Demystifying Weibull

A Practical Means to Locate Performance in the Bathtub Curve

Joe Sipper
9/13/2016

This document does not contain technology or Technical

Data controlled under either the U.S. International Traffic in Copyright © 2014 Raytheon Company. All rights reserved.
Arms Regulations or U.S. Export Administration Regulations Customer Success is Our Mission is a registered trademark of Raytheon Company.
Scope of Presentation
• Intent is foundational, discuss a focused facet of Weibull
• This is about plotting data and identifying phases in the
bathtub curve, it’s not about further Reliability analysis
• Why all these details if Minitab does all the work?

• Key discussion points include:

1. Weibull distribution plotting
2. Data censoring and test type
3. Types of repair to the population
4. Creating proper data structure for Minitab
5. Weibull interpretations and the bathtub curve
6. Process for Weibull plots
7. Using Weibull plots to gain performance insight
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 2
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1. Weibull Distribution Plotting
a) The big picture of today’s presentation

b) Scope

c) Overview: analytical flow

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 3
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1a1. The Big Picture: Today’s Presentation
• There are lots of great Weibull applications to predict system life,
reliability, success rates, preventive maintenance intervals,
warranty structures, repair strategies, etc
• The same source data can be used for everything we’ll discuss
however the format needs to be in a format appropriate for each
assessment
• Today’s outputs can feed into some Weibull applications but that’s
for another day
• Today is about structuring failure data, analyzing it in Minitab, and
making first-tier conclusions about where in the bathtub a
performance falls

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 4
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1a2. The Big Picture: Characterize Performances
• Understanding the problem is often the hardest part of the
problem; getting and arranging good data is a much easier for a
problem that is well understood than for one that is not

• Determine what distribution best models performance

• For the most part, the analytical process is similar or identical
across many scenarios but the data arrangement differs
• Data drives a model, and is then replaced by its model
• The Weibull model, which is highly flexible and often applicable,
gives insight into performance phases of the bathtub curve

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 5
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1a3. The Big Picture: Use Weibull for Systems?
• A lot depends on the question, i.e., the needs of the assessment

• Some believe that Weibull can be used only for an individual failure
mode, and multiple failure modes offset performance characteristics
and muddy the analysis
• Others believe a system is simply a series of components
• System fails when one (random) component fails
• System time-to-failure = smallest of the component failure times
• This is equivalent to Weibull distribution
• Underlying premise to, but not explored in, today’s presentation
• Ideally run a unique Weibull assessment for each failure mode,
however the method given here addresses needs of the assessment

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 6
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1b1. Distribution Plotting for this Presentation
• Scope focuses on expectations of performance when the system or
element is repaired, and the corresponding data structures
Non-Repairable Repairable (System or Element) Repairable
(System or Element)  analyze patterns b/w failures (Fleet)

Typical Single failure mode Individual system based Individual system Fleet based on
characterization based on sample data on field data based on field data field data

Repair state New Like New Minimal Mixed

s.i.i.d. ? Yes Yes No No

Element fails vs. System failures vs. System failures vs. Fleet failures vs.
Data analyzed element usage system usage system usage fleet usage

Ordered usage from Ordered usage from Superimposed Cumulative

Data timeline T=0 T=0

s.i.i.d. = statistically independent and identically distributed

Ultimately make a set of usable, value-added data with appropriate groupings.

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 7
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1b2. Repair Type and Data Structures
Timelines shown on this slide are in terms of each system studied
Replace “Like New” Repair “Minimal” Repair
• Brand new • Expectation is that the next failure • Next failure depends on
hardware is random (independent) because current usage, i.e., the system’s
restarts the history is reset, i.e., the repair age prior to the repair
the clock returned a system that was as good
Data structure: as new • System Reliability is the same
ordered usage
Data structure:
after the repair as it was
from T0 = 0
t1 ordered usage t1 before the repair
I t X from T0 = 0 I t X
2 2 Data structure:
It X It X superimposed
I tX 3
I tX3

t1 t2 t3 t4
I t X I t X . t.n . X
4 4

I XT XT XT XT
I X n
I X n T0 = 0 1 2 3 4 Tn

I XXX
T =0
I X XX
T =0
0 0
t1 t2 t3 t4
Note the differences in cumulative timelines: I XT XT XT XT . t.n . X
T0 = 0 1 2 3 4 Tn
I XXX
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 8
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1b3a. Fleet vs System
• Fleet failures and fleet time can stay random over time
because of the mix of old and new systems in the fleet
• Failures for each individual system are tracked and composed
additively on a cumulative timeline
• Cumulative timeline combines failures from different systems
onto a single timeline

• Individual systems each provide usage data, and

clustering of events becomes evident at similar usages
• Failures are superimposed onto a timeline
• Both use Power law statistically, but treat data differently

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 9
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1b3b. Fleet vs Repairable Systems
Repairable Systems – IX X
System 1 XX X
Systems age in patterns, characteristics are evident
when viewing clustering of superimposed failures
System 2 I X X X XX
System 3 IX X XXX
NOTE: In repairable systems, failures for non-
repairable elements are s.i.i.d. however failures for System n I X X XX
the system or element are dependent
XXX X X X X XXXXXXX
Superimpose IX XXX

IX X XX X System 1 Fleet – Failures become random when

I X X X XX System 2 time modeled because the fleet is
composed of systems both new and
IX X XXX System 3 old, i.e., assumption of independent
failures is not valid
IX X XX System n

IX X XX X X XX XX X X XXX X X XX
Cumulative Timeline

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 10
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
1c. Overview: Analytical Flow
Data
Structure
Yes Censored
Data ?

Test
No (Test) and (Data)
Type Censoring Suspensions
Analyze as Analyze as
Considerations
Censored Data Complete Data

Replaceable = New =
MTTF
Repair
Non-repairable Right Censored – lower bound known,
or Structure upper bound not fully known for all data
(MTTF)
MTBF
? Left Censored – only have the upper
Analyze as Repairable bound, not sure of exact time of failure
MTTF (MTBF)
Interval Censored – all fail, have ideas of
Minimal Fleet times but not exact measures
Expectations
of Repair
Superimpose Cumulative (Test) Type I: Time Censoring – test time
each System Timeline fixed
New or
Like New
(Test) Type II: Failure Censoring –
Identical Ordered Usage number failures fixed
conditions from T=0
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 11
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2. Data Censoring and Test Type
a) Complete data
b) Left censored data also called Interval data
c) Right censored data also called Suspended data
d) Singly censored data
e) Multiply censored data
f) Interval, or grouped, data
g) Test Type I also called Type I Censoring
h) Test Type II also called Failure Censoring

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 12
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2a. Complete data
• All units run to failure, have exact usage times for each

• Another way to say this is that each unit has run through
its entire lifetime and is characterized exactly

• Examples include: Complete Data

I Fail
• Highly structured
Sample 1 of n

Sample 2 of n I Fail
lab testing Sample 3 of n I Fail

Sample 4 of n I
• Fully accessible
Fail

Sample 5 of n I Fail
field data with high
...

...

failure rates Sample n of n I Fail

Usage (Time, Cycles, Distance, etc.)

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 13
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2b1. Left Censored data
• Check performance at some time, t

• Observe the unit(s) have failed, do not know exactly when

• Examples include:
• Testing or field Left Censored Data
data where data Sample 1 of n I Fail

is not fully Sample 2 of n ] Fail, not sure of exact usage

accessible Sample 3 of n I Fail

Sample 4 of n I Fail

Sample 5 of n ] Fail, not sure of exact usage

...

...

Sample n of n I Fail

Usage (Time, Cycles, Distance, etc.)

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 14
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2b2. Interval Censored data
• Different from left-censored in that all intervals do not
start with zero usage
• Check performance at some time, t
• Observe the units have failed, do not know exactly when
• Examples include: Interval Censored Data
• Testing or field Sample 1 of n [ Fail in interval
]
data where data Sample 2 of n [ Fail in interval ]
is not always Sample 3 of n [ Fail in interval ]
Fail in interval
accessible Sample 4 of n [ ]
Sample 5 of n [ Fail in interval ]
...

...

Fail in interval
Sample n of n [ ]
Usage (Time, Cycles, Distance, etc.)

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 15
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2c. Right Censored (or Suspended) data
• Important to include usage of survivors in timelines

• Check performances at some time, t

• Observe some units have failed, while the remaining are

still operational; record all usage
Right Censored Data
Sample 1 of n Operational
• Examples include: Sample 2 of n Operational
• Most field data, Sample 3 of n I Fail
especially those Sample 4 of n I Fail
checked somewhat
regularly Sample 5 of n I Fail
...

...

• Time truncated = Sample n of n Operational

Type I, failure Usage (Time, Cycles, Distance, etc.)

truncated
Minitab Insights
= Type II
This document does not contain technology or Technical Data controlled under either the
Slide 16
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2d. Singly Censored data
• Only one set of samples throughout all observations

• Population started together, data collected before all fail

• Record all usage – impacts cumulative percent fail

• Examples include:
• Typical Type I or Type II testing in 2g and 2h

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 17
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2e. Multiply (muhl-tuh-plee) Censored data
• Different run times and different number of interventions
for each system

• More than one censoring point in a set of observations

• Record all usage – impacts cumulative percent fail

• Examples (CP = censor point) include:

• Field data for a mature repairable product
• Controlled testing
• Test for n = 100 units stopped at 20,000 cycles  CP1
• Remove n = 30 units, re-observe after 5,000 more cycles  CP2
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 18
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2f. Interval or Grouped data
• 2 applications
• Known individual data is plotted and recoded and grouped into
bins after the fact for analysis, e.g., Pareto groupings
• Observations are made periodically across a population, exact
failure times are unknown but exact quantities are recorded as
described in 2b2

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 19
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2g. Test Type I (or Type I Censoring)
• Test n samples, terminate after some predetermined usage
regardless of the number of failures
• Some samples may, and likely do, survive the test (test conditions
may be too harsh if all fail) and are coded as suspensions
• Test time is fixed, number of failures is variable
• Record all usage – impacts cumulative percent fail

• Examples include:
• Reliability growth testing – accelerations impose 10 years of accelerated life
on samples in 600 hours, then terminate test
• Able to obtain usage for each element or system in the population, cross-
reference with those systems or elements that have failed at the time of
the data collection
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 20
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2h. Test Type II (or Failure Censoring)
• Test n samples, terminate after some predetermined number of
failures regardless of the usage
• Apply when time is of the essence, guaranteed enough data to
carry out assessments with statistical significance
• Number failures is fixed, code survivors as suspensions
• Record all usage – impacts cumulative percent fail

• Examples include:
• Reliability growth testing – run test until value-added information is
available for, say, n = 30 failures
• Able to obtain usage for each system in the population, keep selecting
random systems from field population, viewing pass/fail status, until say n =
30 failures are identified
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 21
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
2i. Thought on Censoring (also called Suspensions)
• Implementation also depends on the question asked

• Field data has practical limitations, often obtainable only upon

failure; suspensions in the data must be recorded

• These types of performances often span a lifetime that includes

multiple failures for each system, i.e., system will fail it just hasn’t at
the time of this data collection (sometime have eternal survivors)

• Suspensions may have a different failure mode from that being

studied or may not have failed at all

• Field data is usually a snapshot, survivors at a data collection point

will eventually fail and be treated as appropriate in terms of a data
timeline
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 22
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
3. Types of Repair
a) Non-repairable system or element (Replaceable)
i. Replaceable = return to “new” condition

b) Repairable system
i. Repair to “like-new” condition  “reset” system history

ii. Minimal repair  fix only what is broken, retain historical

information

c) Repairable fleet
i. Treat individual systems as a single population
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 23
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
3a. Non-Repairable System or Element
• Must be replaced; impractical or unsafe to return to
reusable condition
• Hazardous materials
• Disposables
• Inexpensive products

• Each system life, production or repaired, begins at T=0

upon being placed into operation

• Number of failures, and time between failures, is random

• Data analyzed: system or element failures versus system

or element usage
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 24
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
3b1. Repairable System – Like New
• All elements’ usage resets upon repair
• Analyzing the data between failures is, in effect, the
equivalent of pulling all usage back to a common
chronological point, i.e., T=0
• Examples include:
• Overhauls, e.g., jet engines, dirt bikes, race cars, etc.
• Preventive Maintenance
• Customer expectations, viable or not, based on repair costs or
perceptions of failing quickly upon return

• Generally NOT true but sometimes useful to consider

• Data analyzed: system failures versus system usage
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 25
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
3b2. Repairable System – Minimal
• Minimal Repair – fix only what is broken, improve
nothing else

• Each elements’ usage retains its history upon repair 

the next failure is not independent, components
untouched in the repair have a cumulative life and the
next failure is influenced by this life

• Examples include:
• Most personal services, e.g., car repair, dentist, etc

• Data analyzed: system failures versus system usage

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 26
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
3c. Repairable Fleet
• Provide a service where the function is more important
than the means of accomplishing the function
• Fleet failures and fleet usage can stay random over time
because of the mix of old and new systems in the fleet
• Failures for each individual system are tracked and composed
additively on a cumulative timeline
• Examples include:
• “Big picture” questions relevant to fleet, not individual systems
• Rental cars
• Capable computers for a large workforce
• Data analyzed: fleet failures versus fleet usage
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 27
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4. Creating the Proper Data Structure for Minitab
a) The purpose here is to produce shape and scale
parameters and then replace the data with a model;
3-parameter Weibull includes a threshold parameter

b) Prepare as much relevant data as practical – the

necessity is for
• Appropriate operational descriptors of usage (e.g., hours,
distance, cycles, etc) for each failure being repaired
• Constructing the appropriate timeline and groupings

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 28
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4a1. Shape, Scale, and Threshold Parameters
• Shape parameter (used in 2- and 3-parameter Weibull)
• In simple words: an influential value, defined by the underlying
data, that drives the look (shape) of the Weibull plot
• More technically: Measure the rate of change in the unreliability
function over the usage, where both the unreliability and usage
are linearized using logarithms.
• Use median ranks to estimate unreliability for each failure
rank in the failure order − 0.3
• Median Rank =  Bernard’s approximation
number of failures+0.4

1
• xi = ln time to fail i yi = ln ln
1−median rank

• In application: β < 1  decreasing failure rate, β = 1  constant

failure rate, β > 1  increasing failure rate
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 29
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4a2. Shape, Scale, and Threshold Parameters
• Scale parameter (used in 2- and 3-parameter Weibull)
• In simple words: measure of the spread of the Weibull, the
usage from the first viable failure potentially occurring to 63.2%
units failing

• More technically: the characteristic life, which is the level of

usage at which 63.2% of the systems or elements will fail. 
t
−
1 − R = 1 − e−λt = 1 − e τ = 1 − 0.368 = 0.632 when t = τ

• In application: if scale is large the distribution is more spread

out, if scale is small the distribution is tighter; a scale of 10
implies 63.2% of population will fail in the first 10 hours, cycles,
etc. while a scale of 500 implies the failures will occur over a
much longer time
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 30
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4a3. Shape, Scale, and Threshold Parameters
• Threshold parameter (used in 3-parameter Weibull)
• In simple words: distribution’s shift from 0 usage; can be used if
failures do not start at t=0 but instead at some time +t or -t; the
parameter describes the threshold at which failure occurs (also
called shift, failure-free, or location parameter)

• More technically: provides an estimate of the earliest viable

time-to-fail; < 0 negative on x-axis, > 0 positive on x-axis

• In application: track ship date but it takes 30 days for customer

to install and use – this is a failure-free usage period; negative
values may imply failure occurred prior to collecting
performance data such as a quality escape, in transit, etc.
(Minitab defines threshold only for non-negative values)
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 31
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b0a. Prepare as much Relevant Data as Practical
• Key pieces of raw data include, but are not limited to:
• System identifiers such as part number, serial number, etc.
• Usage at the time of the repair (often given as cumulative)
• Other data useful for further characterizations (dates, etc.)
• Ensuring the data is relevant per specific assessment needs

• Rather than trending performances over measures of usage, as in

Paretos, group repairs chronologically by each system identifier

• Build a timeline valid for the application, either pull failures to T=0,
superimpose failures on a common timeline, or system
performances as appropriate

• Note: data accuracy can lead to conservative results, high integrity

data may mean discarding some information
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 32
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b0b1. Prepare as much Relevant Data as Practical
• Establish qualifying characteristics in raw data to decide whether
data is relevant or not – examples include:
• A repair disposition that a field observation is indeed a failure, valid per the terms of
the assessment
• The information is considered complete, the observation is closed
• Able to apply to identified element, failure mode, or date range as appropriate

• Rather than plotting overall quantities chronologically, use Excel

pivot table to group data by each individual system and study each
system chronologically
• Serial number is an obvious choice for identifiers
• Good to lay out the Pivot Table Design in Tabular layout (Pivot table tools > Design >
Report Layout > Show in Tabular Form)
• Also consider (Report Layout > Repeat All Item Labels)

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 33
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b0b2. Prepare as much Relevant Data as Practical

Sample of fictitious
data for illustration only
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 34
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b0b3. Prepare as much Relevant Data as Practical
Excel formulas to drive fields added to pivot shown on prior slide
• Good Record? (cell N8 and drag through column N)
=IF(AND(A8<>"",A8<>"(blank)",B8<>"",B8<>"(blank)",C8<>"",C8<>"(blank)",E8="Yes",F8<>"",F8<>
"(blank)",J8="Closed"),"yes","no")

• Use in Weibull? (cell P8 and drag through column P)?

=IF(O8<=0,"no",IF(COUNTIF(O8,"*data*")=0,"yes",IF(COUNTIF(O8,"*data*")=1,"no","")))

• Failure Free Hours (per structure of the data as presented)

• Replaceable (cell S8 and drag through column S)
=IF(O8<=0,"",IF(N8="yes",IF((A8<>A7),VALUE(F8),IF(AND(N7="yes",B8<>B7),VALUE(F8)-
VALUE(F7),"")),""))
• Minimal Repair (cell Y8 and drag through column Y)
=IF(P8="yes",F8,"")
• Fleet (cell AE9 and drag through column AE)
=IFERROR(AE8+S9,AE8) NOTE: also include in cell AE8 =S8

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 35
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b1a. Prepare as much Relevant Data as Practical
Example 4b1, System Repair State: New and Like New

• Usage for each element

or system put into
operation, whether new
or repaired, starts at T=0

Sample of fictitious • Use columns S, U, V, W of

Excel worksheet in

data for illustration Minitab

• Table implies complete

only data

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 36
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b1b. Prepare as much Relevant Data as Practical
Excel formulas for complete data for “New” or “Like New”
• Bin Grouped by 100_1 (cell Q8 and drag through column Q)
=IF(S8<=0,"",IFERROR(IF(S8<>"",ROUNDDOWN(S8/100,0)+1,""),""))

• Order Bin Size 100_1 (cell T8 and drag through column T)

=IF(ROW()-7<=MAX(Q:Q),ROW()-7,"")

• Grouped Cycles_1 (cell U8 and drag through column U)

=IF(T8<>"",T8*100,"")

• Occurrences Bin Size 100_1 (cell V8 and drag through column V)

=IF(T8<>"",COUNTIF(Q:Q,"="&T8),"")

• Cum Pct Bin Size 100_1 (cell W9 and drag through column W)
Since this is cumulative, initialize U8 with a different formula
=IF(T8<>"",IFERROR(V8/SUM(V:V),""),"")
=IF(T9<>"",IFERROR(V9/SUM(V:V)+W8,W8),"")

Data now structured for Replaceable analysis in Minitab:

Each usage to fail, grouped (binned) occurrences and cumulative percentage
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 37
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b2a. Prepare as much Relevant Data as Practical
• Example 4b2, System Repair State: Minimal
• Each system’s usage starts
at T=0, each failure and
usage is recorded, and
failures are superimposed

Sample of fictitious • Use columns Z, AB, AC, AD

of Excel worksheet in
Minitab
data for illustration
• Table implies complete
only data

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 38
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b2b. Prepare as much Relevant Data as Practical
Excel formulas for “Minimal Repair”
• Bin Grouped by 100_2 (cell X8 and drag through column X)
=IF(Z8<=0,"",IFERROR(IF(Z8<>"",ROUNDDOWN(Z8/100,0)+1,""),""))

• Order Bin Size 100_2 (cell AA8 and drag through column AA)
=IF(ROW()-7<=MAX(X:X),ROW()-7,"")

• Grouped Cycles_2 (cell AB8 and drag through column AB)

=IF(AA8<>"",AA8*100,"")

• Occurrences Bin Size 100_2 (cell AC8 and drag through column AC)
=IF(AA8<>"",COUNTIF(X:X,"="&AA8),"")

• Cum Pct Bin Size 100_2 (cell AD9 and drag through column AD)
Since this is cumulative, initialize U8 with a different formula
=IF(AA8<>"",IFERROR(AC8/SUM(AC:AC),""),"")
=IF(AA9<>"",IFERROR(AC9/SUM(AC:AC)+AD8,AD8),"")

Data now structured for Repairable analysis in Minitab:

Each usage to fail, grouped (binned) occurrences and cumulative percentage
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 39
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b3a. Prepare as much Relevant Data as Practical
• Example 4b3, Fleet Repair
• Each system’s usage starts
at T=0, is tracked
cumulatively, and each
system is then stacked
additively

Sample of fictitious • The bins are much larger

because the failure free
data for illustration cycles are cumulative

• Use columns AG, AI, AJ, AK

only of Excel worksheet in
Minitab

• Table implies complete

Minitab Insights data
This document does not contain technology or Technical Data controlled under either the
Slide 40
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b3b. Prepare as much Relevant Data as Practical
Excel formulas for “Fleet” (Bin sizes = 1000 in this example)
• Bin Grouped by 1000_3 (cell AE8 and drag through column AE)
=IF(AG8<=0,"",IFERROR(IF(AG8<>"",ROUNDDOWN(AG8/1000,0)+1,""),""))

• Order Bin Size 1000_3 (cell AH8 and drag through column AH)
=IF(ROW()-7<=MAX(AE:AE),ROW()-7,"")

• Grouped Cycles_3 (cell AI8 and drag through column AI)

=IF(AH8<>"",AH8*1000,"")

• Occurrences Bin Size 1000_3 (cell AJ8 and drag through column AJ)
=IF(AH8<>"",COUNTIF(AE:AE,"="&AH8),"")

• Cum Pct Bin Size 1000_3 (cell AK9 and drag through column AK)
Since this is cumulative, initialize U8 with a different formula
=IF(AH8<>"",IFERROR(AJ8/SUM(AJ:AJ),""),"")
=IF(AH9<>"",IFERROR(AJ9/SUM(AJ:AJ)+AK8,AK8),"")

Data now structured for Fleet analysis in Minitab:

Each usage to fail, grouped (binned) occurrences and cumulative percentage
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 41
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
4b4. Prepare as much Relevant Data as Practical
• Data structuring stops here due to time limitations
• Censoring can be considered by simply expanding the Excel table
to include a “censoring” column and also importing this
information into Minitab
• Consideration that this data set is complete
• Many data collection processes gather only failures

• For the purposes of the plotting here, each system will eventually fail (either
permanently or will be repaired) and that information will be added to future
assessments

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 42
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
5a. Weibull Interpretations and the Bathtub Curve

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 43
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
5b. Weibull Interpretations and the Bathtub Curve
• The bathtub curve is a compilation of 3 distributions, it is not
continuous
• Weibull can indeed be made to fit all 3 distributions of the
bathtub curve as three separate plots
• Shape parameter is the rate of change in failure rate over time
• Weibull characterization arranges data, plots a representation of
each failure, and determines the rate of failure rate over usage
along with the spread of failures via characteristic life

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 44
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6. Process for Weibull Plots
a) Start with the appropriate data

b) Produce frequency plots for failures

i. Occurrences
ii. Cumulative failure percentage

c) Use Minitab to select the best distribution for the data

d) Use Weibull parameters to create a model for the data

e) Determine times to failure based on the model

f) Final validity check of distribution selected

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 45
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6a1. Start with the Appropriate Data
• The analysis starts to become more systemic once the
appropriate data is established

Sample of fictitious data for

illustration only – note the differences
in occurrences and cumulative
percentages
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 46
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6a2. Start with the Appropriate Data
• Aside from censoring considerations, which may or may
not be influential but are not discussed at this time, the
analysis now becomes a repeatable process
• Censoring impacts cumulative percentage
• Minitab – Stat > Reliability/Survival > Distribution Analysis
(Right Censoring) > Distribution ID Plot > Censor

• Work through the process in the rest of section 6 with

one set of data
• Since the data has been structured per the needs of the
analysis, the specifics of the data are not further discussed

• Section 7 shows analytics that reflect each phase of the

bathtub curve
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 47
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6b1. Produce Frequency Plots for Failure Times
“New” or The scatterplot and Pareto
“Like New” Pareto Chart of Grouped_Cycles

1200 100
chart here look similar to the
Weibull distribution plot in
1000 80 7a of this presentation but
800 Use Minitab Assistant they are different.
Occurrences

60

Percent
600
> Graphical Analysis
> Pareto Chart 40
400
Scatterplot of Occurrences Bin , Cum Pct Bin Size vs Grouped Cycles_1
20
200 160 100.0%

140 Graph > Scatterplot > Simple

0 Occurrences Bin Size 100_1 0
80.0%

Cum Pct Bin Size 100_1

Grouped_Cycles 500 1000 1500 2000 2500 3000 120
3500 4000 Other Y variables X variables
Occurrences 541 255 152 82 47 38 23 22 49 Occurrences Grouped Cycles
Percent 44.7 21.1 12.6 6.8 3.9 3.1 100
1.9 1.8 4.1
Cumulative Pct Grouped Cycles 60.0%
Cum % 44.7 65.8 78.4 85.2 89.1 92.2 94.1 95.9 100.0
80
Multiple Graphs > Multiple Variables >
60 Overlaid on the same graph 40.0%
These are essentially the 40 Plot results with both variables on primary axis
same plot. The first 5 points 20
> Double click Y-axis 20.0%

in the scatterplot make the > Secondary > Assign secondary scale for Cum_Pct
0
first bar in the Pareto. 0.0%
0 2000 4000 6000 8000 10000 12000 14000
Grouped Cycles_1 - "New" or "Like New"
Variable
Occurrences Bin Size 100_1
Cum Pct Bin Size 100_1
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 48
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6b2. Produce Frequency Plots for Failure Times
Compress Occurrences into a Cumulative Timeline (Excel)

Scatterplot of Occurrences Bin , Cum Pct Bin Size vs Grouped Cycles_1

• Copy Failure Free Hours to a
160
new column in Excel, detached
100.0%
from table derived throughout
140
this presentation
Occurrences Bin Size 100_1

80.0%

Cum Pct Bin Size 100_1

120

100
• Sort from smallest to largest
60.0%
80 • Create a pair for each data
60 40.0%
point by adding an adjacent
40
column filled with zeros (value
20.0% = x is now = x, 0)
20

0 • Create scatterplot in Excel

0.0%
0 2000 4000 6000 8000 10000 12000 14000 • Compress entire output to
Grouped Cycles_1 - "New" or "Like New"
Variable
Occurrences Bin Size 100_1
produce format shown below
Cum Pct Bin Size 100_1

Cumulative Timeline for Previous Plot Clearly see a bias toward

1
0 failures earlier in the
0 2000 4000 6000 8000 10000 12000 14000 16000
cumulative timeline
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 49
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6c. Use Minitab to Select Best Distribution for Data
Weibull is a valid model for this particular data
Stat > Quality Tools > Individual
Distribution Identification
Data are arranged as a Single Column
‘Failure Free Hours …’

Probability Plot for Failure Free Cycles_1 - "New" or "Like New"

Goodness of Fit Test
Normal - 95% CI Exponential - 95% CI
99.99 99.99
Normal
90 AD = 98.059
99 50 P-Value < 0.005
90
10

Percent

Percent
Exponential
50
1
AD = 15.622
10 P-Value < 0.003

1
Weibull
0.01 0.01 AD = 1.514
-5000 0 5000 10000 0.1 1 10 100 1000 10000 P-Value < 0.010
Failure Free Hours at each retu Failure Free Hours at each retu
Gamma
Weibull - 95% CI Gamma - 95% CI AD = 3.374
99.99 99.99 P-Value < 0.005
99
90 90

50 50

10 10
Percent

Percent
1 1

0.01 0.01
0.01 0.1 1 10 100 1000 10000 0.001 0.01 0.1 1 10 100 1000 10000
Failure Free Hours at each retu Failure Free Hours at each retu

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 50
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6d. Use Parameters to Create a Model for the Data
Histogram of Failure Free Cycles_1
Weibull
900 Shape 0.8540
Graph > Histogram > Simple Scale 974.3
800 N 1215
Graph variables
700 ‘Failure Free Cycles_1’

600 Multiple Graphs > Multiple Variables > Overlaid on the same graph > OK
Frequency

500 Data View > Distribution > check ‘Fit distribution’

(choose Weibull, leave ‘Historical Parameters’ blank) These parameters
400 drive the Weibull
OK to make plot, then adjust the bins. Right-click bars on histogram model that
300 Edit Bars > Binning > Cutpoint replaces the data

200

100

0
0 2000 4000 6000 8000 10000 12000
Failure Free Cycles_1 - "New" or "Like New"

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 51
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
6e. Determine Times-to-Failure based on the Model
“New” or “Like New”

Explanation:

Per the selected model,

• 1% of the failures occur

between 5 and 8-9 cycles

• 10% of the failures occur

between 73 hours and 96
cycles

• The average failure is

between 922 and 1038
cycles

STAT > Reliability/Survival > Distribution Analysis (Right Censoring) > Distribution ID Plot
Variables: ‘Failure Free Cycles_1’ Specify Weibull
There are options available to draw further distinctions on censoring that are beyond today’s scope.
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 52
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
 6f. Final Validity Check of Distribution Selected
Probability Plot for Failure Free Cycles_1
LSXY Estimates-Complete Data Explanation:
Correlation Coefficient
Weibull Weibull
0.995 For this particular
99.99
set of data, the
95 Weibull model is
80 quite solid between
50 50-5,000 cycles
20 however is a little
shaky outside of
Percent

5 this range.
2
1 We need to pay
close attention to
processes
impacting early life
failures, e.g.,
0.01
0.01 0.1 1 10 100 1000 10000 • Testing
Failure Free Cycles_1 - "New" or "Like New" • Burn-in
• Shipping

STAT> Reliability/Survival > Distribution Analysis (Right Censoring) > Distribution ID Plot.
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 53
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
7a. Using Weibull Plots to Gain Performance Insight
Per the selected model, e.g., 50% fail at ≈ 634.3 cycles
Same source data,
“New” or
Distribution Plot manipulated as
“Like New” Weibull, Shape=0.854, Scale=974.3, Thresh=0 appropriate, used for
7a – 7d
0.0025
Graph > Probability Distribution Plots > View Probability
Distribution: Weibull
0.0020 ‘Shape’ and ‘Scale’ parameters from “Distribution ID Plot”
Probability Density

Shaded Area:
0.0015 Right Tail with Probability = 0.5 Shape parameter < 1
implies performance
is in Infant Mortality
0.0010

634.3 cycles creates

0.0005 an area under the
curve of 50%

0.5
0.0000
0 634.3
Cycles to Failure_1 - "New" or "Like New"
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 54
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
7b. Using Weibull Plots to Gain Performance Insight
Per the selected model, e.g., ≈ 13.3% fail 100 cycles
Same source data,
“New” or
Distribution Plot manipulated as
“Like New” Weibull, Shape=0.854, Scale=974.3, Thresh=0 appropriate, used for
7a – 7d
0.0025
Graph > Probability Distribution Plots > View Probability
Distribution: Weibull
0.0020 ‘Shape’ and ‘Scale’ parameters from “Distribution ID Plot”
Probability Density

Shaded Area:
0.0015 X Value, Left Tail, Value:100 Shape parameter < 1
0.1333 implies performance
is in Infant Mortality
0.0010

100 cycles creates an

area under the curve
0.0005
of 13.38%

0.0000
0100
Cycles to Failure_1 - "New" or "Like New"
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 55
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
7c. Using Weibull Plots to Gain Performance Insight
Per the selected model, e.g., 50% fail ≈ 2300 cycles
Same source data,
Minimal manipulated as
Distribution Plot
Repair appropriate, used for
Weibull, Shape=1.164, Scale=3143, Thresh=0
7a – 7d
0.00025
Shape parameter ≈ 1 implies
performance is in Steady-State
0.00020
(Data from “New” or “Like New”
Probability Density

repairs shows early-life tendencies

0.00015
however the failures upon “Like
New” repairs have higher failure
0.00010 rates than those systems minimally
repaired)
0.00005 2294 cycles
creates an area 0.5
under the curve of 50%
0.00000
0 2294
Cycles to Failure_2 - Minimal Repair
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 56
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
7d. Using Weibull Plots to Gain Performance Insight
Per the selected model, e.g., 50% fail ≈ 2300 cycles
Same source data,
Fleet manipulated as
Distribution Plot
Weibull, Shape=1.424, Scale=66443.5, Thresh=0 appropriate, used for
7a – 7d
0.000012
Shape parameter ≈ 1 implies
performance is in Steady-State
0.000010
however fleet data has
considerations when doing Weibull
0.000008
Density

0.000006

0.000004

0.000002 51,366 cycles 0.5

creates an area
0.000000
under the curve of 50%
0 51366
Cycles to Failure_3 - Fleet
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 57
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
 7e. Using Weibull Plots to Gain Performance Insight
Steady-State Fail 95% CI (hours)
Distribution Plot 1% 24.7 105.9
Weibull, Shape=1.13633, Scale=2827.54, Thresh=0 5% 126.6 351.7
0.00030 10% 259.6 599.8
Scatterplot of Occurrences_Bin_, Cum Pct_Bin_Size vs Grouped Cycles_B
50% 1640.1 2514.0
5 100.0%

0.00025

Occurrences_Bin_Size_100_SS
0.02217

Cum Pct_Bin_Size_100_SS
80.0%
4
Probability Density

60.0%
0.00020 3

40.0% Results for

2 illustration
0.00015 20.0%
of Steady
1 0.0%
State:
0 2000 4000 6000 8000 Variable 12000
10000
0.00010 Example of Steady-State Occurrences_Bin_Size_100_SS
Cum Pct_Bin_Size_100_SS Data not
reviewed in
this
0.00005 presentation

<0.1% fail within 100 hrs

0.00000
100
0 Model: Weibull
Steady-State Cycles to Failure • Shape 1.13633
• Scale 2827.53698
Note early shape of data (spread out) and model (skewed but thick in the beginning).
Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 58
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations
7f. Using Weibull Plots to Gain Performance Insight
Wearout Fail 95% CI (hours)
Example of Wearout Distribution Plot 1% 822.9 5323.3
Weibull, Shape=4.33205, Scale=11777.6, Thresh=0 5% 2182.7 7053.5
• Note how shape is close 10% 3354.0 7996.7
0.00014 Scatterplot of Occurrences_Bin_, Cum Pct_Bin_Size vs Grouped Cycles_B

4 100.0% to normal. This is by 50% 9797.8 11701.7

Occurrences_Bin_Size_100_WO

virtue of the shape value.

Cum Pct_Bin_Size_100_WO
80.0%

0.00012
3

2
60.0%
• Shape parameter =3.4-3 .6
closest to normal
Probability Density

40.0%

0.00010 1
20.0% distribution
0 0.0% Results for
0.00008 0 2000 4000 6000 8000
Example of Wearout
10000 12000 14000
illustration
0.6113
Variable
Occurrences_Bin_Size_100_WO
Cum Pct_Bin_Size_100_WO
of Wearout:
0.00006
Data not
reviewed in
0.00004 this
presentation
0.00002
>61% fail after 10,000 hrs
0.00000
0 10000 Model: Weibull
Wearout Cycles to Failure • Shape 4.33205
• Scale 11777.56298

Minitab Insights This document does not contain technology or Technical Data controlled under either the
Slide 59
9/13/2016 U.S. International Traffic in Arms Regulations or U.S. Export Administration Regulations