Crow AMSAA PDF

Crow/AMSAA Reliability Growth Plots
And there use in Interpreting Meridian Energy Ltd’s, Main Unit Failure Data
by
Nigel Comerford
Areva T&D New Zealand
Purpose
To explore the Crow/AMSAA model and its use in monitoring and projecting reliability
growth or performance of existing plant and apply this technique to the main unit failures
of Meridian Energy’s generating system, allowing confirmation of reliability
improvement, those improvements to be quantified, correlation of operational events
against reliability, and forecast of failure rates.
16th Annual Conference 2005- Rotorua
Page 1 of 22
1.0 Executive Summary
Meridian Energy is a New Zealand state owned enterprise, which operates, among other
assets, 38 hydro machines in a deregulated electricity market. It is the Forced Outages or
unit failures as they apply to this system that is the subject of this paper.
The key objectives are to explore the Crow/AMSAA model and its use in monitoring and
projecting reliability growth or performance of existing plant and apply this technique to
the main unit failures of Meridian Energy’s generating system, allowing confirmation of
reliability improvement, those improvements to be quantified, correlation of operational
events against reliability, and forecast of failure rates.
Although Forced Outages, within this industry, do not have a great affect on availability
they do cost money and expose the business to risk.
The Crow/AMSAA technique involves plotting, most commonly, cumulative failures Vs
cumulative time on a log-log scale with the resulting straight lines’ slope indicating
improving, deteriorating, or constant reliability. Instantaneous failure rate can be
determined, and due to the straight-line nature of the plots, forecasts can be made of
failures into the future. This method handles mixed failure modes, so is therefore suitable
for the complex nature of the generating units, which at a system level exhibit random
failures of mixed modes.
A review of the last five years of Forced Outage data has determined the average
estimated cost of a Forced Outage at $6500 per event.
Crow/AMSAA plots of the last nine years of data show that although on average the
reliability of the system is constant there was a deteriorating situation up until the turn of
the century, then there has been consistent year on year improvements; that is we were at
a level then deteriorated and have since improved to be at our best now.
This performance mapped against operational history shows this improving situation
started after the unsettled period of the late 90’s and the Automation & Remote Control
project finishing and seems to have been driven by the Event Analysis system. The
analysis also shows that although the numbers of machine starts and the amount of
generation have both increased slightly and that the direct maintenance spend has dropped
there has still been a great improvement in reliability.
A sound management tool has been proven which allows forecasting of Forced Outages
allowing their affect on availability and unplanned maintenance costs to be known.
Combining this information with the cost of each Forced Outage, shows that if there had
been no improvement of the performance from the 1999 – 2000 period we could have
expected to see as much as 150 extra Forced outages in the last year alone, the reduction
in Forced Outages that has been achieved equates to savings of $975,000 / year.

This paper is the property of VANZ. Reproduction is not permitted without permission.
Page 2 of 22
2.0 Contents
1.0 Executive Summary ............................................................................................. 2
2.0 Contents ..................................................................................................................... 3
3.0 Introduction................................................................................................................ 4
3.1 Objectives .............................................................................................................. 4
3.2 Meridian Energy’s Generation System.................................................................. 4
3.3 Why should we reduce Forced Outages & how will measuring Forced Outage
Rate in this manner help .............................................................................................. 4
4.0 The Duane Crow/AMSAA Model ............................................................................. 5
4.1 History.................................................................................................................... 5
4.2 Application............................................................................................................. 5
4.3 Crow/AMSAA Model Theory ............................................................................... 7
4.3.1 Further Developments......................................................................................... 9
5.0 Crow/AMSAA Software............................................................................................ 9
6.0 Failure Data & Definitions....................................................................................... 10
6.1 Definition of a Failure.......................................................................................... 10
6.2 Period of Analysis................................................................................................ 10
6.3 Data Sets .............................................................................................................. 10
7.0 The Cost of a Failure................................................................................................ 11
7.1 Why is it important to know what Forced Outages cost...................................... 11
7.2 Forced Outage cost build-up................................................................................ 11
8.0 Meridian Energy’s FO data as Crow/AMSAA plots ............................................... 12
8.1 Complete Nine-Year Plot..................................................................................... 12
8.2 First Four-Years Plot............................................................................................ 13
8.3 Last 5-Year Plot ................................................................................................... 13
8.4 Year by Year Plot................................................................................................. 14
8.5 2003 – 2004 Financial Year plot.......................................................................... 15
9.0 Organisation Operational History & Correlation with Failure Rate........................ 15
1.1.1 9.1 Operational & Maintenance Systems Timeline ................................... 16
1.1.2 9.2 Operational Parameters & Beta Value Comparison ............................ 17
9.3 Correlation Between Failure Rate and Historic Parameters ................................ 18
10.0 Failure Forecasting................................................................................................. 19
10.1 Limitations ......................................................................................................... 19
10.2 Why do we want to Forecast?............................................................................ 19
10.3 Forecasting Method ........................................................................................... 19
11.0 Conclusions............................................................................................................ 20
Acknowledgements........................................................................................................ 22
References...................................................................................................................... 22

Page 3 of 22
3.0 Introduction
3.1 Objectives
The key objective of this paper is to explore the Crow/AMSAA technique and to apply it
to the evaluation of historic and future ‘Forced Outage’ occurrences of Meridian Energy’s
generating plant. This will include an overview of the history and theory of the
Crow/AMSAA Reliability Growth Model to enable a depth of understanding as to apply
the technique; choose software to allow cost effective application, extract and format
Forced Outage data from Meridian Energy’s databases which will produce suitable inputs
for the model, determine an average estimated cost of a Forced Outage to enable the
quantification in dollars terms of improvements, create Crow/AMSAA plots to interpret,
& confirm historic reliability growth and or degradation, and to build a high level time-
line of Meridian Energy’s operational and maintenance history, to allow correlation of
reliability against historic events, with an aim of learning what factors are propagating
growth. Additionally forecasting of failure rate, allowing more certainty around
availability and financial budgets, the quantification of proposed reliability improvement
plans, and the uses for management of Crow/AMSAA plots will all be discussed.
3.2 Meridian Energy’s Generation System

Meridian Energy is the largest producer of hydro electricity in New Zealand. It operates
nine large hydro stations, comprising 38 main generating machines, spread across the
lower South Is of New Zealand with a combined installed capacity of 3500MW. It is this
system or fleet of 38 hydro machines, viewed as a whole, which is the subject of this
paper in respect to applying the Crow/AMSAA technique. This system of machines,
which vary from multi unit, 120MW Francis turbine stations, to a 25MW, single unit
station, Kaplan machine, are all controlled remotely from one location and unmanned.
The system operates within a deregulated electricity market where generation is sold
based on availability of both machines and fuel Vs demand.
3.3 Why should we reduce Forced Outages & how will measuring Forced
Outage Rate in this manner help
Forced outages do not have a great impact on availability however they do impact the
systems exposure to risk, cost money, and they indicate generally the overall health of the
system.
The fewer Forced Outages there are the less exposure to ‘Revenue Opportunity Cost’, the
cost of lost generation due to the units sudden unavailability and the systems inability to
pick-up the generation, and the cost of market imposed penalties for repeated failure to
deliver offered generation.
Overall the health of the generating system is perceived to be better when fewer Forced
Outages are occurring. It is seen as an indicator of the general health of the plant, the
fewer Forced Outages there are, the fewer high priority alarms, the fewer corrective
maintenance work orders, and the fewer unknown plant conditions. Our probable
exposure to “the big one”, that 1 in 200 Forced Outage event that causes considerable
plant damage, cost, or injury is less frequent when our failure rate is less.
Effective measuring, allowing transparency of reliability, cusps in the failure rate, and
Forced Outage forecasting and quantification will allow better management decisions to

Page 4 of 22
be made on factors affecting Forced Outages. This will also allow decisions to be made on
the cost of maintaining a set Forced Outage Rate or wanting to improve it further.
4.0 The Duane Crow/AMSAA Model

Reliability Growth Plots have a variety of names, such as Duane Plots, Crow Plots, or
Crow/AMSAA Plots. The technique involves plotting, most commonly, cumulative
failures Vs cumulative time on a log-log scale with the resulting straight lines’ slope
indicating improving, deteriorating, or constant reliability. The literal beta value of >1
shows failures increasing, 1 indicates constant failures, and <1 shows failure rate
improving. Instantaneous failure rate can be determined, and due to the straight-line
nature of the plots, forecasts can be made of failures into the future. This method handles
mixed failure modes, so is therefore suitable for the complex nature of the generating
units, which at a system level exhibit random failures of mixed modes.
4.1 History
Reliability Growth Modeling has its origins in the tracking of the improvements in
manufacturing times and has been exhaustedly demonstrated as a true log-log
phenomenon. T. P. Wright in 1936 pioneered an idea that improvements in the time to
manufacture an airplane could be described mathematically. Wrights findings showed
that, as the quantity of airplanes were produced in sequence, the direct labour input per
plane decreased in a mathematical pattern that forms a straight line when plotted on log-
log paper.4
Learning curves were used extensively by General Electric and a GE reliability engineer
made log-log plots of cumulative MTBF Vs cumulative time, which gave a straight line,
(Duane 1964). James Duane developed a deterministic postulate for monitoring failure
rates of more complex systems over time using a log-log plot with straight lines.
At the US Army Material Systems Analysis Activity (AMSAA) during the mid 1970’s
Larry Crow converted Duane’s postulate into mathematical and statistical proof via
Weibull statistics in MIL-HDBK-1893.
By the end of the seventies there were several dozen different growth models in use. The
Aerospace Industries Association Technical Management Committee studied this array of
methods as applied to mechanical components and concluded the Crow/AMSAA model
was the best. The U.S. Air Force study, conducted by Dr Abernethy1, including both
mechanical and electrical controls, reached the same conclusion.
4.2 Application
Although the Crow/AMSAA model has its base in measuring reliability growth, defined
as, “The positive improvement in a reliability parameter over a period of time due to
changes in product design or the manufacturing process”, it is the wider definition of
reliability growth management, which is defined as, “The systematic planning for
reliability achievement as a function of time and other resources, and controlling the
ongoing rate of achievement by reallocating of resources based on comparison between
planned and assessed reliability values”, which is more relevant for our purposes.
With reliability growth management in mind the Crow/AMSAA plot has more recently
found other important applications. Many industries are routinely doing Crow/AMSAA
analysis and it is considered best practice for tracking fleets of units to trend reliability,

Page 5 of 22
safety, and maintainability events of interest. It is capable of utilizing dirty data, including
missing data sets, and mixtures of failure modes. It is also applicable for tracking
parameters that interest management and forecasting their future levels.
The Crow/AMSAA model will be applied to the Forced Outage data of Meridian’s
systems of generators. Cumulative failures over cumulative time are plotted using suitable
software, (see below). This delivers a graphical straight-line plot, with a goodness of fit
test, and extrapolation. Failure rate, hazard rate, and the slope of the line are all delivered
which will allow realisation of the following benefits.
Trend charts that make reliability growth or degradation more visible and
manageable.
The progress of the reliability improvement program, the effects of changes and
improvements are measured and displayed.
Reliability predictions are available early to compare against business
requirements.
Adverse system reliability trends are indicated much sooner and more accurately
than rolling averages.
System or component level reliability trends can be measured even with mixed
failure modes.
It is important to note that the reliability improvement process is modeled, not just the
physical system. For example a system may have a number of components that initially
exhibit wear-in characteristics, which as this phase elapses, a generally improving failure
rate is exhibited in the CA model. However the CA model will show equally well the
instigation of a Condition Monitoring program or the doubling of RCFA effort.

Page 6 of 22
4.3 Crow/AMSAA Model Theory
Plotting the observed number of cumulative failures against cumulative time on
logarithmic paper forms the basis of the model from a deterministic point of view.
For example, given the observed times between failures of a machine in Table 1;
Failure Date Cumulative Failure Failure Date Cumulative Failure
Time in Days Number Time in Days Number
2/05/2003 4 1 16/08/2003 110 13
8/05/2003 10 2 25/08/2003 119 14
16/05/2003 18 3 30/08/2003 124 15
27/05/2003 29 4 2/09/2003 127 16
4/06/2003 37 5 14/09/2003 139 17
14/06/2003 47 6 28/09/2003 153 18
20/06/2003 53 7 11/10/2003 166 19
30/06/2003 63 8 25/10/2003 180 20
10/07/2003 73 9 5/11/2003 191 21
14/07/2003 77 10 20/11/2003 206 22
18/07/2003 81 11 4/12/2003 220 23
8/08/2003 102 12 20/12/2003 236 24
Table 4.1
Then by plotting the cumulative failures over cumulative time, as in Figure 4.1, and
assuming 1 to 1 scale log paper is used, the following can be determined.

Page 7 of 22
Log-Log Plot of Machine Failures
1000
Cumulative Failures
100
10 Failure
Number
8.1cm
Beta
1 0.81
Lambda
10cm
0.29
0.1
1 10 100 1000
Cumulative Time Days
Figure 4.1
Cumulative Failure Events at time t is represented by n(t). The scale parameter λ, is the
intercept on the y-axis of n(t) at t = 1 unit. With the data plotted in Figure 1, λ is read off
as 0.29. The slope β can be measured graphically providing log paper with a 1 x 1 scale is
used; in the above case Beta is 8.1 / 10 = 0.81.
The model’s intensity function p(t) measures the instantaneous failure rate at each
cumulative time point. The intensity function is;
ρ (t ) = λβt β −1
The log of the cumulative failure events n(t) verse the log of cumulative time is a liner
plot if the model applies;
n(t ) = λt β
Taking natural logarithms of the equation yields;

Lnn (t ) = Lnλ + βLnt
Which implies the equation above is a straight line on logarithmic paper.

Page 8 of 22
The interpretation of the slope, beta is the same as on a Weibull plot. If the slope is
constant, the failure rate is constant (β = 1), and the process is termed a Homogenous
Poisson Process. If the slope, β, is greater than one, the failure rate is increasing, and if the
slope is less than one the failure rate is decreasing, as indicated by the example machine
in Figure 1.
The cumulative failure rate, C(t), that is the long term average failure rate equals;
n(t )
C (t ) = = λt β −1
t
The reciprocal of p(t) is the instantaneous MTBF, as the reciprocal of C(t) is the
cumulative MTBF, both of which can be modeled as an alternative.
Goodness of fit is indicated by the proximity of the points to a straight line. Curvature or
discontinuities may be observed, but this is part of the process. As improvement in the
reliability of the modeled system becomes apparent, a cusp or corner should appear on the
plot at the point of change. The straight line should be fit from this point onwards to
model the latest process.
The equation, n(t ) = λt β , can be used to predict future failures,
1
⎛ n ⎞β
t =⎜ ⎟
⎝λ⎠
For example the 25th failure, to continue Table 1, will occur at (25/0.29)^(1/0.81) = 245,
therefore the next failure will occur in 245-236 = 9Days.
4.3.1 Further Developments

Maximum Likelihood Estimates can be used to determine lambda and beta from which
p(t) and C(t) can be derived. This method is used with time and failure terminated data or
interval and group data. See references 1, 2, or 3 for more detail.
AMSAA Technical Report 6522, which updates both IEC1664 and Mil-Hbk-189,
describes an unbiased method for determining the parameters suitable for use when the
individual failure times are known. Paul Barringer investigated the accuracy of all three
methods, Rank Regression, MLE, and IEC, with IEC being the best bearing in mind it
requires the failure times to be known. However it should be noted that at least 20 data
points are required for any method to work.
5.0 Crow/AMSAA Software

There are a number of software packages on the market that perform Weibull analysis of
failure data and some of these packages include a Crow/AMSAA or Duane function,
however to fully utilise the model, a number of key features should be included.
Ability to plot multiple datasets and their fit lines on one graph
Ease of import of data from common packages such as MS Excel

Page 9 of 22
Fully scalable, colour, and editable graphs, which can be exported easily to
common software.
Ease of extrapolation of both the x and the y axis
Able to calculate both cumulative and instantaneous failure rates.
Provide a choice of methods, i.e. Rank Regression, IEC, and MLE
Able to convert data from interval to cumulative
Ability to plot in either discrete failures of MTBF.
A number of software packages were trailed for this project, however ‘WinSMITH
Visual’ was chosen as it fulfilled all the above criteria.
6.0 Failure Data & Definitions

6.1 Definition of a Failure
Failures as discussed in the project are made up of four types;
The ‘Tripping’ of a main generating unit, that is, a unit unexpectedly shutting
down while still connected to the grid
A unit being forced from service due to imminent failure, that is, a unit with a
known recently developed fault that is purposefully shut down, before it ‘Trips’,
to repair that fault
A unit failing to start when called upon, that is, the unit being given a start
command and the unit failing to be online within a prescribed time
And a unit failing to stop, that is, after a stop command is given, the unit failing to
come to a complete stop within a prescribed time.
All of these types of failures are classified by ‘North American Electricity Reliability
Council’ codes, such as U1, U2, U3, SF, and all of these failures can occur from a mixture
of failure mechanisms and modes, in addition to the expected electrical & mechanical
failure modes, causes such as PLC code, operational policy, human error, and faults on
the connected transmission grid operator’s system, causing unit failures, are all included.
6.2 Period of Analysis

Failure data, as defined above, is available in electronic form for Meridian Energy’s
generating system dating back approximately 15Years, however due to the diminishing
relevance of past data and the difficulty in matching operational history, it will be the last
9 years data that will be included in the analysis with an increasing focus on the more
recent data. The period covered will be from the end of July 1995 to the end of June 2004.
6.3 Data Sets

Initially the nine-year total, nine year, (year by year), first four years, last five years, and
last 12 months data sets will be analysed, with particular attention being paid to cusps and
discontinuities. Further more detailed plots will also be made where appropriate to
investigate cusps and trends.

Page 10 of 22
7.0 The Cost of a Failure
7.1 Why is it important to know what Forced Outages cost
If the cost of forced outages is known, this can be used by management, in making
decisions on the affect Forced Outages have on business and the cost of any effort in
maintaining or reducing the Forced Outage Rate Vs any financial benefit by that reduction
to the business.
7.2 Forced Outage cost build-up

An estimate of the average cost of a Forced Outage has been determined based on the last
five years of Forced Outage data. It comprises of direct costs including the actual costs to
repair failures and the indirect costs such as the cost of RCFA and ROC.
Direct Costs
Average cost of the Initial Callout Work Order, based $330
on the past 12 months of callouts.
Average cost of the Follow-up Work Orders to actually $815
repair the fault, based on the past 12 months of Forced
Outages.
Average cost per Forced Outage over the last five years $440
of the big failure events. Of which there has been three,
MAN03 Baffle Plate, AVI Battery Bank, and BEN03 CB.
Indirect Costs
Average cost of Asset Coordinators, Tactical Engineers, $1000
and AREVA Engineers time for RCFA per Failure Event
based on a subjective estimate @$100/Hr
Average cost of ROC per Forced Outage based on data $3430
from the last five years.
Total $6015
This estimate is sensitive to the accuracy of the data within our CMMS and is drawn for a
number of systems, as there is no consistent effort to record the true actual cost of each
Forced Outage. It is also of note that the affect of ROC seems much less in recent years
than say four or five years ago. This cost could be further inflated by less tangible costs
such as the cost of disruption to scheduled work due to a Forced Outage or the increased
personal risk associated with responding to a callout after-hours.

Page 11 of 22
So, based on a review of the last five years of Forced Outage data, on average, each
forced outage is estimated to cost, allowing for the time cost of money, and other factors
mentioned above, approximately $6500.
8.0 Meridian Energy’s FO data as Crow/AMSAA plots

8.1 Complete Nine-Year Plot
Figure 8.1
The failure data from the last nine financial years, from July 1995 to the end of June 2004,
has been plotted above in Figure 8.1, using the IEC method. The plot shows a beta value
of 0.969, which indicates that over this period, on average, reliability has been relatively
constant, i.e. no improvement or degradation. However when the discrete points are
examined cusps can be seen indicating periods where improvements have been made and
periods when degradation in failure rate are evident.
The Cumulative Failure Rate is 0.01874 and the Instantaneous Failure Rate is 0.01751,
these two figures are very close, again indicating little overall improvement, on average,
in the last nine years.
Extrapolation of the plot shows that if failures were to continue at the above instantaneous
rate there would be 153 failures in the upcoming financial year.

Page 12 of 22
8.2 First Four-Years Plot
Figure 8.2 shows data from the first four financial years of our period, from July 1995 to
the end of June 1999, using the
IEC method. The plot shows a
beta value of 1.2, which indicates
that over this period, on average,
reliability was deteriorating. The
cusps that appear around the
middle of Figure 8.1 are evident
towards the end of this data set
with an obvious deterioration
from the end of 1996 till mid
1999.
Figure 8.2
again indicating deterioration over time.
Extrapolation of the plot shows that if failures were to continue at the rate they were
around the end of 1999 that is with an increasing rate of occurrence and no added effort to
reduce our failure rate then there would be 240 failures in the upcoming financial year.
8.3 Last 5-Year Plot

Figure 8.3 shows data from the last
five financial years, from July 1999
to the end of June 2004. The plot
shows a beta value of 0.805, which
indicates that over this period, on
average, reliability has been
improving. This indicates there has
been a major turn around in the last
five years as compared with the
1995 to 1999 period. Some minor
cusps are visible suggesting along
this five-year period there have
been smaller periods of
improvement or degradation, with
an obvious flattening of the discrete
points at the end of the period.
Figure 8.3
again indicating an improving reliability.

Page 13 of 22
Extrapolation of the plot shows that
if failures were to continue at the
above rate there would be 126
failures in the upcoming financial
year. This is less than both the 9-
year and first 4-year plots.
Looking at these two fit lines on the
same graph, Figure 8.4, shows this
obvious improvement graphically by
the contrast in slope of the two fit
lines.
8.4 Year by Year Plot Figure 8.4

Figure 8.5 shows the period on a year-by-year basis with a fit line for each year indicating
the beta value. The changing rate of occurrence of Forced Outages can clearly be seen
both graphically and by the changing beta value, 1995 and 1996 show a period of
improving reliability on par with 2000, to 2002, however the financial years 1997 to 1999
show a period of decreasing reliability with 1997 showing the worst performance. The
most recent year shown, FY2003-2004, the financial year just past, shows a strong
improvement compared with the previous three years.
Figure 8.5

Page 14 of 22
8.5 2003 – 2004 Financial Year plot
Figure 8.6
The failure data from July 2003 to the end of June 2004 has been plotted above in Figure
8.6. The plot shows a beta value of 0.562, which indicates that over this period, on
average, reliability has been improving, that is the rate of occurrence of Forced Outages is
reducing.
The Cumulative Failure Rate is 0.01872 however the Instantaneous Failure Rate is
0.01016, which indicates that the rate at which failures are occurring now is less than the
average rate which suggest an improving situation
Extrapolation of the plot shows that if failures were to continue at the above instantaneous
rate there would be 87 failures in the upcoming financial year. Continuing this
extrapolation suggests that if the current rate of improvement was to continue then the
number of Forced outages occurring each year would reduce by about 3 to 4 per year,
however obviously there would be a theoretical floor once the inherent reliability of the
system is met.
9.0 Organisation Operational History & Correlation with Failure Rate

A key objective mentioned above was to build a high level time-line of Meridian Energy’s
operational and maintenance history, to allow correlation of reliability against historic
events, with an aim of learning what factors are propagating growth, or if indeed, any
improvement can be mapped against any operational or maintenance philosophy. This is
important as it will give guidance in management decisions about what is effective in
promoting reliability growth management as per the definition, “The systematic planning
for reliability achievement as a function of time and other resources, and controlling the
ongoing rate of achievement by reallocating of resources based on comparison between
planned and assessed reliability values”.

Page 15 of 22
1.1.1 9.1 Operational & Maintenance Systems Timeline
Legend M AN Lube Oil Analysis Started Condition M onitoring of CB's VA of UW & M W Aux
Red Line - Annual Beta Value started, M AN VA of Aux plant plant started
started
ECNZ
Finishes, M EL M DT Restructuring
ARC Project started Begins
PRISM Started
1996 1997 1998 1999 2000 2001 2002 2003
1996 1997 1998 1999 2000 2001 2002 2003
TKA O/Haul
DGA Technique in use (started in the 80's),
Battery condition m onitoring started,
Therm ography in use from about 93', UW
EA RCFA Process Started 2M TT Cam e on-line
Lube Oil Analysis in place since early 90's,
M W Lube oil analysis started, BN Racks MAN 1/2 Life Project Started
VA started early 90's, PDA started.
Outsourcing of m aintenance started
1995, ARC Project Com plete PRISM Finished, M EMFOS Started
Asset M angers started to m ove from the
stations to centralised offices
Figure 9.1
c25992 18412203 Page 17 13/12/2006
1.1.2 9.2 Operational Parameters & Beta Value Comparison
Sys
MEL Operational Parameters / Beta Value Comparison MW
1800000
Unit
1600000 Starts
1400000 Mtce
Spend
1200000
Scaled Value
Beta
1000000 Value
800000 Poly.
600000 (Sys
MW)
400000 Poly.
(Beta
200000 Value)
Poly.
0 (Unit
Starts)
Poly.
n
p
ar
ar
ar
ar
ar
ar
ar
ar
ar
Ja
Se
Se
Se
Se
Se
Se
Se
Se
M
M
(Mtce
M onths, 95' to 04' Spend)
Figure 9.2
GEG7144 Reliability Applications Unit Advisor Emile Eerens

9.3 Correlation Between Failure Rate and Historic Parameters
It can be seen from the Crow/AMSAA plots above, the Timeline in Figure 9.1, and the graph
in Figure 9.2 that over the period in question the failure rate, represented by the beta value,
has climbed to reach its worst performance during the 1997 to 1999 period then steadily
improved to reach its best performance today.
Looking firstly at the Timeline, it can be seen that most of the condition monitoring programs
were in place by the start of the study period, bearing in mind that the predominant condition
monitoring activities are manual inspections and tests, with the only outstanding techniques
by early this century being VA in the Upper and Mid Waitaki areas. Although there may not
appear to be any obvious correlation between failure rate and the Condition monitoring
program this may be reflective of the time it takes to realize the full benefit of these types of
programs.
One comment often made is the perceived correlation between the increase in Forced Outages
and major project work; the time line shows that the period of the Automation & Remote
Control project is also the period of highest failure rate, however this is also the period
preceded by outsourcing of maintenance activities where there was possibly feelings of
insecurity among staff, different contracting companies were competing for work, ECNZ
finished and Meridian Energy was formed and Meridian station management and controllers
were moving from being onsite at the various stations to being centralised.
The period from 2000 to current date is the period of greatest improvement, especially the last
year, it seems that this is typified by two factors. The first is the relative stability in field
maintenance staff and the second is the RCFA process, which began, in the robust form we
know as the Event Analysis Process at the start of this period.
Moving onto Figure 9.2 which shows very little correlation between beta value and the Mega
Watts the system produces, the number of machine starts, or the direct maintenance spend
(sourced from the CMMS). In fact the MW’s produced and the number of machine starts have
both trended up slightly and the direct maintenance spend has trended down slightly while the
failure rate has plummeted.
Other factors considered were the numbers of field staff and support staff in the maintenance
effort. Although it was to difficult to ascertain the exact numbers of staff in each type of role
over the study period it is generally felt from speaking with long standing staff that the
numbers of field staff has slightly reduced and the numbers of support staff, supervisors, asset
managers, engineers etc has stayed constant to slightly increased. One comment made and
supported by the author is that the ratio of field staff to support staff has changed over the
years, that is, for each hands-on field person, there tends now to be more, scheduling,
planning, technical support, and management for that person.
Although it is difficult to draw concrete conclusions from this high level view, the
improvement in failure rate over the last four years, which is most definitely evident, seems to
be a characteristic of improving long term stability for the skill sets involved, the continuing
development of condition monitoring and I believe most importantly the effort in determining
why the plant fails and then rectifying those failure modes. Whether major refurbishment

Page 18 of 22
projects have an affect is not clear due to the number of factors in play. However it seems
logical, based on the ‘Bath Tub’, curve that due to the type of work carried out during the
Automation & Remote Control project this must have contributed at least in part to the blow
out of the failure rate seen in this period and therefore it must have also contributed, due to
infant mortality and the introduced failure modes now passing, to the improving failure rate
we now have. The improving failure rate does not appear to have much to do with direct
maintenance spend, within reason, the amount of power the units produce or the number of
starts they are exposed to.
10.0 Failure Forecasting

10.1 Limitations
Predicting and modelling are concepts that have generated much attention, literature and
controversy in the reliability field. Reliability is not a parameter that is inherently predictable
on the basis of the laws of nature or of statistical extrapolation. It is useful that the extent of
uncertainty is always considered when attempting to model reliability5. It is therefore
important to realise that any forecast given via the Crow/AMSAA method is accepted
knowing that any change in the physical system being modelled or the management systems
being applied to the physical system will result in an altered outcome.
10.2 Why do we want to Forecast?

Given the limitations above a forecast will still give valuable information in respect to
budgeting. By forecasting the number of outages we expect to have over the next year for
example and combining this with the distributed time to repair an estimate of the affect
Forced Outages will have on availability can be made. Additionally by combining the
forecasted outages and the estimated average cost of a Forced Outage the total cost of
unreliability for the up coming year can be determined.
10.3 Forecasting Method

Many ratios of historic data to forecast period were fitted and extrapolated using the
crow/AMSAA model to determine the most accurate combination. All combinations suffered
the same failure to account for the cusps evident in the data, therefore unless the system is
relatively stable and in a period of relative consistency it is very difficult to forecast based on
just the historic data alone.
However the model produces very good reliability in forecasting the next failure event,
consistently predicting the event to within one day based on the previous six months data.
The forecasting of the number of Forced outages expected in the next financial year should be
done by considering not only the extrapolated figure from the Crow/AMSAA model but also
other factors such as any major refurbishment work planned and any changes in the systems
used to manage and control reliability.

Page 19 of 22
11.0 Conclusions
So in conclusion it is without doubt that the Crow/AMSAA reliability growth plots deliver
much more value in respect to tracking reliability than simply counting the number of forced
Outages. The three parameters, beta, cumulative failure rate, and instantaneous failure rate
alone give a clearer picture than a single count. The method models the outcome of the
maintenance effort allowing a comparison between planned and achieved reliability showing
trends and tracking progress.
What the examination of the observed failures rates show is a number of points.
Over the examined period on average, as a whole reliability is tracking at a steady
unchanging rate.
But, the data shows that the rate was deteriorating up to the turn of the century and
since then has be making large year on year improvement
Even with what appears to be robust maintenance practices in place, the period 95’ to
00’, for example other factors can undermine any maintenance effort.
Combining this information with the cost of each Forced Outage, shows that if there had
being no improvement of the performance from the 1999 – 2000 period we could have
expected to see as much as 150 extra Forced outages in the last year alone, the reduction in
Forced Outages that has been achieved equates to saving of $975,000 / year. This is
graphically represented in Figure 11.1. Moreover it seems that this saving has been made not
by spending more money in direct preventative measures but by utilising the existing
resources in different ways. This is born out by the fact that even though the generating output
from the fleet and the number of starts those units were asked to perform have increased, the
number of staff and the direct maintenance spend has remained constant or dropped. This
outcome mirrors well the concept of reliability growth management, which is defined as,
“The systematic planning for reliability achievement as a function of time and other
resources, and controlling the ongoing rate of achievement by reallocating of resources based
on comparison between planned and assessed reliability values”.

Page 20 of 22
Savings
Figure 11.1
If further reduction in the number of Forced Outages is desired, bearing in mind the potential
savings, then any improvement can either be made by further controlling the ongoing rate of
achievement by reallocating of existing resources based on comparison between planned and
assessed reliability values or by a greater direct spend on maintenance. However any program
or physical change to plant designed to reduce the failure rate needs to be done for less cost
than the combined value of the ongoing Forced Outages it is expected to save.

Page 21 of 22
Acknowledgements
• Mark Cain, Meridian Energy, Team Leader Asset Coordinators Mid-Waitaki
• Terry Smith, Meridian Energy, Business Analyst
• Kelvin Jopson, Team Leader Asset Coordinators, Upper -Waitaki
References
1. Abernathy, R ‘The New Weibull Handbook, Forth Edition’, Robert B Abernathy 2000
2. Broeman W, et al, ‘Technical Report No. TR-652 AMSAA Reliability Growth
Guide’, AMSAA 2000
3. ‘Military Handbook – 189 Reliability Growth Management’, US Department of
Defence 1981
4. Barringer P, ‘Problem of the Month, Nov 2002 – Crow/AMSAA Reliability Growth
Plots’ www.barringer1.com, 2003
5. O’Connor, P ‘Practical Reliability Engineering, Forth Edition’ Wiley, 2002

Page 22 of 22

Crow AMSAA PDF

Uploaded by

Copyright:

Available Formats

Crow AMSAA PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Crow AMSAA PDF

Uploaded by

Copyright:

Available Formats

Crow/AMSAA Reliability Growth Plots

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

3.2 Meridian Energy’s Generation System

16th Annual Conference 2005- Rotorua

4.0 The Duane Crow/AMSAA Model

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

Taking natural logarithms of the equation yields;

Which implies the equation above is a straight line on logarithmic paper.

16th Annual Conference 2005- Rotorua

4.3.1 Further Developments

5.0 Crow/AMSAA Software

16th Annual Conference 2005- Rotorua

6.0 Failure Data & Definitions

6.2 Period of Analysis

6.3 Data Sets

16th Annual Conference 2005- Rotorua

7.2 Forced Outage cost build-up

16th Annual Conference 2005- Rotorua

8.0 Meridian Energy’s FO data as Crow/AMSAA plots

16th Annual Conference 2005- Rotorua

8.3 Last 5-Year Plot

16th Annual Conference 2005- Rotorua

8.4 Year by Year Plot Figure 8.4

16th Annual Conference 2005- Rotorua

9.0 Organisation Operational History & Correlation with Failure Rate

16th Annual Conference 2005- Rotorua

1996 1997 1998 1999 2000 2001 2002 2003

1996 1997 1998 1999 2000 2001 2002 2003

1.1.2 9.2 Operational Parameters & Beta Value Comparison

GEG7144 Reliability Applications Unit Advisor Emile Eerens

16th Annual Conference 2005- Rotorua

10.0 Failure Forecasting

10.2 Why do we want to Forecast?

10.3 Forecasting Method

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

16th Annual Conference 2005- Rotorua

You might also like