A variety of tools can aid in making decisions about the best time to repair, rehabilitate, or replace equipment. The newest tool – economic risk-based analysis – provides a way to optimize reliability while minimizing costs.
By Hans W. de Meel
There are many strategies for planning major maintenance and capital improvement at hydroelectric projects. Life-cycle planning is one such strategy. This planning method involves defining long-term expectations for equipment, then using these expectations to determine actions needed today. Life-cycle planning can be achieved using qualitative or quantitative assessments of future conditions. Most of the methods used in the past focused on qualitative evaluations, using primarily engineering judgment. However, today, statistical analysis can be performed to evaluate the risk of future equipment failure and, most importantly, the economics associated with that risk. It is the economic evaluation that allows determination of the optimum life extension measure, its optimum timing, and optimum equipment reliability.
Various planning methods used at hydro facilities over the past 40 years are:
– Maintain and repair as required (used before 1965);
– Condition assessment (used 1965 to 1985);
– Condition monitoring (used 1985 to today);
– Remaining life evaluation (used 1990 to today);
– Health indexing analysis (used 1995 to today); and
– Economic risk-based analysis (used 2000 to today).
The first five methods listed above are qualitative, and each has strengths and weaknesses.
Maintain and repair as required
This was the primary strategy before the mid-1960s, when most power plants were less than 50 years old. Considering the longevity of hydro equipment and structures, as well as the plentiful supply and low price of electricity, minimal maintenance was undertaken. Repairs were carried out as needed when breakdowns occurred. However, hydro project owners did not seriously plan for future major maintenance and capital expenditures.
In the late 1960s, maintenance costs began to increase as equipment aged and experienced associated wear-and-tear. Hydro project owners realized that equipment condition was an important factor in unit availability.
Initially, hydro project owners attempted to classify equipment through a priority allocations process. Using this process, owners defined categories related to whether the equipment was likely to fail or need maintenance, and assigned priorities. Maintenance efforts were then directed toward the highest priorities. The need to make provision for future expenditures became more important as project owners focused more on capital budgets. However, these estimates were based exclusively on engineering judgment and experience.
Equipment instrumentation systems allow for monitoring of performance and deterioration, as well as for advanced warning of impending problems. Condition monitoring is an effective method that assists in the planning of maintenance and in defining the condition of equipment components.
Condition monitoring permits extrapolation of monitoring data for short-term maintenance and/or repair planning. However, if used to predict reliable mid- and long-term maintenance and capital requirements, extrapolation by definition has its limits. Another shortcoming of this method is that planning for life-cycle extension is not primarily a technical issue. It is driven by the risk-cost of failures; the economics of major repair, rehabilitation, or replacement programs; and reliability considerations.
Although it remains difficult to determine which quantities to monitor and how to monitor the equipment, condition monitoring systems are valuable for improving maintenance. However, for life-cycle planning, their main use is limited to providing the base input for condition assessments.
Remaining life evaluation
The remaining life method was applied in the early 1990s to get a better estimate of future requirements for major maintenance, equipment rehabilitation, and replacement. The method, based on industry experience regarding normal expected service life, determines remaining life based on the results of a representative age. This representative age takes into account past performance, behavior, and condition.
Figure 1 on page 56 demonstrates the remaining life concept and is based on an industry survival curve prepared by Hatch Energy that contains information from several hundred transformer retirements. The curves express the variation of average remaining life with equipment age. In this example, at age 38 years (on the left curve in Figure 1) the survival percentage is 60 and the probable end of life (on the right curve) is about 58 years. This gives a remaining life of about 20 years. However, at age 58 years, with a corresponding survival percentage of about 39, the probable remaining life is still about five years. Therefore, as equipment survives, the remaining life also increases.
Concerns about this method relate to the meaning and use of the terms “normal expected service life” and “remaining life.” Normal expected service life is not a fixed parameter. Rather, it is a statistical definition that has a mean value and a standard deviation. As such, remaining life is not a constant; it varies with the age of the equipment.
A second issue is that decisions to replace or rehabilitate a piece of equipment are not just a function of its age or remaining life. They also are a function of the consequence cost incurred when the equipment fails. Consequence cost is the sum of the replacement cost and downtime cost (lost generation). The probability of failure multiplied by the consequence cost is the risk-cost associated with equipment failure. The higher the risk-cost, the earlier the equipment should be replaced. The optimum reliability of a piece of equipment depends on its exposure to risk-costs. The remaining life concept does not take risks into account.
A third issue is that the remaining life method does not indicate the type of measure that should be implemented (maintenance only, rehabilitation, replacement, etc.), its implementation timing, or how its reliability can be optimized for the least overall cost.
Health indexing combines the benefits of condition assessment with that of remaining life evaluations to develop a rating (index) system for the relative health of a piece of equipment. This rating system can then be used to set a standard for future maintenance needs.
Shortcomings of this method are the same as those for remaining life. However, the method is most usable when a large number of the same components are analyzed, such as electrical equipment in a transmission/distribution system (transformers, breakers, etc.) and varying health classes are introduced to prioritize maintenance and rehabilitation requirements.
Quantitative method: economic risk-based analysis
Economic risk-based analysis was developed in the mid-1990s to overcome concerns associated with the condition monitoring, remaining life, and health indexing methods. The method was developed separately by the U.S. Army Corps of Engineers and Acres International. This method has been applied at about 80 power stations, involving hundreds of pieces of equipment.
The method uses statistically defined failure-probability (survival or retirement) curves derived from large samples of industry-recorded failures and retirements. It then introduces evaluated condition assessments (providing a representative age instead of a calendar age) as input parameters for risk calculations.
The method recognizes that remaining life is highly dependent on:
– Number and type of failures the equipment might be exposed to during its future life;
– Selected repair methods;
– Extent of renewal that may occur from failure repairs; and
– Contribution of each of these failures to the ultimate retirement of the equipment.
Once this (existing) failure environment is uniquely defined, the method examines the economic effect of various remedial measures to extend equipment life. This examination also includes sensitive analyses on varying risk-costs, maintenance costs, capital costs of interventions, and potential performance improvement benefits.
The result is a comparison of the various alternatives and a definition of the optimum option for life extension, its timing of implementation, and its cost, with a view to optimize reliability of the equipment and plant and to minimize costs. Economic risk-based analysis also allows examination of the failure interaction between different equipment, as well as the scheduling of multiple equipment intervention measures.
Setting failure-probability curves
Reasonable event forecasts can be made using historic data. For example, the use of past hydrology to predict future electricity generation is well accepted. Such an approach is no less valuable for assessing the future behavior of equipment components. Failure-probability (f-p) curves reflect such historically recorded failure behavior patterns.
The premise behind the use of generic f-p curves is that it is nearly impossible to capture all of the unique and specific failure characteristics of equipment systems or components for the various manufacturers, materials used, designs, operating environments, maintenance conditions, etc. It would require a massive database to derive such separate classes of f-p curves, and failure data simply is not available at that scale. However, it is possible to develop generic industry average f-p curves for each equipment component. The task at hand, then, is to find a mechanism to introduce the uniqueness of each individual component into the generic f-p curves.
This approach may be better understood by considering a theoretical “human f-p curve.” The best way to assess individual life expectancy is by comparing ourselves with average statistics of mean human life. We are all unique, but we know that our mean age (at least in the western world) is about 80 years. We then would introduce our unique characteristics into the human f-p curve by considering life style, heredity, medical history, quality of medical attention, living environment, etc. The same customized approach is valid for equipment components.
The economic risk-based method adopts f-p curves based on industry experience. The uniqueness of individual components is introduced as follows:
– A representative age is used, based on a detailed condition assessment and using the REMR (Repair, Evaluation, Maintenance and Rehabilitation by the Corps) research program, health indexing, or similar methods. This representative age reflects past operating conditions, past maintenance, manufacturers, materials, etc., relative to calendar age.
– A decision tree repair approach is incorporated, in which the user specifies an unlimited number of minor and major failure scenarios, each with their degree of effect on future life-cycle expectation (retirement). It allows definition of the frequency of minor and major failures, and permits both the method and extent of repair to be selected. The latter will affect future behavior in different ways. Finally, it recognizes that, after each repair, some extent of equipment rejuvenation will occur and can be defined. This approach permits a degree of aggressiveness in maintenance and repairs to be uniquely introduced.
– The mean and standard deviation of the f-p curves are varied to reflect past and future operating conditions.
– The effect of many different intervention modes to extend life are introduced and examined, so that economic return and reliability can be optimized.
Customizing the f-p curve
A typical example of customization of the generic f-p curve is a study carried out in 2003 and 2004 for a 650-MW station in Québec, Canada, that contains 14 turbine-generating units. The station had been converted from mainly baseload to pure peaking operation. The owner wanted to determine the effect of increased wear and tear on the asphalt-insulated stator windings.
For this scenario, Hatch Energy converted the age-based f-p curve to an hourly-based curve. The effect of starts and stops on the aging process was evaluated using the VDGW (German Association of Power Producers) criterion that each start and stop will age the winding equivalent to ten hours of operation. Past operating hours were reviewed as a percent of mean life-span operating hours, and the number of anticipated starts and stops per year as a result of the future peaking operation were determined. The f-p curve was then adjusted, as shown in Figure 2 on page 58.
The customized f-p curve was then used for economic risk analysis after equipment condition and failure/repair strategies were introduced through the decision tree feature of the model. The optimum timings of rewinds for all 14 stators were determined, permitting development of a rewind implementation schedule for the entire plant.
Accuracy of f-p curves
Discussions regarding the various methods for life cycle analysis naturally focus on the accuracy of the results. Foremost, attempting to look into the future is not an accurate science. To do so using only engineering judgment or not using well-defined end-of-life criteria overlooks the economics of optimizing intervention timing, the type of intervention, or reliability. The economic risk-based method attempts to quantify both the uncertainties associated with engineering judgment of future equipment behavior and the optimization of reliability at minimum costs.
F-p curves are only one part of the risk-based economic life-cycle analysis. Other parameters that play a key role are:
– Condition of the equipment (representative age);
– Value of lost revenue from downtime;
– Type, extent, and cost of the repair method;
– Frequency of failure;
– Extent of renewal introduced by the repair; and
– Economic discount rate.
The question arises about the importance of these parameters relative to the accuracy of the f-p curves. In other words, is there a single parameter in the method, such as the f-p curves, that drives the solution?
The following example illustrates the answer to this question for one facility. An 80-year old two-unit station in Ontario, Canada, still had its original transformers. Technical experts agreed the transformers had reached end-of-life and failure was imminent. This conclusion was supported by industry failure curves. However, closer examination using the risk-based economic method revealed that, in case of failure, all power flow could be diverted to a nearby parallel station. Thus, while the probability of failure was high, consequence cost from lost revenue could be avoided. Therefore, there was no risk if the transformers failed.
Despite overwhelming technical evidence to the contrary, the optimum recommendation was a do-nothing scenario. In this case, it was not the f-p curves or their accuracy, but the downtime cost that drove the solution. The transformers functioned for another three years before they were replaced.
Extensive use of the economic risk-based method over the past ten years has involved many sensitivity analyses. It has become apparent that the accuracy of the f-p curves is no more dominant than any of the other parameters of the analysis. And therein lies the answer to the questions about both the generic nature and the associated accuracy of the f-p curves.
An analogy may again be found in human behavior. It is how healthy we live (operating environment) and the quality of medical care available to us (maintenance/repair practice) that greatly influences our life expectancy, rather than just average human life expectancy statistics.
Traditional life cycle planning methods do not consider the probabilistic nature of future equipment failures, nor the economics associated with the timing of life extension measures. A quantitative, as opposed to qualitative, approach that takes these factors into account is available through risk-based economic analysis.
This unique approach uses recorded industry-average f-p curves and customizes the risk environment of specific equipment components using a decision-tree approach. This method also considers maintenance costs before and after implementation of life extension measures, capital costs, downtimes, and performance benefits associated with these measures. The method is embedded in the HydroVantage model, which has been used for life cycle planning of about 80 power stations.
Mr. de Meel may be reached at Hatch Acres, 6 Nickerson Street, Suite 101, Seattle, WA 98109; (1) 206-352-5730; E-mail: firstname.lastname@example.org.
Sturtevant, Mark A., Darin E. Johnson, and Scott E. Anschell, “A Process for Prioritizing Upgrades and Life Extensions at PacifiCorp,” Hydro Review, Volume 25, No. 7, November 2006, pages 30-37.
Welt, F., M. Morgenroth, and Hans W. de Meel, “Optimization of Equipment Capital Investment Planning for Hydro Plant Rehabilitation,” Waterpower XIII Technical Papers CD-Rom, HCI Publications, Kansas City, Mo., 2003.
Welt, F., and Hans W. de Meel, “Analysis of Failure Risk Interdependencies between Hydro Equipment Components,” HydroVision 2004 Technical Papers CD-Rom, HCI Publications, Kansas City, Mo., 2004.
Westermann, G., K. Bhan, and Hans W. de Meel, “Generator Rewinds: A Model to Predict Optimum Timing,” HydroVision 2000 Technical Papers CD-Rom, HCI Publications, Kansas City, Mo., 2000.
Westermann, G., and H. Zhu, “Timing for Generator Rewinds: Bridging the Gap between Statistical Methods and Condition-based Monitoring,” Waterpower XII Technical Paper CD-Rom, HCI Publications, Kansas City, Mo., 2001.
Hans de Meel, P.Eng., is manager of operations in the western U.S. for Hatch Acres Corporation (formerly Acres International). He was instrumental in developing the economic risk-based method discussed in this article.
µ Peer Reviewed
This article has been evaluated and edited in accordance with reviews conducted by two or more professionals who have relevant expertise. These peer reviewers judge manuscripts for technical accuracy, usefulness, and overall importance within the hydroelectric industry.