March 2000

SIM-00-001

Financially Based Inspection and Repair Planning

Maintenance engineers are finding themselves competing more and more for financial resources. Until recently, sound engineering analysis alone was adequate to convince management that maintenance expenditures were necessary. However, corporate resources in many industries have tightened, and engineering analysis alone can no longer justify maintenance resources. Increasingly, the control of maintenance resources in many industries is in the hands of financial decision-makers using quantitative financial and decision analyses. Traditionally, engineers are ill equipped to convincingly tell their story in these terms. Structural Integrity (SI) offers a process that aids in this communication and guides maintenance resources into financially viable repair/replacement strategies. Equipment Aging The pressure on maintenance budgets comes from equipment that is in the aging period of its life cycle. A typical life cycle curve, Figure 1, shows
20
Unit Forced Outage Rate, %

in manufacture or design. The center section is a period when minimum prudent maintenance keeps the failure rate relatively constant. The right section is the aging phase and has an exponentially increasing failure rate because of the dominance of age-related failure mechanisms. This curve is classic and predictable. In most industries, the original plan was to replace equipment at the end of its 30 to 50 year life cycle. The advent of improved worldwide transportation, reduced single country market domination and multi-national corporate ownership have resulted in extraordinary downward pressure on unit product pricing. This competitive pressure is limiting maintenance resources at the time in equipment life cycles when failure rates are increasing. As a consequence, many corporations are faced with the challenge of keeping aged equipment operating with limited maintenance resources. The limitation on maintenance resources due to lower prices, and the increased need for maintenance resources because of aging, has placed maintenance decision-makers in a dilemma. Quantitative financial and decision analyses techniques are now needed to properly select maintenance actions. Basic Decision/Financial Model

15 10 5 0 0 5 10 15 20 25 30 35 40
Unit Age (Years)

Fig. 1 NERC-GADS Data for Forced Outage Rate for Coat Fired Units from 50MW to 200MW.

the classic shape of the Weibull life cycle curve.1,2 To the left is the decreasing failure rate of the curve, usually called infant mortality. It is dominated by failures from initial defects, either

Decision analysis, supported with financial analysis and optimization techniques, is routinely used for large investment decisions. This decision-making method was developed almost 30 years ago3 to account for conditions of uncertainty. It is well suited for maintenance decision-making because of the uncertainties in maintaining equipment during its aging phase.

Many companies are already addressing financial decisions on a fully quantitative basis using classical methods. It is reasonable, therefore, to use the same techniques for maintenance decisions since maintenance is competing for the same resources. Decision Criteria In decision analysis, the first step is to select the criterion to be optimized. Textbooks used in MBA corporate finance courses typically recommend Net Present Value (NPV).4 This criterion insures there is maximum return for the invested corporate dollar when decisions have impact over multiple years. This makes NPV the most robust of the financial criteria used for optimization. In the case of maintenance, the Net of Net Present Value can be defined by addressing a choice between two fundamental maintenance options - one to do nothing, i.e. run the equipment as is, performing only routine maintenance, and the other to take a mitigating (non-routine) maintenance action.5,6 A diagrammatic representation of this decision is shown in Figure 2. The decision criterion to be
Consequential Cost of Failure W ithout Action
Benefit Cost

difference between the Benefit and the Cost is the Net of Net Present Value. The Present Value of Net Present Value considers the effect of taxes, time value of money, and required rate of return on investments. In taxed corporations, the tax effects on maintenance expenditures, as well as losses declared from shutdowns or deratings, are significant and must be considered. Because maintenance decisions on aging equipment include timing, the time value of money is an important consideration in any maintenance decision analysis. And finally, the expected return from the invested maintenance dollar must meet or exceed the minimum desired return to the shareholder over time. The Link Between Engineering and Finance Perhaps one of the most important steps in the fully quantitative decision analysis process is the conversion of engineering analysis results into financial consequences. The engineer usually speaks in engineering units, such as time to failure or probability of failure. He must quantify not only the probability of the occurrence, but the consequence of the occurrence as well. Financial decision-makers are not concerned with whether or not a component will fail (i.e. the occurrence). They are concerned only with what the failure will mean to the financial health of the company (i.e. the consequence). A widely used term for this is risk.6 Risk is defined as the product of probability of occurrence and consequence of occurrence. In the financial community, this same formulation is termed the expected value of the consequence.5 Because of this similarity in definitions, we now have a link between the engineering world and the financial world. Risk or expected value of consequence of failure is key to the formulation of a model for fully quantitative maintenance decision making. This key can be illustrated by enhancement of the Figure 2 Maintenance Action Decision Influence Diagram. Note in Figure 3 the addition of the Probability of Failure, Year to Perform Maintenance Action, and Consequential Cost of Failure nodes, and conversion of the Consequential Cost of Failure nodes into the Expected Consequential Cost of Failure nodes. The with and without maintenance

Decision

Outcome

No Maintenan ce Action vs Perform M aintenance Action

Net Present Value

Consequential Cost of Failure W ith Action Cost of M aintenance Action

Fig. 2 Maintenance Action Decision Influence Diagram

optimized is shown at the right of the figure. The consequential cost of failure without the mitigating maintenance action is the potential Benefit of the decision, i.e. the consequences avoided by taking action. The cost of the mitigating maintenance action plus the consequential cost of failure with the action (less drastic, or later, than the failure without action) is the potential Cost of the decision, i.e. the consequences incurred by taking action. The

Probability of Failure W ithout A ction

D ecision

Expected Consequential Cost of Failure W ithout A ction

Ou tcom e

No Maintenance Action vs Perform M aintenance Action

U ncertainty

Consequential Cost of Failure

Net Present Value

Probability of Failure W ith A ction Year to Perform M aintenance Action Decision

Expected Consequential Cost of Failure W ith A ction

determine a maintenance action year that maximizes NPV, while not exceeding constraints such as maintenance budget limits or forced outage rate limits. In other words, the goal is to optimize NPV by maintenance action timing.5,6 SI utilizes an MS Excel workbook version of this model, with an operations research type optimizer incorporated as a visual basic macro.5,6,7 Financial Inputs The financial assumptions used in this type of analysis are usually well established in the financial organization of a corporation, and available to the engineer. These assumptions must be included because they have significant effect on the impact of maintenance action timing. The first set of financial assumptions affect the time value of money and include the projected inflation rate, to account for the eroding value of the dollar with time, and the corporate discount rate used to set required return on investment. The second set of financial assumptions has to do with the composite income tax and the property tax rates. Of these two, the composite income tax rate is the most significant and has a large effect on a maintenance actions benefit. Financial loss consequences can be viewed as a business loss, and a significant portion can be written off. The impact of the maintenance cost itself can be greatly reduced for the same reason, if the maintenance cost is classified in an expense category. If the maintenance cost is capitalized, it still can be written off, but at a lower rate, over time. Engineering Inputs A prudent, fully quantitative engineering probabilistic analysis or assessment, and appropriately justified financial failure consequences, lend added credibility to risk or expected value of the consequence estimates. It is prudent to use all possible technologies in assessing equipment condition. Probabilistic fracture mechanics (PFM) is such a technology. PFM establishes a damage propagation mechanism mathematical model with a failure criterion. This model is then integrated over the range of input distributions with a

Cost of M aintenance Action

Fig. 3 Maintenance Action Decision Influence Diagram with Expected Consequential Cost or Risk.

action cash flow streams create the Net value of the decision criterion. The arrows pointing from Year to Perform Maintenance Action to Probability of Failure Without Action, and from Probability of Failure Without Action to Expected Consequential Cost of Failure With Action require explanation. These arrows establish the relationship for the expected consequence of failure if the maintenance action is delayed. During the delay to the year in which maintenance action is performed, the expected consequence of failure will be the same as if the action had not been taken. After the year to perform the maintenance action, the probability of failure with action is used to determine the expected consequential cost of failure with the action. Probability of Failure with Time The use of probability of failure with time is an all-inclusive expression of the engineering condition of the equipment. For many years, worst case analyses were performed, with probability of failure set to an absolute certainty, the value of one. The advent of sophisticated inspection techniques and probabilistic fracture makes it possible to express the complete engineering state of equipment in terms of a probability of failure versus time curve. It is clear from the discussions above that an expression of equipment condition in this form directly integrates into established classical probabilistic, decision, and financial analysis methods. Optimizing the Decision/Financial Model The objective in optimizing the model is to

Monte Carlo program. This integration produces a probability of failure verses time curve.5,6 Less rigorous, but fully quantitative, methods are available for obtaining probability of failure versus time. These methods are useful in obtaining preliminary engineering inputs for the maintenance decision analysis allowing a first level look at possible allocation of resources. One such method is trending failure history data using standard linear regression analysis and the Weibull distribution function. This function was derived to express equipment reliability as a function of time. Another method is structured interviews with plant personnel regarding potential future equipment failures. This method is referred to as probabilistic assessment interviewing or expert opinion elicitation. It was developed by cognitive psychologists in the decision analysis community over the last 15 years, and has been extensively used with senior corporate executives for large corporate decisions.5,6,7 The Bayesian-like update analysis is a classical method of combining failure history projections with plant personnel structured interviews. The Bayesian method, developed several hundred years ago,8 combines current probabilistic information of an event with newly found information for the same event, resulting in updated information.9 This technique provides better projections of the future failure probability because of the use of both data sources. In addition, these techniques allow the decision to be assessed inexpensively as a first iteration to assess the need for a more expensive, fully quantitative engineering analysis. Combustion Turbine Blade Refurbishment Example The Situation Because of the high capital cost, it is prudent to delay blade refurbishment or replacement until a turbine wreck is a real possibility, but not a real threat. Replacing too soon will result in a large expense when there is no real threat. Replacing too late could result in a very costly turbine wreck and forced outage.

The Model Measurement of blade coating and temperature state, coupled with a remaining life engineering model, provides the projected probability of failure. The decision analysis model, Figure 4,
D ecisio n

Ye ar of A c tio n

Te m p e ra tu r e M e a s u re m e n t/ In s p e c tio n C ost

C onstraint

B u d ge t L im it

D ecisio n

R e fu rb is h m e n t C ost

M a in te n e n c e C ost

R un vs In sp ect-R efu rbish m en t/ In sp ect-R ep lac e B la d es C oa tin g D eg rad ation M eas ure m en t

B la d e R ep lac em en t C os t

O utcom e

C oa tin g D eg rad ation M ode l

Te m p era ture M eas ure m en t

U ncertainty

N et P res ent Value Tu rbin e W re ck C os t F orc ed O utage C os t

R em aining L ife M ode l

E s tim a te d P rob ab ility o f Fa ilu re

F orc ed O utage H ou rs

F orc ed O u ta ge R ate Lim it

C onstraint

Fig. 4 Decision Model for Timing Refurbishment or Repair of Gas Turbine Blades

serves as a bridge between engineering analysis and financial analysis. The model illustrates use of measurement data, life prediction analysis, turbine wreck costs, and forced outage costs. The model also includes the competing refurbishment and replacement costs that must be held within the operators budget limit. The workbook constructed for this model is exercised not only for each candidate year of action, but also for each candidate type of action. The goal is to find the action/year combination that maximizes NPV and remains within specified constraints. Turbine Overhaul Timing Decision Example This example deals with the delay of a turbine overhaul beyond the OEM recommendation. The objective is to determine the optimum timing for the overhaul. The model for this decision, Figure 5, is constructed in the same manner as above. In this case, however, we are dealing only with the timing of a maintenance decision. The upper half of the diagram represents the benefits from timing the overhaul to avoid the consequence of a shutdown and its associated damage. The overhaul is estimated to be $350,000. If there were a failure in the hot section, the down time

Probability of Failure W ith ou t O v erh au l

D ecisio n

Exp ected C o ns equ ential C o st of F ailu re W ith ou t O v erh au l

provides an indication of the sensitivity of value with time. It particularly highlights the impact on the corporation of delaying the overhaul too long.
O u tcom e

N o O verhau l vs Perfo rm O v erh au l

C o ns equ ential C o st of Failure

N et Pres ent Valu e

Boiler Tube Component Replacement Example This example addresses a boiler tube component of the superheater or reheater type. The concern is if, and when, to spend $1,000,000 for replacement of the entire component. The failure risk in this example is the burst of any tube segment, resulting in the shutdown of the entire power unit. It is assumed that the unit will be down for 24 hours for repair of each tube burst. The unit has a 150MW maximum dependable capacity, with a projected capacity factor increasing from 25% to 55%, with replacement power cost increasing from 20$/MWH to 55$/ MWH over the next twenty years. The probability of failure versus time is based on a remaining life analysis, shown in Figure 8
Boiler Tube Probability of Failure vs Time
Cumulative Probability
45.00% 40.00% 35.00% 30.00% 25.00% 20.00% 15.00% 10.00% 5.00% 0.00%

U ncertain ty

Probability of Failure W ith O verhau l Year to Perfo rm O v erh au l

D ecisio n

Exp ected C o ns equ ential C o st of Failure W ith O verhau l

C o st of M ainten ance O v erh au l

Fig. 5 Decision Model for Timing of an Overhaul

costs are expected to be $1,000,000 for repairs and 500 hours of lost revenue (due to the shutdown) at $2,000 per hour, weighted by a utilization factor of 60%. The probability of hot section failure is represented by the cumulative probability of failure versus time curve in Figure 6. The Net Present Value versus Maintenance
0.60

Probability of Failure

0.50 0.40 0.30 0.20 0.10 0.00

19 97 19 99 20 01 20 03 20 05 20 07 20 09 20 11 20 13 20 15

Base Case Alternative Case

Years

Figure 6 Gas Turbine Hot Section Component Probability of Failure Vs. Time

96

98

00

02

04

06

08

10

12 20

19

19

20

20

20

20

20

20

Year

Action Year curve is shown Figure 7 with a peak in 1998 of $79,495. This is the optimum time to
$100,000 $50,000 $0 ($50,000) ($100,000) ($150,000) ($200,000) ($250,000) ($300,000) ($350,000) ($400,000) ($450,000) ($500,000)

Fig. 8 Boiler Tube Probability of Failure Vs. Time.

Year

Figure 7 Net Present Value Vs. Maintenance Action Year

perform the overhaul. The curve is fairly flat for another year, indicating that the overhaul could be delayed until 1999 with essentially the same positive value result. From 1999, however, the curve drastically decreases with a negative net value being created after 2001. This curve

as Base Case. This is the expected probability of tube burst versus time curve, given no replacement and given that the unit continues to operate, causing the next less efficient unit in the dispatch to be operated upon tube burst. The Alternative Case curve assumes that the boiler tube component is replaced. Here, the probability of failure curve is lowered due to the replacement action. NPV is generated by decreasing the probability of failure by replacement of the boiler tube component. Note from Figure 9 that the highest NPV is created by replacement of the component in 1998. Note, also, that the highest NPV is positive, which means that the boiler tube component should be replaced, creating the highest value, if performed in 1998.

Net Present Value

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

20

14

Net Present Value versus Maintenance Action Year for Boiler Tube Replacement Example
$400,000 $200,000 $0 ($200,000) ($400,000) ($600,000)

estimated NPV improvement resulting from the optimization was in the tens of millions of dollars. Conclusions

Net Present Value, $

1996

1998

2000

2002

2004

2006

2008

2010

2012

2014

Year

A fully quantitative maintenance decision analysis model can: 1) Be optimized for the timing of the maintenance actions, maximizing the value of the maintenance dollar to the corporation; 2) Create a link between engineering and the financial decision maker by the common definition of risk and expected value of consequence of failure; and 3) Allow engineering to compete for corporate resources during times of tight budgets. For more information regarding financially based inspection and repair planning or related SI capabilities, please contact:

Fig. 9 Net Present Value Vs. Maintenance Action Year

Concept of Multiple Maintenance Decisions In reality, numerous maintenance projects are competing for the maintenance budget simultaneously. Financial risk optimization is the performance of a decision analysis process that maximizes NPV for many projects simultaneously. In a utility, all projects may be annually constrained by the maintenance budget limit, which means that financial risk optimization must be conducted with all individual decision models linked.5,6,7 Financial risk optimization has several benefits. The more maintenance projects there are, and the larger the portion of the maintenance budget that is modeled, the larger the expected value benefit that can be realized. In addition, decision analyses are performed using a systematic, fully quantitative and reviewable process. For optimization of multiple projects, the decision models previously described would be expanded. For each additional project, the additional probabilities of failure and years of maintenance action nodes would be added to the influence diagram and the spreadsheet. The multiple projects optimization then provides a prioritized list of decision alternatives and years of maintenance action that creates the maximum NPV for the integrated group, while staying within specified constraints. This is accomplished in the spreadsheet by inserting different combinations of years of action seeking that combination that maximizes total NPV. This is accomplished through an operations research algorithm approach. An example of the benefit of this approach was demonstrated with a PC-based workbook model applied to 140 forced outage critical fossil plant tube components across a power system.5 The

Dave Mauney dmauney@structint.com

6110 Executive Boulevard, Suite 1080 Rockville, MD 20852 Phone: 301-231-7746 Fax: 301-231-7752 www.structint.com

References

1.

Heiges, H. H. and Stoll, H. G., Power Plant and Turbine-Generator Upgrading Economics, Proceedings: Fossil Plant Life Extension Conference and Workshop, Electric Power Research Institute, Palo Alto, CA, August 1985, p. 12-3.

2.

Miller, I. and Freund, J. E., Probability and Statistics for Engineers, Prentice-Hall, Englewood Cliffs, NJ, 1965, pp. 365-367 and 379-380.

3.

Raiffa, H., Decision Analysis, Addison-Wesley, Reading, MA, 1968.

4.

Brealey, A. and Myers, S. L., Principles of Corporate Finance, McGraw-Hill, 4th Ed, 1991.

5.

D.A. Mauney, Economic Optimization of Multiple Component Replacement/Inspection in the Power System Environment, ASME Publication, PVP Vol. 251, Reliability and Risk in Pressure Vessels and Piping, Ed., J. H. Phillips, July 1993, pp. 1-16.

6.

Fossil Fuel Fired Electric Power Generating Station Applications, Risk-Based Inspection-Development of Guidelines, ASME Research Report, CRTD, Vol. 20-3, ASME, New York, 1994.

7.

ASME Application Handbook for Risk-Based Methods For Equipment Life Management, ASME Research Report, CRTD-41 (to be published 1999), ASME, New York.

8.

Bayes, T., An Essay Towards Solving a Problem in the Doctrine of Chance, Philosophical Transactions of the Royal Society, 53, 1763, pp. 370-418. Reproduced with biography of Bayes in G. A. Barnard, Studies in the History of Probability and Statistics: IX, Biometrika, 45, 1958, pp. 293-315.

9.

Schmitt, S. A., Measuring Uncertainty: An Elementary Introduction to Bayesian Statistics, AddisonWesley, Reading, MA.