How to run a Weibull analysis
Iron Lake Reliability is a reliability workbook for hardware life data. You enter times and F/S, fit a two-parameter Weibull model, then reuse those parameters for B-lives, MTTF, hazard, fleet risk, accelerated life, and demonstration tests. It is not a PDF/CDF sketching toy, and it is not a certified life assessment.
Open the Iron Lake Reliability workbook Open Weibull Calculator
The 2P Weibull model
Unreliability (the CDF) and reliability are
F(t) = 1 − exp[−(t/η)β]
R(t) = exp[−(t/η)β]
Characteristic life η is the scale: 63.2% of units have failed by t = η, for any β. Mean life (MTTF) is η Γ(1 + 1/β) and equals η only when β = 1. Shape β sets the hazard trend — falling, flat, or wear-out — which is why reliability engineers estimate both parameters from data instead of drawing a CDF by eye.
How to do it in the workbook
- Open the Weibull Workbook. Enter time, F (failed) or S (suspended), and quantity. Optional STRESS is for ALT, not for a single-stress endurance set.
- Leave Distribution model on Arrhenius-Weibull unless you have a temperature or voltage acceleration story. Fit method MLE is the default.
- Read β and η, then check the probability plot. Points should track the fitted line; a 95% Fisher band is available when the grid has no numeric STRESS.
- On Weibull Calculator, evaluate R(t), B10, MTTF, or h(t) from the same LifeFit. Fleet risk and RDT live on their own tabs.
- Export Excel or PowerPoint if you need a record. Recalculate after edits — the workbook is client-side and does not silently keep a stale fit.
Worked example: spindle bearings (right-censored)
The built-in bearing sample is 12 failures and 7 suspensions (quantity on the S rows). Times in hours: failures at 167, 223, 284, 318, 376, 419, 468, 512, 571, 640, 718, 805; two suspensions at 450, two at 600, three at 850.
The Iron Lake Reliability MLE on that set is β = 2.168257 and η = 716.212 h (about β ≈ 2.168 and η ≈ 716.2 h). Shape above 1 is wear-out, which is what a bearing endurance set is supposed to show. The figures below are that pair run through the closed-form identities, not a second fit.
Reliability at 500 h. The cumulative hazard is H(500) = (500 / 716.212)2.168257 = 0.458772. Then R(500) = exp(−0.458772) = 0.632059, so F(500) = 0.367941. About 37% of this population has failed by 500 h. Those three digits match 1 − e−1, but that is this dataset only. Characteristic life is 716 h, and the failed fraction there really is 63.2%.
B10 is the quantile t0.10 = η [−ln(0.90)]1/β. −ln(0.90) = 0.105360516 and 1/β = 0.461200, so 0.1053605160.461200 = 0.354209. Then 716.212 × 0.354209 = 253.69 h, which rounds to the 254 h in the table. The same steps at 50% give B50 = 604.83 h. Mean life is 716.212 × Γ(1.461200) = 716.212 × 0.885603 = 634.28 h, below η, as wear-out requires. Hazard at that age is h(500) = (2.168257 / 716.212) × (500 / 716.212)1.168257 = 0.00302740 × 0.657156 = 0.001989 / h. From those two numbers:
| Quantity | Formula | Value |
|---|---|---|
| B10 life | η [−ln(0.9)]1/β | 254 h |
| B50 (median) | η [−ln(0.5)]1/β | 605 h |
| Characteristic life | η at F ≈ 63.2% | 716 h |
| MTTF | η Γ(1 + 1/β) | 634 h |
| R(500 h) | exp[−(500/η)β] | 0.632 |
When a two-parameter Weibull is the right model
Use it for non-repairable hardware with one dominant failure mode, times in one unit, and a mix of failures and suspensions. That is the ordinary endurance-test and warranty-return problem. Fix β = 1 (exponential) when the hazard really is flat, or when you do not have enough failures to identify a shape: one failure cannot pin down both parameters. Choose the workbook’s lognormal when the probability plot bends the other way, not because its mean formula looks simpler. If the rows were run at two temperatures or two loads, do not pool them into one Weibull. A hot unit and a cold unit are not the same population; that is an accelerated-life fit, which the tutorial walks through separately.
Common mistakes
- Calling η the mean or “the life.” On this fit the mean is 634 h, the median is 605 h, and B10 is 254 h. A drawing that says “life = 716 h” has not said which of those it means.
- Dropping suspensions. A unit removed alive at 850 h is evidence. Deleting it, or marking it failed, shortens the estimated life. Quantity on an S row is several units, not a weight to ignore.
- Trusting β from a thin failure count. B10 depends on a power 1/β out in the tail. A small move in a poorly identified shape moves B10 more than it moves η. Read the probability plot before you quote a B-life.
- One Weibull for two failure modes. A seal leak and a raceway fatigue are different shapes. A single β is then a mixture, not the physics of either mode.
- Reading the header bathtub as this fit. A single 2P Weibull hazard only falls, stays flat, or rises. See hazard rate.
Load bearing sample on the workbook to reproduce the fit, then open Weibull Calculator for B10 / MTTF / hazard with the same LifeFit. For the closed-form identities with round numbers, see the B10, MTTF, and hazard rate pages (β = 2, η = 1 000 h).
FAQ
What is two-parameter Weibull analysis?
It fits a shape β and a characteristic life η to failure and suspension times. The CDF is F(t) = 1 − exp[−(t/η)^β]. Reliability is R(t) = 1 − F(t). η is the 63.2% failed life, not the mean.
How should right-censored units be entered?
Mark failures as F and survivors as S (suspensions) at the time they were last seen unfailed. Quantity can repeat identical rows. Maximum-likelihood estimation lets failures contribute the density f(t) and suspensions contribute R(t).
What does β tell you?
β < 1 is a falling hazard (infant mortality). β ≈ 1 is roughly constant hazard (exponential / useful life). β > 1 is a rising hazard (wear-out). A wide interval on β means the shape is poorly identified — add failures or test longer.
Is Iron Lake Reliability a PDF/CDF curve sketching tool?
No. It is a reliability workbook: life-data grid, 2P Weibull or lognormal fit, Arrhenius / IPL accelerated life, reliability demonstration (RDT), fleet risk, and Excel / PowerPoint export. Plotting F(t) is a result of the fit, not the product.
Open the Iron Lake Reliability workbook Open Weibull Calculator