Weibull Analysis

Definition: Weibull analysis is a statistical method that fits failure data to the Weibull distribution to model time-to-failure and forecast reliability. The shape parameter (beta) describes how failure risk changes with age, and the scale parameter (eta) sets the characteristic life in the same units as the failure times.

What Is Weibull Analysis?

Weibull analysis is a statistical method for studying equipment failure: it takes the times at which units failed, fits a two-parameter Weibull distribution to them, and produces a model of time-to-failure. From that model, a reliability engineer can forecast how many units in a population will survive to a given age, estimate the risk of failure over the coming months, and decide when replacement is justified. The two-parameter form uses beta, a dimensionless shape parameter, and eta, a scale parameter expressed in the same units as the failure times.

The Two Parameters: Beta and Eta

Beta controls the shape of the failure curve: whether failure risk falls, holds steady, or climbs as units age. It is dimensionless, so a beta of 2.1 points to the same aging pattern in a pump as in a gearbox.

Eta controls the scale. It is the characteristic life, defined as the age by which 63.2% of the population is expected to have failed. At t = eta the reliability function returns exp(-1) = 0.368, so eta is easy to read from a Weibull plot: find the 63.2% cumulative failure line and read the corresponding time. Both parameters come from the fit itself, which makes checking the fit the analyst's first job before acting on the numbers.

What Beta Tells You

Beta is the output with the most direct maintenance implications, because it classifies the failure pattern and points to a response.

Beta Value Failure Pattern Example Failure Modes Typical Response
Beta < 1 Decreasing failure rate; failures concentrate early in life (infant mortality) Installation errors, manufacturing defects, early-life software faults Tighten commissioning and run-in procedures; screen parts and workmanship before service
Beta = 1 Constant failure rate; failures behave as random events Random overload, human error, unpredictable external events Replacement on a fixed schedule adds little; address design and error-proofing instead
Beta > 1 Increasing failure rate; risk grows with age (wear-out) Bearing wear, belt wear, seal degradation, corrosion Replace ahead of the wear-out region; monitor condition to catch degradation early

A beta below 1 usually signals a workmanship or quality problem, which is where precision maintenance practices pay off: correct alignment, torque, and lubrication remove many of the defects that otherwise surface as infant-mortality failures. A beta near 1 warns that time-based replacement of the component is buying little, since failures are not age-driven. A beta above 1 is the case where scheduling works, and it is where the rest of this page concentrates.

Worked Example: Forecasting Pump Reliability

Consider 10 identical pumps that failed at these running hours: 6,200; 8,100; 9,400; 10,500; 11,600; 12,700; 13,900; 15,400; 17,600; and 21,300. Fitting a two-parameter Weibull distribution to these times, using the methods described below, yields beta = 2.1 and eta = 13,500 hours.

With beta above 1, these pumps show a wear-out pattern: failure risk rises with age. Eta says 63.2% of a pump population is expected to fail by 13,500 hours, and each failure on a continuously running line is also equipment downtime. The Weibull reliability function gives survival probability at age t: R(t) = exp(-(t/eta)^beta).

To find the share of pumps expected to survive to 10,000 hours:

  1. Divide the age by the characteristic life: t / eta = 10,000 / 13,500 = 0.741.
  2. Raise the result to the power of beta: 0.741^2.1 ≈ 0.533.
  3. Apply the exponential: R(10,000) = exp(-0.533) ≈ 0.59.

About 59% of the pumps are expected to survive to 10,000 hours, and 41% are expected to have failed by then. Running the same function at eta gives R(13,500) = exp(-1) ≈ 0.368, which confirms the characteristic life. The fit also implies a median life of about 11,300 hours, the age at which half the population is expected to have failed.

How the Fit Is Done

Two fitting methods dominate practice. Rank regression plots each failure time against its median rank on Weibull probability paper and fits a straight line: the slope estimates beta, and the time at the 63.2% line estimates eta. The plot doubles as a visual check, since curvature suggests the data mix failure modes. Maximum likelihood estimation (MLE) finds the beta and eta values that make the observed failure times most probable, and software implementations usually prefer it for small samples and heavily censored datasets.

Real studies rarely observe failures alone. Units still running when the study ends, or removed for an unrelated reason, are right-censored (suspended). These units carry information: a pump still healthy at 20,000 hours proves the population can reach 20,000 hours, and leaving it out biases eta downward. Median rank methods adjust the plotting positions to account for suspended units, so right-censored data contributes to the fit instead of distorting it.

Maintenance Applications

A fitted Weibull model supports four recurring maintenance decisions:

  • Setting replacement intervals. The distribution shows the failure probability at each age, so the interval can be set where risk is still acceptable. In the pump example, the B10 life (the age by which 10% are expected to fail) works out to roughly 4,600 hours.
  • Comparing failure modes. Fitting each mode separately shows which ones are aging and which are random, so effort goes where the pattern justifies it.
  • Justifying condition monitoring. When beta is above 1 and the degradation is measurable, the fit builds the business case for predictive maintenance: bearing wear shows up in vibration data long before the bearing seizes, which converts an unexpected failure into a planned replacement.
  • Sizing the spares inventory. The model forecasts how many failures to expect in a quarter, which tells planners how many spares to hold and shortens unplanned downtime when a failure does occur.

Limitations

Weibull analysis is only as strong as its inputs, and three limitations come up repeatedly:

  • Mixed failure modes. A dataset that lumps several mechanisms together produces a curved Weibull plot, and a single beta then describes nothing real. Separating modes before fitting is standard practice.
  • Small samples. Ten units, as in the pump example, produce useful point estimates but wide confidence intervals; treat beta and eta as ranges, not precise constants.
  • Poor failure data. Missing failure dates, unknown operating hours, or records that do not say what broke leave the fit with nothing reliable to model. Data cleanup precedes the analysis.

Frequently Asked Questions

What do beta and eta mean in Weibull analysis?

Beta is the shape parameter. It describes the failure pattern: values below 1 indicate early-life failures, a value near 1 indicates a constant failure rate, and values above 1 indicate wear-out. Eta is the scale parameter, also called the characteristic life: it is the time by which 63.2% of the population is expected to have failed, and it carries the same units as time.

What does a Weibull beta below 1 mean?

A beta below 1 means a decreasing failure rate, often called infant mortality. Failures concentrate early in life and are typically caused by installation errors, manufacturing defects, or early-life software faults. The response is better commissioning and screening rather than more frequent scheduled maintenance.

How is Weibull analysis used in maintenance planning?

The fitted distribution sets replacement intervals by showing the failure probability at each age, often expressed as a B10 or B50 life. It also compares failure modes, supports the case for condition monitoring when wear-out is detectable before failure, and forecasts failure counts for spare parts stocking.

What is a suspended unit in Weibull analysis?

A suspended unit, also called a right-censored unit, is one that has not failed by the end of the study or was removed for an unrelated reason. It still contributes information, because surviving to a given age proves the population can reach that age. Median rank methods account for suspended units in the fit.

What data do you need for Weibull analysis?

You need failure times recorded in consistent units, the failure mode identified for each event, and suspended units tracked through the end of the study period. Missing failure dates or lumped failure modes distort the fit, so data cleanup comes before curve fitting.

The Bottom Line

Weibull analysis turns a column of failure times into two numbers you can act on: beta identifies the failure pattern, and eta anchors the characteristic life in real units. Read beta first, because it decides the response: workmanship fixes for beta below 1, design changes for beta near 1, and scheduled replacement or condition monitoring for beta above 1. Fit one failure mode at a time, keep suspended units in the dataset, and refresh the fit as new failures arrive so the forecast tracks the fleet.

Catch Wear-Out Before It Becomes a Failure

Tractian's multimodal sensors capture vibration, temperature, and operating context in one system, so degradation on assets with a beta above 1 surfaces while replacement is still a planned task.

Explore Condition Monitoring

Related terms