Improving Asset

Performance Using
Statistical Methods
Improving Asset Performance Using Statistical Methods

Copyright© Meridium, Inc. 2004

All rights reserved. Printed in the U.S.A.

All rights including reproduction by photographic or electronic process and translation into other
languages of this material are fully reserved under copyright laws. Reproduction or use of this material in
whole or in part in any manner without written permission from Meridium, Inc. is strictly prohibited.

Meridium is a registered trademark of Meridium, Inc. All trade names referenced are the service mark,
trademark, or registered trademark of the respective manufacturer.

MERIDIUM | Improving Asset Performance Using Statistical Methods 1

Introduction

Event data analysis can be a very useful in understanding how and when assets fail. It can also provide
insight to help plant personnel understand what action to take and when to take it. In other words, it
helps in the process of building a strategy for asset performance management (APM).

This paper will cover several data analysis techniques that can be used to help understand the
dynamics of asset failure. These methods are as follows:

• Pareto Analysis
• AMSAA Growth Analysis (MTBF Trending)
• Distribution Analysis (e.g. Weibull Analysis)
• System Reliability Modeling

Before we can discuss the use of these analytical tools we need to first address the data inputs. In
order to use data analytics you must consider where your data is coming from and how valid it actually
is. Primarily when using the analytics mentioned above, we will be using failure event data. This
should bring up the question, what is a failure? Depending on whom you ask you get a different
answer. For instance, some may say that it is only a failure when there is a production loss. Others say
the asset has to have a complete loss of function; therefore partial loss of function would not count.
The fact is, for this type of data analysis we need to have a common definition of failure.

Many of our customers use a definition that encompasses the following three levels of consequence:

• Complete Loss of Function

• Partial Loss of Function
• Potential Loss of Function

Let’s explore some examples and reasoning for utilizing such a broad definition. Consider that you
have a centrifugal pump that is designed to pump 200 gpm. If that pump had a component failure (e.g.
bearing failed) and the pump could not pump at all that would certainly be considered a failure. Now
suppose that for some reason that pump had a defect that only allowed the asset to pump 100 gpm. Is
that pump meeting its intended function? Probably not. Lastly, suppose that that pump had a defect
but did not cause a complete or partial loss of function (e.g. high vibration). When the decision is
made to take the asset out of service to correct the problem, is that not a failure? Again, it is indeed a
failure of the asset to perform to its intended function. If there were not a defect there would be no
reason to take the asset out of service!

The point is that you need to know when assets and components do not perform their intended
function. By having a very restrictive definition of failure you miss valuable data about the
performance of individual assets.

Once the definition of a failure event is in place, you have to determine what data is important to fully
describe the failure event. For instance, when a failure event occurs you need to know things like
when the failure event occurred, what component failed, the failure mode, maintenance activity (e.g.
replace, repair, inspect) and many others. Below is a table and description of critical information that
is important to collect for any failure event.

MERIDIUM | Improving Asset Performance Using Statistical Methods 2

Identification History Dates Consequence
Event ID Functional Loss Event Date Maintenance Cost
CMMS ID Functional Failure (ISO Failure Mechanically Unavailable Production Cost
Mode) Date/Time
Functional Location Effect Mechanically Available
Date/Time
Functional Location Hierarchy Maintainable Item Mechanical Downtime
Level 1 (Site) Condition Maintenance Start
Date/Time
Level 2 (Area) Cause Maintenance End
Date/Time
Level 3 (Unit) Maintenance Action Time to Repair
Level … Narrative
Level n (System)
Equipment ID
Equipment Name
Equipment Category (Rotating)
Equipment Class (Pump)
Equipment Type (Centrifugal)

Event ID - This is the unique identifier for each failure event.

CMMS ID – This is useful if you are using a CMMS system as the base data collection system for failure
events.

Functional Location - The functional location is typically a "smart" ID that represents what function
takes place at a given location. (Pump 01-G-0001 must move liquid X from point A to point B)

Functional Location Hierarchy - Functional hierarchy to roll up metrics at various levels

• Level 1
• Level 2
• Level 3
• Level …
• Level n (System)

Equipment ID - The Equipment ID is usually a randomly generated ID that reflects the asset that is in
service at the functional location. The reason for a separate Equipment ID and Functional Location is
that assets can move from place to place and functional locations

Equipment Name - Name or description of Equipment for Identification purposes

Equipment Category (e.g. Rotating) - Indicates the category of equipment the work was performed on.
Generally by discipline (Rotating, Fixed, Electrical, Instrument)

Equipment Class (e.g. Pump) - Indicates the class of equipment the work was performed on. Failure
Codes can be dependent on this value

Equipment Type (e.g. Centrifugal) - Indicates the type of equipment the work was performed on.
Failure Codes can be dependent on this value

MERIDIUM | Improving Asset Performance Using Statistical Methods 3

Functional Loss - This indicates whether the equipment experienced a functional loss as part of this
event. A functional loss can be defined as any of the following three types: (1) Complete Loss of
Function, (2) Partial Loss of Function, (3) Potential Loss of Function

Functional Failure (ISO Failure Mode) - Basically the symptoms of a failure if one has occurred. Any
physical asset is installed to fulfill a number of functions. The functional failure describes which
function the asset no longer is able to fulfill.

Effect - The effect of the event on production, safety environmental, or quality

Maintainable Item - This is the actual component that was identified as causing the asset to lose it
ability to serve. (e.g. bearing)

Condition - This indicates the type of damage found to the maintainable item. In some cases this also
tends to indicate failure mechanism as well.

Cause - The general cause of the condition. This is not the root cause. It is recommended to use RCFA
to assess root causes.

Maintenance Action - Corrective action performed to mitigate the damaged item

Narrative - Long text description of work and suggestions for improvements

Event Date - This is the date that the event was first observed and documented

Mechanically Unavailable Date/Time - This is the date/time that the equipment was actually taken
out of service either due to a failure or the repair work.

Mechanically Available Date/Time - This is the date/time that the equipment was available for service
after the repair work had been completed.

Mechanical Downtime - Difference between Mechanically Unavailability Date and Mechanically

Available Date (in hours)

Maintenance Start Date/Time - This is the date/time that the equipment was actually being worked on
by maintenance.

Maintenance End Date/Time - This is the date/time that the equipment was actually finished being
worked on by maintenance.

Time to Repair - This is the total maintenance time to repair the equipment

Maintenance Cost - This is the total maintenance expenditure to rectify the failure. This could be
company or contractor cost. This cost could be broken out into categories such as Material, Labor,
Contractor, etc.

Production Cost - This is the amount of business loss associated with not having the assets in service.
This cost includes Lost Opportunity, when an asset fails to perform its intended function and there is no
spare asset or capability to make up the loss.

Once the work process is in place to collect the data, it is then possible to begin analyzing it. There
are many ways to analyze data. From very simple methods like Pareto to more sophisticated methods
like trending and distribution analysis. Do not mistake simple as not useful or effective. For instance,
Pareto analysis is a very simple data analysis technique but it is extremely useful and valuable.

MERIDIUM | Improving Asset Performance Using Statistical Methods 4

Pareto Analysis
The Pareto Principle or the 80/20 rule as it is sometimes referred was develop by an Italian Economist
in the early 20th century. The principle was first based on income distribution. In other words, a very
small portion of the population holds the majority of wealth. Since then, the principle has been
applied in many different ways. In manufacturing, we use it to track the defects or failures that occur
and their overall importance. For instance, you will typically find that 20% or less of your assets
represent 80% or more of the losses within a typical facility. This can be represented financially or by
number of events.

This type of analysis has been used successfully at most industrial sites for the purpose of determining
where the most significant problems and defects are occurring in the facility. Once they are
identified, they are further analyzed with other tools and methodologies like Weibull Analysis, Root
Cause Analysis and others.

Below is a sample Pareto Analysis:

This Pareto chart shows the top 30 centrifugal pump failures by the total cost each has experienced. Within each pump
asset you can see the component or maintainable item that failed in the event.

MERIDIUM | Improving Asset Performance Using Statistical Methods 5

The basic workflow for developing a Pareto analysis is as follows:

1. Determine your measures. For instance, what is your measurement criterion (e.g. Total
maintenance cost? Total number of failures? Total lost profit opportunity? etc…)
2. Determine your x-axis dimensions. What do you actually want your measures plotted against?
For example, a common Pareto would measure the total cost of maintenance plotted against
individual equipment types. You can see measures plotted against many different types of
dimensions. Below is a list of common dimensions that can be used for a reliability Pareto
analysis.

• Unit • Maintainable Item

• Equipment Category • Failure Mode
(e.g. Rotating) • Cause Type
• Equipment Group • Date (Year, Month)
(e.g. Pump) • Failure Type (Total,
• Manufacturer Partial or Potential
• Equipment Location Loss)
• Equipment ID

3. Sort the data in descending order from highest importance to lowest importance
4. Plot the data in a histogram chart.

There are a number of effective tools to assist with Pareto analysis. It can be as simple as using a
spreadsheet program like Microsoft Excel® or as sophisticated as using an OLAP (On-Line Analytical
Processing) engine like the one that comes with Microsoft SQL Server®. Microsoft Excel® offers
excellent data manipulation tools as well as a very sophisticated engine for performing Pivot Tables.

Sample Pivot Table in Microsoft Excel®

MERIDIUM | Improving Asset Performance Using Statistical Methods 6

 Sample Pareto chart based on a Pivot Table in Microsoft Excel®

OLAP or On-Line Analytical Processing is a sophisticated tool to help analysts easily mine or drill down
into their data. OLAP allows for the development of multi-dimensional cubes that analysts use to drill
down from one x-axis dimension to the next lower level x-axis dimension. Below is an example of the
drill down capability in Meridium utilizing OLAP technology. The following series of graphs presents the
failure cost of a refinery by production unit. By clicking on any of the units, the analyst is presented
with the next level of detail. In this case, the asset IDs within the selected unit with the highest
failure cost. By clicking on any of the asset ID’s the analyst is then presented with detailed information
about the Asset subunit or failed item.

OLAP Graph – Refining Units

MERIDIUM | Improving Asset Performance Using Statistical Methods 7

 OLAP Graph – Equipment ID’s within the Alkylation Distillation unit

OLAP Graph – Asset Subunit for PMP-4543, which is within the Alkylation Distillation unit

MERIDIUM | Improving Asset Performance Using Statistical Methods 8

AMSAA Growth Analysis (MTBF Trending)

Army Materiel Systems Analysis Activity or AMSAA Growth Modeling is an analytical tool to trend Mean
Time Between Failure or MTBF. This tool has multiple practical applications in the field.

The purpose of this tool is to plot MTBF data over time to determine if MTBF is increasing, decreasing
or remaining constant. For example, assume that you have 10 pumps in similar service. The plant
underwent a new predictive maintenance strategy on these pumps two years ago. AMSAA can be used
to trend the MTBF to determine the effectiveness of the new strategy. It can also be used in reverse as
well. Perhaps the maintenance department wants to create a new predictive strategy on specific
equipment and needs to demonstrate that MTBF has been trending down for a significant period of
time. This helps make the case to create a new equipment strategy based on past MTBF performance.

The second main purpose for using AMSAA is to determine the validity of conducting a distribution
analysis on specific failure modes. Weibull distribution analysis is used to help determine the pattern
of failure an asset experiences for a specified failure mode. In order to perform distribution analysis it
is important to make sure that MTBF is not trending significantly higher or lower as time goes by. In
order to use distribution analysis effectively, there has to be a constant failure rate to ensure that the
data will provide a good “fit”.

Let’s take a look at an example dataset that might be used to perform AMSAA growth analysis.

The dataset is based

on 24 failure events
for two force draft
fans in similar
service. Note that
this data has mixed
failure modes which
are acceptable with
a growth analysis as
opposed to a
Weibull analysis
where it would not
give accurate
results.

AMSAA Growth typically

offers two parameters.
The Beta and Lambda
values.
Beta is the slope of the
growth plot. Similar to
Weibull, but growth
looks at cumulative time
to event (failure) and
not just individual times

Lambda is the
Scale parameter
It equals the y-intercept
of the growth plot at
time t=1

MERIDIUM | Improving Asset Performance Using Statistical Methods 9

 The MTBF plot provides
a visual representation
of the MTBF as it has
performed over a
particular time period.
It this case the MTBF for
these two fans has
trended positively over
time indicating an
improvement in overall
reliability.

Weibull Distribution Analysis

Weibull distributions have been used effectively to help determine the pattern of failure that a specific
component experiences for a specified failure mode. In addition to identifying the failure pattern it
also provides an accurate assessment of the characteristic life of the component for the same failure
mode.

The failure pattern is a very important factor when determining what type of strategy to employ for a
given component. The pattern of failure is based on the ‘reliability bathtub curve”

Early Wear-out
Failure Failure
Failure Rate

Useful Life
(Random)

Time
Reliability Bathtub Curve

The three patterns represented by the “reliability bathtub curve” are:

• Early Failure or “Infant Mortality”

• Useful Life (Random Failure)
• Wear-Out Failure

MERIDIUM | Improving Asset Performance Using Statistical Methods 10

Early failure indicates that the component has a higher likelihood of failure early in its life than later.
In other words, it will likely fail close to the time it comes on line or into service. If it survives the
initial startup period it will likely have a long life. This type of pattern is common for certain types of
equipment like electronics. For instance, if you purchase a TV and it works for the first few weeks it
will likely fail due to obsolesce rather than a specific failure. This pattern is also indicative of
personnel avoidable problems like poor workmanship, incorrect startup procedures and other personnel
avoidable issues. Many studies have shown than most failures occur in this pattern, which clearly
shows that there is a lack of knowledge and skill in the maintenance and operation of our assets. When
experiencing an infant mortality problem, it does not make sense to do planned or time based
replacement maintenance, as it will only increase the chance of failure when the component comes
back on-line.

Random failure patterns indicate that time is not a factor in our failures. For instance, a component
may fail at 10 days, 100 days or 1,000 days. The probability is the same for each of the time periods.
Therefore planned replacement maintenance is not effective in this type of situation. Since time is not
a factor in the failure there is no obvious time to do the planned replacement.

Wear out failure patterns indicate that the component has a useful life and that time is definitely a
factor in the life of the component. For instance, piping corrosion at an oil refinery would typically
have a wear out pattern. This simply means that some period of useful runtime takes place before the
component begins to show signs of deterioration. Depending on the failure mechanism, the wear out
rate can be very rapid or very slow. This is measured as the time from the first identifiable defect to
the actual loss of the component.

Each of the failure patterns are represented by a parameter known in Meridium as the Beta value (β).
The beta value is simply a measure of the slope of the probability plot. Below is a table with Beta
value interpretations.

General Rules for Beta:

β<1 Indicates infant mortality
β=1 Indicates random failure
1<β<4 Indicates early wear-out failure
β>4 Indicates rapid wear-out failure

MERIDIUM | Improving Asset Performance Using Statistical Methods 11

 Early Wear-out
Failure Failure
Failure Rate
β<1 β>1
Useful Life
(Random)
β=1

Time

“Reliability bathtub curve” with beta value interpretations

Weibull distributions also measure the characteristic life of a component at the point in time where
63.2% of the population in a given dataset has failed. This value is represented in time (e.g. hours,
days, years, etc.). The parameter for this measurement is called the Eta (η)

Weibull Distribution Workflow

There are several steps you must go through to complete a Weibull distribution. These steps are also
known as a workflow. Before is an example of a Weibull distribution workflow.

1. Determine the asset(s) that you would like to analyze

2. Determine the component/failure mode for that asset(s)
3. Collect the dates when the failures have occurred
4. Determine the TTF or Time to Failure values
5. Determine the Eta and Beta values for the supplied TTF values
6. Determine if the data provides a “good fit”
7. Determine next steps (PM optimization, Failure Probability, Root Cause Analysis)
8. Take action on results!

Step 1 - Determine the asset(s) that you would like to analyze

Weibull analysis can be used on multiple assets although it is important to make sure that the assets
are similar in design and in the service they provide. For instance, you do not want to perform Weibull
Analysis in a reciprocating pump in a water service and a centrifugal pump in hydrocarbon service. The
failure modes and rates could be significantly different and consequently will provide inaccurate results
in the analysis. So the general rule is to select a single asset or multiple assets if they are similar in
design and service.

Step 2 - Determine the component/failure mode for that asset(s)

Once the asset is identified, you must isolate the component/failure mode because combining multiple
components/failure modes will cause inaccurate results and many times will create a failure pattern of
random due to the different failure rates of each component/failure mode.

MERIDIUM | Improving Asset Performance Using Statistical Methods 12

Step 3 - Collect the dates when the failures have occurred

Determine the dates of each failure that has taken place. This is best described as the date that the
asset was unavailable to perform its intended service. For instance, this is the date that operations
took the asset out of service and made it available for maintenance to make repair.

Step 4 - Determine the TTF or Time to Failure values

This is the time between the first failure date to the next failure date. For example if you experienced
a failure on 03/04/1996 and the next failure date was on 5/10/1998 than the TTF value would be 797
days.

5/10/1998 – 03/04/1996 = 797 Days

Step 5 - Determine the Eta and Beta values for the supplied TTF values

The TTF values must be processed utilizing the Weibull calculations to determine the Eta and Beta
values. This can be done either manually utilizing a manual method utilizing special Weibull graph
paper to more sophisticated tools built especially for performing Weibull analysis. They can also be
done to some degree utilizing a generic analytical tool like Microsoft Excel®.

Step 6 - Determine if the data provides a “good fit”

There are several methods to determine the goodness of fit for your analysis results. Common fit tests
might include the following:

• Kolmogrov-Smirnov
• R-Squared Values

Step 7 - Determine next steps (PM optimization, Failure Probability, Root Cause Analysis)

The output of a Weibull analysis will help to determine if time-based replacement is a suitable strategy
for a particular component. It will also help you to determine what the most cost effective time
interval should be for wear out failure patterns. In addition to determining the appropriate time based
strategy, a Weibull distribution will allow you to determine when a failure might occur so that proper
proactive action can be taken to avoid the secondary failure. A common result of a Weibull analysis is
to conduct a disciplined Root Cause Analysis (RCA). In many cases, the Weibull will indicate that a
problem exists that is uncharacteristic for a particular component. An RCA is useful in determining the
underlying causes that might be attributing to the poor component performance.

Let’s take a look at a practical example. We have a fan that has had several bearing failures. To
perform this analysis you have three potential tools to perform your analysis:

• Weibull graph paper (manual)

• Spreadsheet
• Weibull Software

For visual clarity we will use a Weibull software tool to perform our analysis.

MERIDIUM | Improving Asset Performance Using Statistical Methods 13

 TTF values are derived by subtracting
the first failure date for an asset from
the next subsequent failure date.

The analysis determines

that the Beta value is
4.5937, which is indicative
of a rapid wear out
pattern. In other words,
when the bearing begins to
first show signs of
deterioration, it rapidly
progresses to a secondary
failure. The Eta value
shows that the
characteristic life of these
bearings is 104.2417 days.

MERIDIUM | Improving Asset Performance Using Statistical Methods 14

 The cumulative distribution
Function shows time on the
x-axis and probability of
failure on the y-axis and in
this case it has reliability
plotted on the z-axis. To
interpret this graph, you
can look at any point along
the estimate curve and
draw a horizontal line to
the y-axis and a vertical
line to the x-axis to
determine the probability
of failure at a give time
period. You can also see
from the graph that this is
a wear failure due to the
fact that there is no
chance of failure until
about the 27th day of
operation.

The Probability Plot is the

plot most often used when
conducting a Weibull
analysis. It takes the data
from the Cumulative
Distribution Function and
plots it onto Log paper to
create a straight line. The
interpretation is similar to
the Cumulative
Distribution Function.

MERIDIUM | Improving Asset Performance Using Statistical Methods 15

 Weibull is often used to
determine the optimal
replacement period for a given
component. This is only valid
when a wear out pattern exists
for the given component (e.g.
Beta > 1)

This plot shows that if the

unplanned repair cost for this
component is $10,000 and the
planned cost is $3,500 then the
recommended replacement
interval is 65.14 days.

The Weibull allows the analyst

to easily determine when
another failure might take
place. For instance, if FDF101
and FDF102 are both in service
for 24 and 45 days respectively
then the likelihood of not
surviving another 50 days is
18.62% and 46.83%
respectively.

Weibull analysis has been used successfully in a number of industrial applications. It has been
particularly useful for helping to determine which turnaround/shutdown to replace heat exchanger
bundles. For instance if you have a turnaround/shutdown coming up in two months and there will not
be another planned outage for five years after that then the analyst can review the bundle failure data
and run a Weibull analysis to determine if the bundle should be replaced in the upcoming outage or
what the probability of survival will be until subsequent planned outage. This has been successfully
applied at many refineries with dramatic financial and environmental benefits.

System Reliability Modeling

System Reliability Modeling takes component reliability (e.g. Weibull Analysis) to the next level. For
instance, a Weibull can tell you when you can expect another component failure on an asset but it
cannot tell you how the failure will affect the system in which that asset resides. That is where system
modeling comes into play. System modeling allows the analyst to draw existing and planned systems
with all of the combined assets and asset relationships (e.g. parallel vs series). In addition to defining
the physical relationships, system modeling also takes into account individual reliability calculations for
each asset/component.

MERIDIUM | Improving Asset Performance Using Statistical Methods 16

To take it one step further, system modeling can also take in financial data to determine what the
financial impact would be for a given system failure. This is often done with Monte Carlo simulation,
which enables the analyst to simulate many scenarios of a situation to see what the effects would be
on a system.

System Modeling Example

Suppose that in your plant you had a control valve, tank and two pumps in the following configuration.

P1
CV T

P2

Let’s assume that the P2 is full capacity spare for P1 and is only needed when P1 has a failure and
cannot perform its function.

The next thing we need to do is determine what the reliability is for each of the assets within the
system. In order to determine reliability you will need an accurate MTBF value for each asset. This
data can be acquired from plant information systems or from industry standards if plant data does not
exist. For simplicity we will use the exponential reliability distribution. This is the equivalent of using
the Weibull distribution when Beta is equal to 1.

R(t)=e(-λt)
Let’s take a look at the interpretation of this formula.

R(t) – Represents the probability of an asset/component to reach its specified mission time

e – Natural Logarithmic Base (2.718)

λ – 1/MTBF (Lambda)

t – Mission Time

MERIDIUM | Improving Asset Performance Using Statistical Methods 17

Let’s say you have an asset that has a MTBF of 900 days. What would be the probability of that asset
surviving until its specified mission time of 365 days.

R(t) = 2.718 –1/900(365)

R(t) = 2.718 –.4055
R(t) = .6666 or 66.66%
We will now use this calculation as the basis for our system reliability model. Let’s build a model of
the simple system defined above.

0.91

0.98 0.99
P1
CV T 0.81

P2

Reliability of series systems is the product of the individual component reliabilities:

Rs = R1 * R2 * ... Rn

Reliability of parallel systems is represented by the following equation:

Rs = 1- (1-R1) * (1-R2) *... (1-Rn)

Let’s determine what the reliability of the system is using the specified Reliability values defined in the
diagram.

RS = RCV * RT * 1 – [(1-RP1) * (1-RP2)]

RS = 0.98 * 0.99 * 1 – [(1-0.91) * (1-0.81)]

RS = 0.95

This calculation determines that the overall reliability of the system is .95 or 95%. This means that it
has a 95% chance that there will be not be a failure that will take the entire system down during the
mission time. This is due in large part to the redundancy built into the system on the pump trains.
Neither pump train has a reliability value greater than 95% but since they run in parallel we can easily
see the increase in system reliability.

As with Weibull Analysis, we have a variety of tools to help us to perform this type of analysis. It can
be as simple as doing it by hand or using other software tools like Microsoft Excel® to using very
sophisticated tools designed specifically for this type of application.

MERIDIUM | Improving Asset Performance Using Statistical Methods 18

Case Studies

Marathon Ashland Petroleum

Marathon Ashland Petroleum (MAP), an $8 billion leader in refining, marketing and transportation
services uses statistical analysis to determine when a heat exchanger bundle should be replaced. They
performed the analysis at the Robinson, IL refinery to determine if the scope on heat exchangers would
be accurate for their upcoming turnaround.

They performed a series of statistical analyses and return on investment (ROI) calculations to
determine if they should replace bundles in the upcoming turnaround and compared it to the
subsequent turnaround. Based on the results of the analysis, they determined that they should spend
an additional $478,490 to replace 26 additional bundles that had a high probability of failure within the
time period between the upcoming and subsequent turnarounds. MAP estimated that they would avoid
nearly $3 million in lost profit opportunity by taking a more proactive reliability position and replace
the bundles instead of risking the greater probability of failure in between turnarounds.

MAP has now institutionalized the heat exchanger analysis as a corporate best practice and is now
performing the analysis at their other 6 refineries in the United States.

ChevronTexaco
ChevronTexaco wanted to investigate the current reliability of their electrical equipment at their
Pascagoula, Mississippi refinery. They built system reliability models of 13.2 kV Substations to
determine what options they had to improve reliability, maintainability, operability with an emphasis
on the overall simplicity of the configuration.

The analysis team was able to simulate many different configuration possibilities and had the cost
benefit for each scenario. The scenarios simulated doing nothing against the benefit of doing a
complete redesign of the system. Sam Preckett, reliability focused maintenance project leader for
ChevronTexaco, determined that "ChevronTexaco will be able to avoid approximately $9 million in
future lost profit opportunity by using Meridium System Reliability.” He went on to say that “we’re
now preparing to use the product (System Reliability Modeling) for all applicable projects across our
enterprise.”

Summary
Statistical Reliability methods are applicable to use in heavy process manufacturing. The important
thing to note in working with statistical methods for reliability is to make sure the base data accurately
reflects the situation within the facility. There are many training courses on the topic as well as an
array of tools and techniques to help get you started. For more information on this subject please
contact Ken Latino, Meridium Senior Asset Performance Consultant at (540) 344-9205 ext. 1176 or
klatino@meridium.com.

MERIDIUM | Improving Asset Performance Using Statistical Methods 19

 Corporate Offices
10 South Jefferson Street
Roanoke, Virginia 24011
540.344.9205
540.345.7083 fax

Regional Offices
11200 Richmond Avenue
Suite 670
Houston, Texas 77082
281.920.9616
281.920.9190 fax

www.meridium.com
info@meridium.com

MERIDIUM | Improving Asset Performance Using Statistical Methods 20