Executive Summary

A study on the applicability of multi-gas monitors was made on an important population of Southern California Edison’s transformers commonly referred to as the “A Bank Population”. This population represented both a significant financial and functional investment. While the Edison transmission and sub-transmission system is designed to handle a single transformer failure with little or no customer or system impact, multiple failures as experienced recently do represent an operational threat. With the average age of this “A Bank” fleet approaching 40 years there is a realistic possibility that multiple or cascading “A Bank” failure events will take place again.

In order to reduce the risk associated with an “A Bank” failure and ensure maximum operating life is obtained from the fleet, the implementation of a multi-gas on-line monitoring system has been shown to be both technically and economically prudent.

Figure 1 Aging "A Bank" Transformer Fleet

Beyond the risk of failure, there is a need to consider pre-emptive replacement of older units but unfortunately there is not a clearly identifiable age at which transformer retirement should take place. While it is expected that the retirement of older units will improve overall fleet reliability and reduce risk, it comes with a significant capital cost.

Current PM programs and operation practices have provided Edison with excellent results over the years, kept the failure rate of the “A Bank” fleet at a very low level and resulted in long transformer operating lives. These maintenance programs and operation practices, as they stand now, are optimized and will not alone address the reliability concerns of an aging fleet. The implementation of continuous on-line monitoring of dissolve gases in transformers has been technically proven as a way to identify incipient insulation failures and provide the Maintenance and Operations Departments with adequate time to respond before a functional failure occurs. Until now, this approach has not been economically justified nor proven to reduce overall risk of failure appreciably at Edison.

The business case that follows conclusively shows that the implementation of continuous multi-gas on-line monitoring on “A Bank” transformers is of significant benefit to Edison. The key benefits this technology provides are:

· Economic: Greatly reduces the impacts of failure. A payback period of less than 5 years is expected for each multi-gas monitor installation (see Table 1).

· Risk Reduction: Significantly reduces the overall risk of transformer failure even when all modes of failure and current maintenance practices are included. A 39% reduction in the overall failure risk for the existing fleet is anticipated (see Figure 2)

· Preservation of Capital: Allows Edison to extract nearly all of the operating life from a transformer without having to experience a failure. A one year of addition operating life more than pays for on-line monitoring investment.

Figure 2 "A Bank" Risk Profiles for Various PM Approaches

2 BUSINESS Opportunity

Power transformers represent a significant capital investment by Southern California Edison. Replacement of these devices occurs for three general reasons:

· Technical inadequacy such as capacity or voltage

· Failure of the transformer or one of its key subcomponents

· Risk of imminent failure

Current PM Practices significantly reduce the likelihood of a transformer failure by correcting deficiencies, replacing worn components and determining the transformer’s operating health. As the operating health of the transformer diminishes there is now a heightened awareness of imminent failure. For many critical transformer components, there is no economically effective method for correcting these health deficiencies thus continuing to operate the transformers adds an increased burden of risk. If the health deficiencies indicate a quickly deteriorating condition, immediate response is required, if the health deficiencies indicate a slowly deteriorating condition, a delayed response maybe appropriate. Unfortunately, it can be very difficult to determine the rate of deterioration, especially if the health analysis is based on data gathered at periodic intervals of low frequency. To reduce this burden of risk created by the uncertainty of when an incipient failure may become catastrophic, retirement and replacement of the transformer is many times the most prudent response. The resulting irony is that while maintenance reduces the probability of certain modes of failure, it heightens the risk awareness for other modes of failure.

The PM Practices employed by Edison, being periodic by design, have been optimized to return the highest reliability for the invested maintenance dollar. In order to further reduce the number of catastrophic failures, increase transformer operating life and minimize risk, current PM practices must be augmented with continuous condition monitoring data. The opportunities presented by continuous condition monitoring are:

· Improved reliability-reduction in catastrophic failures

· Extended operating life-longer return on initial capital investment

· Reduced risk

· An ability to overload the transformer without significant loss of life

3 Objective of this Study

The objective of this study is to determine if on-line, continuous measurement of dissolved gasses in transformer oil is an effective method of reducing in-service failures of large power transformers, extending their in-service life and significantly reducing overall transformer operating risk. This study focuses on maintenance, financial and risk issues; it assumes that the on-line monitoring system being employed is accurate and reliable and that all other forms of maintenance are performed in a responsible manner. The study will also determined if the expense of installing and operating such monitoring systems provides significant benefit compared to traditional alternate forms of diagnostic testing and maintenance.

In order to meet this objective, the study will examine:

· Critical functions performed by the transformer

· Common modes and causes of transformer failure

· Failure mechanisms

· Industry failure statistics

· Edison failure statistics

· Costs

· Risks

· Data management

5 Critical Transformer Systems and Functions

A transformer is considered not to be an assembly of electrical and mechanical components but rather the integration of a number of specialized functional systems. Each of these systems performs a unique and important function having its own unique modes of failure. These critical functions must be preserved in order to prevent both major and minor failures. Important subsystems and their functions will be described in the following subsections.

Dielectric System:

The dielectric system provides electrical isolation between windings, phases and ground planes, included are all major and minor insulation elements found in the power transformer. The insulation system must be capable of withstanding specified operational electric stresses, considering a permissible level of overloads. The elements of a dielectric system include:

· Paper insulation used in windings

· Solid insulation used for blocking

· Lead insulation

· Insulating oil

· Electrostatic shields

Electromagnetic Circuit:

The electromagnetic circuit includes those magnetic elements that create, contain and couple the magnetic flux. Elements include:

· Core

· Windings

· Magnetic shields

· Grounding circuit

Current Carrying Circuit:

The current carrying circuit includes all conducting elements that carry load current. Included are:

· Winding leads

· Winding conductors

Mechanical System:

The mechanical system provides structural support of the current carrying and magnetic circuits.

· Clamping

· Lead support

Voltage Regulating System:

The voltage regulating system includes controls and tapchangers used to change the effective turns-ratio of the transformer. Included are:

· Load tapchangers

· Diverter switch

· Selector switch

· Contacts

· Drive mechanism

· No load tapchangers

· Voltage regulating controls

Containment System:

The containment system provides a physical boundary between the transformer and the outside world. It insures that the oil does not get contaminated with air or water and that the oil does not leak out into the environment. Included in the system are:

· Tank

· Inert gas system

· Conservator

· Pressure relief device

External Interface System:

The transformer interfaces with other equipment through sets of bushings. These bushing themselves are comprised of:

· Bushing dielectric system

· Insulating paper

· Oil

· Bushing conductor

· Bushing containment system

· Porcelain

· Ground sleeve

· Seals

Cooling System:

The cooling system includes all peripherals and auxiliary equipment required to ensure the transformer can carry rated load at a temperature that does not lead to a pre-mature loss of life. Components included in the cooling system include:

· Pumps

· Fans

· Radiators

· Temperature gages

· Cooling controls

7 Application of Condition Monitoring Strategies

Effective condition monitoring strategies requires that an event or condition take place prior to failure so that intervention can take place to prevent the failure. To be successful, the condition monitoring strategy must be able to clearly differentiate between an acceptable operating condition and an approaching failure yet provide adequate time for pre-emptive maintenance response. Several of the previously described age-reliability curves, suggested that a condition monitoring approach may be appropriate. To better determine if condition monitoring is applicable, the failure pre-cursor conditions must be understood. Four general pre-cursor conditions are described below:

Some failures occur with little or no warning or require an external event to initiate. These types of failures, while random must be prevented by design enhancements or the addition of “safety features”. For power transformers, insulation failures caused by lightning or switching surges are of this type and difficult to predict; installation of Surge Arresters is an appropriate preventive approach.

Figure 10 "No Warning" Failure Pre-cursor

Some failures have a recurring pattern of failure followed by a short recovery period. These types of failure patterns can be corrected if the root cause of the problem is properly identified. Lubrication problems associated with infrequently used mechanical devices are a typical example; operation of the mechanical device temporarily rejuvenates the lubrication.

Figure 11 "Temporary" Failure Pre-cursor

Slowly deteriorating items are good candidates for condition monitoring if a reliable condition indicator can be identified. An “end-of-life” condition trigger must be properly identified so to allow pre-emptive action can take place and that a surprised increase in the rate of deterioration does not result in a failure that could have been avoided.

Many of the pre-cursor failure conditions identified by DGA are of this type.

Figure 12 Slowly Deteriorating Pre-cursor

Rapidly deteriorating items are also good candidates for condition monitoring if a reliable condition indicator can be identified. An “end-of-life” condition trigger is easier to identify since the probability of prematurely taking pre-emptive is low. Some of the pre-cursor failure conditions identified by DGA are of this type.

Figure 13 Rapidly Deteriorating Pre-cursor

It is clear that three of the four pre-cursor patterns identified above are good candidates for application of a condition monitoring maintenance strategy. The strategy must:

· Significantly reduce the probability of failure

· Be technically effective

· Be economically effective

· Provide Maintenance and Operations with sufficient warning so they can intervene

· Result in higher reliability than other traditional strategies

· Be continuously managed since the time between the onset of a failure and an actual failure may be quite short.

· Result in an overall reduction in risk.

8 Failure and Trouble Experience

The previous sections focused on the ways a transformer can fail. In order to develop a predictive failure model, these theoretical failure modes must be “calibrated” with actual failure and trouble experiences.

Industry Experience

The utility industry has been relatively active in sharing information about failures, but very in-active when it comes to sharing failure statistics; in-other words, the industry knows about specific catastrophic failure events but knows very little about major failures that do not cause widespread outages, minor failures, failure rates and retirements; information that is necessary to develop accurate statistical transformer life models. One of the leading utility insurers, Hartford Steam Boiler Inspection and Insurance Company has published some of its observations based on data collected by the International Association of Engineering Insurers (IMIA) for the years 1997 through 2001 involving 94 reported failures. These observations are summarized in Table 3 below and give us only limited insight into the failure process.

Cause	Impacted System	Percent of Reported Failures
Insulation failure	Dielectric	26%
Manufacturing failure	Unknown	24%
Unknown	Unknown	16%
Loose connections	Current Carrying	7%
Improper maintenance	Unknown	5%
Overloading	Dielectric	5%
Oil contamination	Dielectric	4%
Line surges	Dielectric/Mechanical	4%
Fire/explosions	Dielectric/Containment	3%
Lightning	Dielectric	3%
Floods	Containment	2%
Moisture	Containment	1%
Total		1.00%

Table 3 Cause of Transformer Failures as reported to IMIA-Source: HSB and Cigré

A summary of failures aggregated by the affected transformer system is shown in Figure 14. It is only possible to make some generalized conclusions from the data published by HSB. These generalized conclusions include:

· The number of reported failures appears to be much smaller than the actual number of failures experienced by the industry making statistically valid conclusions impossible.

· Only relative relations between failure modes can be made.

· Unknown and insulated related failures dominate the modes of failure.

Figure 14 Cause of Transformer Failure by Impacted System

Edison Experience

Edison has experienced an average of one A-bank failure per year since 1989 (see Figure 15). Most of these failures were associated with aged transformer and resulted in their replacement.

Figure 15 “A Bank” Failure Events

While root cause analysis did not take place for many of the failures, the system initiating the failure could be determined by assuming that the cause of the defect was associated with the system containing the defect. The distribution of failures by impacted systems is summarized in Figure 16 below.

When comparing the failure distribution experienced by Edison to that reported by HSB, it is interesting to note that insulation failures are the dominant mode of failure for both yet LTCs are also a significant mode of failure for Edison but not mentioned by HSB. It is theorized that the omission of LTC failure modes by HSB is deliberate or that the population of transformers included in their study did not have LTCs.

Figure 16 “A Bank” Failure Distribution by Impacted System

Many times troubles are a sign of pending failure. Current troubles being experienced by Edison are listed in Table 5 and Figure 17 below. While many of these troubles are easily corrected, they can an indicator of increasing failure risk.

Table 5 Current “A Bank” Problems

Figure 17 “A Bank” Trouble Distribution

It is interesting to note that while dielectric problems are the dominant cause of failures, they are not the leading cause of troubles experienced by Edison. This observation could imply that dielectric failures tend to manifest themselves fairly quickly and that the trouble experience takes a much longer time to evolve into a failure.

Summary of Experiences

Both Edison and industry experience concludes that nearly half of the causes for transformer failure originate in the dielectric system. The remaining causes of failure are only vaguely identified at the industry level while Edison data strongly identifies Load Tapchangers, Electromagnetic Systems and Bushings as dominant causes of failure.

The linkage between troubles and eventual failures appears to be clear for bushings but the linkage between insulation failures and reported trouble experiences is much less obvious even though one could argue that inert gas problems and oil leaks result in a degradation of the dielectric system.

9 Predicting End-of-Life

The utility industry does not have a good grasp on the expected life of a power transformer. Insurance companies predict an expected life of 35 years, regulators and accountants use a book life of 40 years yet Edison has a significant population of transformers over 45 years of age with no signs of pending failure. Developing a good failure model is paramount for building an accurate business case. Three failure models have tested in the development of this business case and to predict the expected reliability of a power transformer. The models are:

· Linear or constant failure rate

· Hartford Steam Boiler (HSB)Transformer failure model

· Weibull equation based on Edison experience

Linear Model

The linear or constant rate failure model is the simplest and widest used predictor of transformer life. Simply stated, Edison’s “A Bank” failure experience in which 15 out of 188 transformers have failed over the past 15 years results in an average failure rate of 0.53% per year. The expected annual failure rate over time predicted by the linear model is shown in Figure 28.

Figure 28 Linear Failure Model (Estimated and Actual)

HSB Model

The Hartford Steam Boiler Inspection and Insurance Company has published several transformer failure models based on the 1825 work of a statistician named Benjamin Gompertz. The latest variation of the Gompertz model developed by W. M. Makeham is used in this analysis. The HSB model is:

f(t) = A + αe^βt

Where:

A is a constant for random failures caused by lightning, vandalism, switching, etc.

A = 0.5

α = scaling factor

= 0.00007346

β = time constant

= 0.176190651

The expected annual failure rate over time predicted by the HSB model is shown in Figure 29 below.

Figure 29 HSB Failure Model (Estimated and Actual)

Weibull Model

The Weibull distribution is one of those generic models that has worked well to accurately describe the age-reliability relation of many failure modes. The primary advantage of Weibull analysis is the capability to provide accurate failure analysis and risk predictions with extremely small samples. Solutions are possible at the earliest stage of a problem without requirements to “fail a few more”. For purposes of determining optimum maintenance interval, the two-parameter density function is use. The Weibull function is:

Where: t = failure time

β = wear characteristics

β < 1.0 indicates infant mortality

β = 1.0 indicates random failures

β > 1.0 indicates wear out failures

η = Characteristic life (time when 63.2% have failed)

The expected annual failure rate over time predicted by the Weibull model is shown in Figure 30 below.

Figure 30 Weibull Failure Model (Estimated and Actual)

Prediction Comparison

Each of the above models appears to reasonably estimate the expected failure rate of a relative young transformer population. Application of the models to the current Edison “A Bank” transformer population yields interesting results as shown in Figure 31. The HSB Model predicts a very high but decreasing failure rate for the existing fleet of “A Bank” transformers. The Linear Model predicts a low but constant failure rate and Weibull Model predicts a low but slowly increasing failure rate.

Figure 31 Predicted Failure Rates for the Current Edison Transformer Population

Failure Pre-cursor Model

In order for any condition monitoring technique to be effective, there must be a monitored attribute that changes state prior to failure. This monitored attribute must have at least three distinctive states:

· Acceptable state

· Incipient failure state

· Failed state

The change of states must be recognized with ample response time to allow intervention and failure prevention. For transformer dielectric failures, the failure precursor or incipient state is the appearance of various dissolved gasses in the oil. While the industry has significant successful experience of detecting and mitigating slowly evolving dielectric failures using traditional DGA sampling techniques, it has less favorable experience with quickly developing failures. It is this latter set of failures that multi-gas online monitors will detect at an early stage and prevent the occurrence of a major failure.

To account for this varying failure pre-cursor, a simple model was employed that assumed a wide range of times between the onset of a failure and the eventually catastrophic loss of the transformer. The model assumed a normal distribution or “bell curve” of failure pre-cursor gas evolution. Normal distributions are a family of distributions that have the same general shape. They are symmetric with more events concentrated in the middle than in the tails. Normal distributions are sometimes described as bell shaped. The area under each curve is the same. The normal distribution can be specified mathematically in terms of two parameters: the mean (μ) and the standard deviation (σ).

In this analysis, parameters used to develop the normal distribution were:

· Mean incipient failure time in months (mean time from abnormal gas generation to failure)

· Standard deviation of incipient failure times.

A graphical presentation of the Failure Pre-cursor model is shown below in Figure 32. The model assumes:

· Mean time between the evolution of abnormal combustible gas and a major failure is 14 months (incipient fault time).

· The standard deviation for an incipient fault is 6 months

· Periodic DGA sampling is on a 12 month interval

· The probability of detecting a major failure with periodic DGA is 63%

Figure 32 Failure Pre-cursor Model

10 Financial Model

The financial model looks only at the multi-gas sensor application for a typical transformer and will include:

· Installed cost

· Annual probability of failure

· Probability of catastrophic failure

· Collateral damage

· Probability of collateral damage

· Various PM Responses

· No Maintenance

· Periodic DGA

· On-line monitoring

· Replacement cost

· Indirect failure effects

· Environmental

· Political

· Contractual

· Regulatory

· NPV

· IRR

· Etc.

The Analysis

An economic analysis on the impacts of dielectric maintenance was performed on an “average A Bank” transformer. The analysis looked at the expenditures and benefits expected over a 20 year period. The analysis examined three maintenance scenarios:

· No maintenance of the dielectric system

· Periodic DGA

· On-line monitoring

Important parameters and assumptions employed by the model were:

	Parameter
●	System Analyzed
●	Average Age
●	Replacement Cost (on the pad)
●	Cost of Non-Catastrophic repair
●	Percent of Major Failures Catastrophic
●	Probability of Collateral Damage
●	Failure Model
●	DGA Effectiveness
●	On-line Monitoring Effectiveness
●	Insurance coverage
●	Annual DGA Cost-Loaded (1 sample per year)
●	Cost of On-line Monitor
●	On-Monitor O&M Cost
●	Weighted Average Cost of Capital (WACC)

Table 6 Key Financial and Technical Parameters

Benefits Used in the Model

The model included direct benefits associated with reduced O&M costs as well as some indirect costs associated public relations. Benefits available in the model the model are listed in Table 7.

	Benefit	Included
●	Reduced Failure Rate	Yes
●	Reduced Maintenance Costs	Yes
●	Cost of Non-Catastrophic repair	Yes
●	Probability of Collateral Damage	Yes
●	Outage time	Yes
●	Environmental Cleanup	Yes
●	Customer Claims	Yes
●	Public Relations	Yes
●	CPUC	No
●	Replacement Power	No

Table 7 List of Benefits Available and Included in the Financial Model

The Failure Model

The Weibull failure model was chosen over the other two models for three key reasons:

· The model was derived from 15 years of actual Edison failure experience

· The HSB Model predicts a failure rate much higher than Edison is currently experiencing

· The insulation failure mechanism is partially a function of the strength of the insulating paper. The strength of the paper decreases with age resulting in an expected higher probability of failure. A constant failure rate does not accurately model this known aging process.

The resulting Weibull aging model derived from 15 years of failure history is shown in Figure 33. The X-axis is the Cumulative Probability of Failure and the Y-axis is age. Since the available data only included 15 years of failures it predicted that 63% of the transformer insulation systems would survive until 165 years of age. This is not a flaw in the model but represents the fact that 61 years of failure were omitted from the analysis since no failure data was available for years 1928 through 1988. In order to correct for this missing data, the model was “calibrated” to achieve a 2005 failure rate that equals the 15 year average. This “calibration” now predicts that 63% of the transformer insulation systems would survive until 101 years of age.

Figure 33 Weibull Model of Edison Insulation Failure Experience

The Results

The economic analysis revealed positive results for on-line monitoring as compared to no maintenance. The on-line monitoring program has excellent financial results:

· Initial Investment = $34,000

· Annual O&M Costs = $3,000

· IRR = 42.03%

· NPV = $110,738

· Payback = 5 Years

Figure 34 Cumulative PV Cash Flow of On-line Monitoring vs. No Maintenance

While the business case for installing an on-line monitoring system is excellent compared to performing no maintenance, it must also be compared to the current periodic DGA program. It is no surprise that the current periodic DGA has even better financial performance since there are no initial capital costs and the technical aspects of DGA are well proven. The economic results associated with periodic DGA testing were:

· Initial Investment = $5,000 (need to start a new program)

· Annual O&M Costs = $250

· IRR = N/A

· NPV = $104,289

· Payback = Immediate

Annual cash flow comparisons of all three options are shown in Figure 35 below.

Periodic DGA can be summarized as a low-cost investment with a fantastic rate of return. The benefit it has historically provided Edison is well documented and universally accepted. On-line monitoring cannot match the overall economic value of DGA but is still an excellent investment and a low cost method of achieving even greater power transformer reliability at a modest cost. The real benefit of on-line monitoring will be exposed in the following section that discusses risk reductions.

Annual Reductions in Expenditures associated with On-line Monitoring

Annual Expenditures associated with No PM Maintenance program

Annual Expenditures associated with On-line Monitoring

Annual Expenditures associated with DGA

Annual Reductions in Expenditures associated with DGA

Figure 35 Cumulative PV Cash Flow of On-line Monitoring vs. DGA vs. No Maintenance

Deferred Transformer Replacement

One benefit excluded from the above financial model that has the potential of overshadowing all the other benefits is life extension. While most on-line monitoring will not renew an item (reducing its effective operational age), there is sometimes a desire retire transformers because they are “old” and have served their expected useful life. The main reason for this early retirement is risk reduction and assumed improvement in overall reliability. While there is merit to this approach, there is also the possibility that a portion the transformers available life goes unused. The implementation of on-line monitoring would solve the risk and reliability issues and potentially result in additional operating years. These extra operating years result in a substantial amount of deferred capital expense which can amount several hundred thousand dollars per year (see Figure 36).

Figure 36 Financial Impacts of Extending the Useful Life of a Transformer with On-line Monitoring

11 Risk Model

It is of low benefit to make an investment in a transformer on-line monitoring system if its overall impact on transformer reliability is minimal. The investment must be framed in the context of all dominant modes of failure and the impact on long term risk. The risk model for the four dominant transformer failure modes will be analyzed along with the risk impacts of various PM responses.

For purposes of this document, Risk is defined as:

Risk = Probability of Failure X Severity of Failure

The No Maintenance Risk

Figure 37 shows the expected modes of failure distribution for the current fleet of Edison “A Bank” transformers if no maintenance is performed. The figure is based upon the probability of failure of each transformer over the next 10 years and the expected impact of the failure. The prediction agrees well with both Edison and Industry experience forecasting that nearly half the failures will be initiated by the insulation system. These results reinforce the need to monitor the condition of the main insulation package.

Figure 37 Transformer Failure Risk Distibution if No Maintenance is Performed

How Maintenance Affects Risk

Obviously the implementation of a good PM program will reduce the risk of “A Bank” transformer failures. The question of how much risk reduction is achieved if various PM programs are implemented must be answered.

Using the same models described in sections 12 and 13, an analysis of the risk Edison should expect to experience over the next 10 years was made. This analysis examined:

· No Maintenance

· Existing PM program

· Existing PM with the inclusion of Multi-gas on-line monitoring

The results of this analysis showed a significant decrease in risk (see Figure 38) is currently achieved with the present PM program and meaningful additional reduction in risk can be achieved by implementing a multi-gas on-line monitoring system.

Figure 38 Transformer Failure Risk Associated with Various PM Approaches

Executive Summary

1 Introduction

2 BUSINESS Opportunity

3 Objective of this Study

4 Scope of Study

5 Critical Transformer Systems and Functions

6 How Transformers Fail

7 Application of Condition Monitoring Strategies

8 Failure and Trouble Experience

9 Predicting End-of-Life

10 Financial Model

11 Risk Model

12 Data Requirements

13 Conclusions