Building a Weibull model from a sample of data that covers a prolonged duration assumes that all failure times observed are “independently and identically distributed” (iid). That means that no change occurred in the inherent reliability of the item over the duration of the data. However there are at least three reasons why this assumption may not apply:
- Design modifications and improvements may have been made on the equipment thereby increasing its reliability.
- Imperfect repairs were carried out.
- Severity of usage increased or decreased.
Andrew Jardine points out[1] that conducting a Weibull (or any failure age) analysis in the above cases (described as either a “happy” or “sad” system in the slides below) will be inappropriate. The following slides (not accessible in Firefox) describe a method for confirming whether the histories in the sample are iid.
© 2015, Murray Wiseman. All rights reserved.
- [1]Maintenance, Replacement, and Reliability CRC Press(2005) ISBN 10: 0849339669 ISBN 13: 9780849339660↩