Weibull in Excel: when it helps and when it misleads
A Weibull plot is a way of looking at life data — times to failure, or times to a defined event — to see whether a single wear-out or infant-mortality story is even plausible, and to estimate a characteristic life and a slope (the shape parameter). In Excel, that usually means a carefully prepared data set, a median-rank or similar plotting position, and a fit you can show. The Weibull Analysis Toolkit is built for that workflow.
When it helps
It helps when you have a reasonably clean set of similar items, a defined failure mode, and enough events (and suspensions) to see a pattern. It can support an interval discussion, a spare-part conversation, or a decision that a mode is still in a random-failure regime where a time-based restoration is a poor bet.
When it misleads
It misleads when the data mix several failure modes, when n is tiny, when “as-bad-as-old” repairs are treated as good-as-new lives, or when a straight line is forced through a clearly curved plot. A precise β to two decimal places is not a finding if the underlying sample cannot support it. A historical rate n/T can describe the observed events per unit of exposure, but it does not resolve mixed failure modes, a changing event rate or inappropriate repair assumptions. A predictive constant-rate model requires evidence that a homogeneous Poisson process is a reasonable approximation; model-based uncertainty intervals depend on those assumptions. The funding method is a scenario model, not proof of a constant failure rate.
Repairable systems: check the trend before predicting
For a repairable asset, examine cumulative events against operating age and the times between successive events. Distinguish restoration that leaves the asset approximately as-bad-as-old from replacement or renewal that starts a new life. Where the event rate trends over time, a repair-process model such as an appropriate nonhomogeneous Poisson process may be more useful than fitting independent Weibull lives.
Illustrative example: 12 events in 6,000 operating hours gives a historical average of 0.002 events/hour. If ten occurred in the final 1,000 hours, that average conceals a substantial change; it is not a justified forecast of two events in each future 1,000 hours. Check changes in duty, failure modes, recording and repair practice before selecting a model.
Primary reference: NIST/SEMATECH, trend and growth plotting for repairable systems. A Weibull life-distribution shape parameter and a repair-process trend parameter describe different models and are not interchangeable.
If you are setting PM intervals, a Weibull fit is an input to a discussion, not a substitute for auditing the program or for a P-F argument on a predictive task.
Related reading
- Weibull Analysis Toolkit
- PM audit checklist
- Risk quantification method