At a recent site engagement, I had the opportunity to test a new cloud software platform that automated Weibull analysis and generated component replacement recommendations.
This site had a large fleet of mining trucks and loaders. All their CMMS work order data was used by the platform to generate the analysis.
A particular example caught my attention. The operator seats in the cabins of the trucks were assessed as exhibiting mild wearout (Weibull shape parameter slightly above 1), and the recommended replacement interval was 1600 operating hours (not even three months at average utilisation).
This is clearly a ludicrous suggestion, and digging into the analysis soon revealed why.
On some occasions, the technicians would elect to replace the entire seat rather make a repair. At the seat’s part number level, the failure data in the CMMS consisted of work orders documenting multiple failure modes.
Mixing of failure modes of varying shape parameters in a single Weibull analysis will often return a shape parameter around 1, leading to the assumption of a random failure distribution. This isn't useful for reliability improvement, especially if it is clearly wrong.
This phenomenon is known as "Drenick's Law" (F. Drenick, R. (1960). The Failure Law of Complex Equipment. Journal of The Society for Industrial and Applied Mathematics.).
Reliability engineers are familiar with the Weibull statistical distribution and its utility in representing component failure patterns. In a two-parameter Weibull distribution, the distribution is defined by the shape and scale parameters. The shape parameter indicates the general pattern of the failed components.
A shape parameter less than one indicates a component that suffers from a burn-in or infant mortality failure, and a shape parameter greater than one indicates a component that exhibits a wear-out characteristic, ie, the rate of occurance of failure increases as time in service increases.
A shape parameter of precisely one is a special case and, the Weibull function reduces to the exponential distribution, which suggests failures are completely random and cannot be predicted (and therefore eliminated with an appropriate maintenance intervention).
Weibull analysis is appropriate when used for simple, non-repairable components. An extreme example would be an incandescent globe, or even an LED light.
When Weibull analysis is applied to complex, repairable systems - like our mining equipment - care must be taken to ensure the analysis is applied to failure data exhibiting a common failure mode. This requires the reliability engineer to review the work order data and, ideally, the failed components themselves. Otherwise, "Garbage In, Garbage Out" applies.
In the case of the seat analysis, several failure modes were evident when the work order text was reviewed:
Worn suspension mechanism
Control system faults, both electrical and pnematic, and
Worn cushions and trim.
If the seat exhibited a combination of failure modes, say, worn mechanism and a failed bladder, the technician might elect to replace the entire seat, rather than incur extended downtime effecting a repair.
Each of these different failure modes, if viewed independently, would return individual shape and scale parameters. However, when they are mixed together and assessed, the software simply returned the calculated value based on this mixed input. In this case, that valued tended to be slightly higher than one, leading to the software's conclusion of "mild wearout" and suggesting the unrealistic 1600-hour replacement interval.
If not detected, the result could have been a recommended maintenance strategy for the seat that did not align with its real-world failure mode, and therefore would have been completely inappropriate. The site would then have seen either unplanned failures requiring replacement or, more likely in this case, higher costs due to over-maintaining the seat through unnecessary replacements.
Drenick's Law involves a number of caveats if applied strictly, but is a reminder to reliability engineers to ensure their Weibull analysis does not mix multiple failure modes. The result can be the incorrect assumption of "random failure rate", and a failure to take appropriate action to eliminate each failure mode.
Bluefield have conducted numerous projects on reliability modelling, failure investigations and defect elimination – read about some of our other experiences here and here.