present with occasionally bad years (see R054B in figure 14) and plants that (given their coding at the same location or sister units, see K141.3-K151.5) in general have an EFOR above average. For those plants with years with over 5 % EFOR, the data in the VGB Nicht-Verfugbarkeits Statistik were analysed with regard to the nature of the problems. In total over the period 19881997 14.390 data were present. Given this large number of data, analysis must be structured and therefore only components have been investigated that are dominant with regard to EFOR.

26

911569/07.81244

With modern power markets, 2 important costs are the consequence of plant failures: A costs for replacement energy in order to fulfil power contracts that have already been signed. B unbalance costs in order to keep up the grid frequency and supply short-term energy by other units generally not belonging to the same company. Unbalance costs can be limited to a 2 hr period if the company that is causing the unbalance solves this by buying additional power or curtailing load from its customers. The presence of term B means that not only classical (E)FOR as a fraction of time not delivering power is present, also reliability defined by the number of failures per hr is important. Now, components that are dominant with regard to EFOR are not necessarily dominant with regard to reliability. An example is failures due to control and instrumentation C&I in power plants. Given their failure rate and (small) repair times, C&I failures are dominant with regard to reliability, but generally not dominant with regard to FOR. Therefore, in the VGB data both dominances have been analysed.

911569/07.81244

27

figure 14

Individual VGB plants

28

911569/07.81244

4.2

Dominance with regard to unavailability

In the VGB Nicht-Verfugbarkeit database, all data were selected with EMS-code A2 = damage. In general these failures have a somewhat larger repair time compared to for instance operation errors. Components (systems and sub-systems) are indicated with the Kraftwerk Kennzeichen System (KKS). For this dataset, dominance with regard to unavailability was analysed. The procedure was: Calculate the contribution of a KKS-code (f.i. MAA = HP turbine, MAD = turbine bearings, etc.) to the total unavailability present in the dataset Sort KKS-codes on contribution According to the Pareto-principle, only a very limited number of components will cause the majority of the problems In the dataset over the period analysed, 251 KKS-codes are present. Only 21 codes (8 %!) are responsible for 80 % of the EFOR in the dataset. This is graphically shown in figure 15.

figure 15

Components dominant to forced unavailability, VGB data

Failures at classical components such as HAD = evaporator, HAC = economiser, HAH = superheater and HAJ = reheat define in the VGB database for plants over 25 years of age about 50 % of the forced unavailability EFOR. These components are classical since they are also dominant with regard to FOR for younger plants. However, now other components are dominant that are interesting for life extension analysis, such as HLD = air heaters, HNC = induced draft (flue gas) fans, LAD = feedwater heating, LBA & LBB = steam piping, MAA & MAB = HP & IP turbine en MK(A) generator. A full list of dominant components with KKS-codes, number of failures per hr, average repair time and equivalent average repair time is kept at NRG offices.

911569/07.81244

29

The dominant components are components that should be investigated in any life extension plan. Also the pattern in time was investigated. When, for the majority of units, a peak occurs in failures (for instance at HAD = evaporator at 25 years of age), the timing of life extension activities could be optimised. Per power plant therefore the pattern in time was investigated. From that analysis, it can be concluded that there is no systematical pattern in the damages as a function of age. In figures 16 to 20 a number of examples are given. These examples should not be used to calculate the failure rate, since in the database also plants are present that do NOT have failures on these specific components.

figure 16

Pattern in time, HAD = evaporator

figure 17

Pattern in time, HNC = flue gas induced draft fan

30

911569/07.81244

figure 18

Pattern in time, LBA = steam piping

figure 19

Pattern in time, MAA = HP turbine, MAB = IP turbine

911569/07.81244

31

figure 20

Pattern in time, MK = generator, MKA = generator rotor/stator

The average number of failures per component per year and the average repair time are shown in figure 21 and 22. Both do not easily show trends. In the average repair time, incidents with large repair times are visible due to HILP failures. Only for evaporator problems (code: HAD, yellow in figure 21) there appears to be an increase between 25 years and 30 years of age. This is contrary to the NERC data, in which boiler unavailability for the 30-year-old plants is less than for 20-25 years old plants, while evaporators are dominant with regard to forced unavailability. The failure / damage characteristics of boilers as a function of age can be described as: a number of power plants older than 25 years operates without large problems at evaporators, superheaters and reheaters for some plants, a large number of failures occurs on specific components (for instance cyclone burners for "Schmelzfeuerung, a system that is no longer used in modern plants. Failures are solved with minimal repair. On average, a failure is not solved directly. The probability of reoccurrence is large.

32

911569/07.81244

figure 21

average no of damages per year, VGB data

figure 22

average repair time, VGB data

The occurrence of minimal repair at certain plants can easily be shown by examples such as figure 23. In both boilers of this so-called duoblock problems with the superheaters occur. Per occurrence, a limited number of pipes are replaced, the problem is not solved once-and-for-all. Also for Dutch power plants in their last years of operation such as the units A-C at Schiehaven, given their limited operation time, at about 1998 management opted consciously for minimal

911569/07.81244

33

repair instead of replacement of large superheater sections. The three units were demolished shortly afterwards.

figure 23

Minimal repair

In figures 24-27 examples of other failures are given. It is expected that failures on components HLD = air heater (bearing problems) and LAD preheater (pipes leaking) are a function of age. For some components (HNC = flue gas induced draft fan), problems are specific for certain plants. This also means that when identical units are operated, both failures and solutions can be expected and prepared for sister units when 1 of the plants has a failure.

34

911569/07.81244

figure 24

Example of HLD = air heater failures

figure 25

Example of HNC = flue gas induced fan failures

911569/07.81244

35

figure 26

Example of LAD = preheater failures

figure 27

Example of MKA Generator stator and rotor failures

36

911569/07.81244

4.3

Dominancy with regard to trips

From the VGB Nicht-Verfugbarkeits Statistik all data were analysed with code SMS1 = A (direct automatic trip) or B = (direct) manually out of operation. From the Pareto analyses, also for trips components such as HAD = evaporator, HAH = superheater, etc. can be shown to be dominant. For trips however, components related to control & instrumentation are present. A full list with KKS-codes, number of failures per hr, average repair time and equivalent average repair time is kept at NRG offices. When carrying out Life Extension studies, for dominant components one should generically investigate whether replacement is necessary. However, over the full operational life of a plant, C & I problems will be dominant with regard to trips. Only for a limited number of components, the frequency of trips will increase beyond 25 years of age, being C = process control, HFB = coal milling control, as shown in figure 28. Also the total number of C & I problems is on an increased level beyond 30 yrs. Both with VGB and NERC data, it can be shown that average repair time of control failures as a function of age increased Average repair time for VGB data is shown in figure 29.

figure 28

Pattern for trip frequency

911569/07.81244

37

figure 29

Repair time for C&I failures, VGB data

Application of failure data for Life Extension Studies

Power plant failures can be divided into normal failures without ageing, failures showing an ageing pattern (either increasing or decreasing) and High Impact Low Probability HILP failures. Normal failures without ageing can adequately be modeled using a Reliability Block Diagram technique. NRG uses the RBDA computer program for this purpose. An example of a block diagram is shown in figure 30. Systems (shown in green) are nested to subsystems and components. Components (shown in purple) are accompanied with a failure rate (/hr), and average repair time (hr) and a capacity. For instance, when when in figure 30 TLAC = turbine driven main feedwater pump fails, the parallel E-motor driven pump will allow the plant to carry on with 60 % output. Failure data are sourced from plant-specific information and data from generic databases. Bayesian updates can be used if plant-specific information is to be combined with generic information in order to have more accurate estimates.

38

911569/07.81244

figure 30

Example of a Reliability Block Diagram

911569/07.81244

39

The Block Diagram is used in a phased approach for Life Extension planning as a reference model that can be checked with plant specific data for the period when ageing is not present. The reference model is an auditable starting point for decision-making. An example is given in figure 31. The RBDA model is only valid for the period without ageing. In the data for these sister units, infant mortality as well as HILP failures can be discerned. Recent plant experience for these sister units showed ageing, for which Asset Management decisions were necessary. However, in the figure given, recent plant experience also included minimal maintenance and failures due to cycling. Therefore, ageing should be separated from these effects by analysis of the individual failures records and the operating regime.

figure 31

Use of a reference block diagram model

Ageing failures either can be deduced from trend-analysis of plant specific records or generic databases such as presented in the report. Trend-analysis is necessary anyway to separate infant mortality problems and random failures for correct failure rates as input in the RBDA model. A next step is optimization of replacement decisions for those components where ageing either is present or can be expected. In a recent analysis, these components were subtracted from the RBDA model and added as a time series in an Excel spreadsheet for optimization. The risk on High Impact Low Probability (HILP) failures should be estimated taking the condition of components into account. When the condition of the equipment, based on inspection and overhaul records, is less than adequate, the failure probability over the remaining plant life might even be high (HIHP= High Impact High Probability failures). This is easily estimated using a Delphi technique to estimate cumulative failure probability. An example is shown in figure 32.

40

911569/07.81244

During the Delphi, experts should be not in doubt on operating conditions, are not to confer on their estimate and may change their answers in a second round. If an expert stays an outlier, he is to state why. Clearly equipment no. 2 and 3 in figure 32 are candidates for replacement.

figure 32

Example of a Delphi exercise

HILP failures or HIHP failures are adequately modeled in a spreadsheet, allowing both for an analytical approximation as well as Monte Carlo analysis in order to calculate uncertainty bounds. Given their low probability, HILP failures can be optimized with f.e. strategic spare parts to lower the consequences. HIHP failures, given their higher probability, can be optimized by both replacement decisions and consequence control measures.

911569/07.81244

41

Expert judgment can also be used in combination with a Markov Decision model. This Decision model is based on a number of discrete system states. In practice, 4 states ranging from as good as new to failed are sufficient to describe degradation.

figure 33

Finite number of system states in a Markov Decision model

In figure 33, over a constant time step the probability is 0.75 that in 1 time step the component still is As Good As New (state 1, cell B2). The probability that in 1 time step the component will be appreciable degraded is 0.05 (cell D2). Similarly, given that the component has started to degrade (state 2, cell E3), the probability that the component over 1 time step the component has failed is 0.20. Probabilities like this can conveniently be estimated from maintenance records as shown in figure 34 or with the help of expert judgment. The example is for an electric motor bearing driving a gypsum pump, however exercises like this can also be carried out for steam turbines, boilers, etc.

42

911569/07.81244

figure 34

System states derivable from maintenance records

The Markov model with discrete system states, when simulated with Monte Carlo analysis, results in a probability distribution for the degradation as shown in figure 35. A next step is overhaul and/or inspection optimization. Stochastic optimization techniques allow varying the interval instead of using a fixed interval.

911569/07.81244

43

figure 35

Condition as a function of time by Markov modelling

Some optimization results are shown in figure 36. Comparison with figure 36 shows that the (genetic) optimization technique used tends to a first inspection about half way the 50 % failure time, followed by subsequent inspections closer to the average failure time while still in the degradation area, in order to replace the equipment without failure. It is recommended to use this technique for the inspection and overhaul of ageing equipment instead of the usual interval extensions when nothing is found during inspection.

figure 36

Inspection optimisation

44

911569/07.81244

Finally, the RAM data should be input to an economical Return on Investment ROI model as shown in figure 37.

figure 37

RAM data

911569/07.81244

45

Conclusions and recommendations

From data analysis on generic power plant databases and from plant specific failure information it is found that only for a limited number of power plant components one can expect an increase in failure rate or repair time. The increase in failure rate is generally called ageing, the increase in repair time may be due to obsolescence, spare parts that are difficult to acquire or failures with larger consequences (for instance generator failures). Also decreases as a function of age occur in the failure rate of specific components. On average, when maintained properly, the Forced Unavailability FOR does not increase. A type of roller coaster behaviour in FOR seems to exist between 20 years and 40 years of age with a large spread between individual units. A number of power plants older than 25 years has many problems with specific boiler components. However, failure records show that these are more related to the vintage design and to minimal maintenance than due to ageing. Components that are worth to investigate during formal life extension programs are: o Cold & hot reheat steam o Boiler controls o Specific boiler components (f.i. bottom ash) o Desuperheaters o Condenser (if no titanium is applied) o Flue gas fans, air heaters, duct expansion joints, dampers o Preheaters o LP en IP turbine o Turbine controls and valves While for most electrical component no ageing seems to occur, recent experience shows that generators and step-up transformers are suspect. If calculated indicators for remaining life are followed well, maintenance and inspection is adequate, and attention is given to some typical ageing items, an increased unavailability and trip frequency between 25 and 40 years is not to expect. When ageing is found, it should be investigated whether this is due to ageing or to minimal maintenance in combination with cycling operation. From unavailability data, it was found that regulation authority attention to emission aspects of the old plant might increase. Maintenance and inspection intervals for old power plants should not be simply extended when nothing is found during inspection. It could be optimum to go for increased inspection near expected failure time on the basis of measurements and/or life models. Inspection frequencies can be optimised using a Markov model. In order to assist in making replacement decisions in a formal Life Extension program, one should use reliability modelling for the plant before and after life extension to balance the costs for modification / renewal of components with gains in availability, efficiency, etc. The models

46

911569/07.81244

should incorporate normal failures, ageing and High Impact Low Probabilty (HILP) failures. Models should be validated with plant specific data. With limited amount of failure information, expert judgement is necessary. This can be based on a Delphi estimate of the failure probability of components.

911569/07.81244

47

Literature
[1] Nelson, A statistical Prediction Interval for Availability, IEEE Transactions in Reliability, vol. R-19, 1970

48

911569/07.81244

appendix A Unavailability of individual VGB plants


(the plots show examples of unavailability, some 20 plants were analysed)

911569/07.81244

49

50

911569/07.81244

911569/07.81244

51

52

911569/07.81244

911569/07.81244

53

You might also like