Weibull
Analysis: When Your Product Starts Talking About Its Own Mortality — and
You Gain the Power to Predict When It Happens
A Story That
Begins at the Component Graveyard
It was in November when Peter got a call from one of his largest
customers. Electric motors had started failing on a production line in
Switzerland — not all at once, not every single one, but with an
unsettling regularity that felt like the ticking of a clock. Three
failures in two weeks. Then five in a month. No obvious cause, no single
parameter that could explain why those particular units had decided to
die.
Peter knew the classic approach — take the failed unit, tear it down,
find the root cause — wasn’t going to cut it this time. The problem was
that failures weren’t happening because of a single manufacturing
defect. They were happening because something in the design itself or in
the operating conditions was accelerating the aging of the components.
And when Peter looked at the data — operating hours of each failed
motor, date of manufacture, usage conditions — he realized he was
looking at a story about product mortality. And there is one tool,
purpose-built for telling exactly this kind of story.
It’s called Weibull Analysis.
What
Is Weibull Analysis — and Why It Isn’t Just Another Statistical
Tool
Weibull analysis is a statistical method for analyzing product and
component lifetime and reliability, used to model time-to-failure. It’s
named after the Swedish engineer Waloddi Weibull, who published it in
1951 in a paper describing the probability distribution that now bears
his name.
The Weibull distribution — as this function is known — is remarkable
in that it can model three completely different failure types depending
on a single parameter:
-
Shape parameter (β) < 1: Failures are caused by
defective units — they fail early, in the “infant mortality” period. The
longer a component survives, the lower the probability that it will
fail. -
Shape parameter (β) = 1: Failures are random —
constant failure rate over time. Typical for electronic components in
stable conditions. -
Shape parameter (β) > 1: Failures are
wear-out — the failure rate increases with time. The older the
component, the more likely it is to fail. This is wear-out failure,
typical for mechanical parts, bearings, seals.
And that flexibility is precisely what makes Weibull so powerful. It
doesn’t force your data into a predefined shape. The data itself will
tell you what type of failure is occurring.
How It Works in Practice
— Step by Step
Step 1: Collecting Failure
Data
The foundation of every Weibull analysis is quality failure data.
You need:
- Time-to-failure for each failed unit — in operating
hours, kilometers, cycles, days, whatever is the relevant usage
metric. - Censored data — information about units that
haven’t failed yet. If you delivered 100 motors and 12 failed, the
remaining 88 are still running. You must not ignore these “survivors” —
they carry critical information that failures aren’t as frequent as they
might appear if you only looked at the failed units.
In Peter’s case, he had 23 failed motors with precise operating
hours and knew of another 180 motors of the same type still in service
with varying hours of deployment.
Step 2: Sorting and
Assigning Ranks
Failures are sorted from shortest to longest time. Each failure is
assigned a so-called Bernard’s approximation formula
or median rank, which determines the probability of
failure at that position:
Median Rank = (i – 0.3) / (n + 0.4)
where i is the failure sequence number and n is
the total number of observations.
This step is critical — you’re not just calculating the percentage of
failed units, but attempting to estimate the population failure
probability based on a limited sample.
Step 3: Creating the Weibull
Plot
On Weibull paper — graph paper with transformed axes — or in
software, a plot is created where the X-axis represents time to failure
and the Y-axis represents failure probability. If the data roughly
aligns into a straight line, the Weibull distribution is a suitable
model.
The slope of this line determines the shape parameter β. The point
where the line crosses the 63.2% failure probability determines the
scale parameter η (eta) — the so-called characteristic life.
Step 4: Interpreting the
Parameters
-
β (shape parameter): Tells you WHAT type of failure
is occurring. Infant mortality? Random? Wear-out? -
η (scale parameter): Tells you WHEN. The
characteristic life is the time by which 63.2% of the population will
have failed. Not the mean, not the median — it’s a distribution
parameter. -
B10 life: The time by which 10% of the population
will have failed. A critical parameter in automotive — if your B10
doesn’t meet the required value, the customer will know. -
γ (location parameter): The minimum time before
which failure cannot occur. Used in the three-parameter Weibull model —
typical for situations where a “warranty period” exists during which
failure is physically impossible.
Peter’s Case — What
the Analysis Revealed
When Peter plugged the data into a Weibull analysis, the picture was
clear and brutal:
-
β = 2.7 — The failures were wear-out. Components
were wearing down and the failure rate was increasing over time. -
η = 14,300 hours — The characteristic life was
approximately 14,300 operating hours. -
B10 = 5,200 hours — The first 10% of components
failed after just 5,200 hours.
The customer was running the motors 24/7, which meant B10 was
reached in approximately 217 days — just over 7 months. And that
matched the failure timeline the customer had been reporting
exactly.
But the story didn’t end with diagnosis. Weibull analysis enabled
Peter to do something far more powerful — predict.
Based on the model, he was able to calculate that with 180 motors
currently in operation, the probability of another 8–12 units failing in
the next 30 days was over 75%. That’s no longer a statistical exercise
— that’s management intelligence that drives decisions.
Where Weibull
Analysis Is Used in Practice
Automotive Industry
In automotive, Weibull is a cornerstone of reliability engineering.
Predicting B10 life for critical components — brake systems, control
units, sensors — is part of virtually every PPAP package for
safety-relevant components.
Aerospace Industry
Here they don’t work with B10, but with far more extreme targets.
B0.01 — the time by which 0.01% of the population fails — is the metric
for critical aerospace components. Weibull analysis combined with
predictive maintenance determines when components are preventively
replaced.
Healthcare and
Pharmaceuticals
Weibull is used to model time-to-failure of medical devices,
durability of implants, and even patient survival analysis in clinical
trials.
Energy Sector
Turbines, transformers, cables — all have a lifecycle that can be
modeled with the Weibull distribution. Preventive replacement before
reaching the critical point saves millions.
Critical Pitfalls —
What to Watch Out For
1. Mixed Failure Mechanisms
If your data contains failures from different causes — for example,
mechanical wear and electrical overload — the Weibull plot will show a
curved line instead of a straight one. That’s a signal you’re mixing
apples and oranges. The solution: separate the data by failure mechanism
and analyze them individually.
2. Ignoring Censored Data
This is the most common beginner mistake. If you analyze only failed
units and ignore those still functioning, your analysis will inevitably
be pessimistic. The information that 180 motors are still running is
just as important as the information about 23 failures.
3. Too Little Data
Weibull analysis is robust — it works with as few as 5–6 failures —
but with small samples, the confidence intervals around parameters are
very wide. The idea that you’ll precisely determine B10 life from five
failures is an illusion. You’ll get a range, not a number.
4. Extrapolation Beyond Data
If your longest failure occurred at 10,000 hours, you can’t
confidently claim what will happen at 20,000 hours. You can extrapolate
the model, but the responsibility for such a prediction is enormous.
Always report confidence intervals.
Weibull and Other
Quality Tools — Synergy
Weibull analysis doesn’t exist in a vacuum. In practice, it
integrates with other tools:
-
FMEA: Weibull provides quantitative estimates for
failure detection and severity that FMEA requires. -
Reliability Block Diagram (RBD): For systems with
series or parallel component arrangements, Weibull parameters of
individual components are used to calculate overall system
reliability. -
Preventive maintenance: Weibull determines the
optimal interval for preventive replacement — before the point where the
failure rate starts increasing sharply, but not so early that you’re
throwing away components that still have life in them. -
Warranty analysis: Combining Weibull analysis with
warranty claim costs allows you to precisely calculate how much the
warranty period will cost and where to set the boundary.
Practical Guide — How to
Get Started
-
Get software. Minitab, ReliaSoft Weibull++, JMP —
all of these tools have built-in Weibull analysis. For simple analyses,
even Excel with an add-in will suffice. -
Start small. Take one component family where you
have at least 10 failures and a few dozen censored observations. Do your
first analysis. See if the data “talks.” -
Pay attention to data quality. Weibull is only as
good as your data. Inaccurate operating hours, incomplete installation
records, missing information about operating conditions — all of this
diminishes the value of the analysis. -
Contaminate the analysis with domain knowledge.
Don’t just plug in numbers. Think about what you’re seeing. If β came
out at 3.5 but you know the component is electronic and shouldn’t
exhibit wear-out, the observation isn’t wrong — your understanding of
failure physics is. -
Communicate results visually. A Weibull plot is a
straight line — understandable even to non-statisticians. Show
management the curve, show them B10, show them the prediction. Numbers
convince, but pictures convince faster.
Conclusion — Why It’s Worth
Knowing
Weibull analysis is one of those rare tools that truly transforms
reactive quality management into proactive. You’re no longer just the
person analyzing what went wrong. You’re the person who knows what will
go wrong next — and when.
Peter’s case with the electric motors ended the way it was supposed
to. Based on the Weibull analysis, he initiated preventive replacement
of all motors that exceeded 4,500 operating hours, and simultaneously
worked with the supplier on redesigning the bearing system. The new
motor revision had B10 shifted from 5,200 to 11,800 hours — more than
double.
The customer never saw the Weibull plot. Never saw β or η. But they
saw the result — motors that started performing the way they were
supposed to from the beginning. And that’s what reliability truly means.
Not statistics, but the confidence that a product will endure what it
promises.
Peter Stasko is a Quality Architect with 25+ years
of experience in automotive, manufacturing, and continuous improvement.
He doesn’t just teach quality — he lives it. His approach bridges
Japanese philosophy with the hard reality of European manufacturing.
