Given-Shape Weibull Analysis and Zero-Failure Demonstration Test Planning

Description

When few or no events occur, borrow the Weibull shape from past experience to bound percentiles, and size a test that passes only if every unit survives.

With few events, a two-parameter Weibull fit is hopeless. If the shape parameter (how fast the hazard rises or falls) is known from similar products, times raised to that power behave like exponential data, so the scale can be estimated from total transformed exposure divided by the number of events, and a conservative lower bound exists even with zero events. The same algebra plans a demonstration: to show at a stated confidence that the p-th percentile life exceeds a requirement, run n units for a test time with no failures allowed, where n follows from the confidence, the shape and the ratio of test time to required life. Because the conclusion depends heavily on the assumed shape, analysis should be repeated over a plausible range of shapes. It generalises the exponential zero-event bound behind the rule of three.

Attributes

Loss type
Siloed
Effort
Days

Steps

  1. Obtain a plausible Weibull shape, or a range, from earlier data on similar products and failure modes.
  2. For analysis, raise each unit's exposure time to the given shape, sum them, and estimate the scale from total transformed exposure and the event count; compute bounds on percentiles such as B10 or on the failure probability at a target time.
  3. With zero events, use the exponential zero-failure bound on the transformed scale to get a conservative lower bound on the scale and hence on any percentile.
  4. Repeat with several shape values (for example 1.5, 2 and 3) and report whether conclusions change.
  5. For planning, fix the requirement (percentile p and required life), the confidence (for example 90 or 99 percent) and the feasible test time; compute the smallest n with n greater than log(alpha) divided by (k to the power shape times log(1 - p)), where k is test time over required life.
  6. Run the test; the demonstration passes only if no unit fails, and plan generous margins because a product only just meeting the requirement will often fail the test.

Sources

  • Statistical Methods for Reliability Data (Meeker & Escobar, 1998)

Also known as

  • Weibull analysis with given shape parameter
  • Zero-failure demonstration plan
  • Minimum sample size demonstration test
  • Success-run test
  • Given-shape (Weibayes-type) analysis and zero-failure reliability demonstration test plans

Categories

  • Analytics & Behaviour
  • Statistics & Modelling