Superimposed failure modes

[Kieran Gillick]

Hi. I'm looking at gas leakage from a pressurised (200 bar) aluminium container. Gas leaks out between the metal surface of the container and a rubber O-ring. The seal is lubricated with grease. A Weibull plot of the distribution of leak rates seems to be split into two sections; a steep straight line for low leak rates and a more shallow line at higher leak rates. Does this indicate two independent leakage modes? Or two superimposed leakage modes? Or is it simply saying that a Weibull distribution doesn't apply to this data? Finally, is there any way of separating out these two distributions and estimating Weibull parameters for each one?

Thanks in advance.
Kieran

[RS Support]

Without looking at your data, plots and failure analysis results all I can give you are generic guidance and answers to your some of your questions:

1. Yes you can easily separate them and analyze them. See Competing Failure Modes Weibull (href: http://www.weibull.com/LifeDataWeb/competing_failure_modes.htm)
2. A Mixed Weibull model may also be used (href: http://www.weibull.com/LifeDataWeb/the_mixed_weibull_distribution.htm)

To the question of “are they independent?” (or even different modes) – this is more of an engineering question i.e. you have to look at your failure analysis to determine that. What you indicated does suggest that, however FA should confirm it. If you do use a competing failure modes analysis -- the assumption is that there are two modes and they are independent.

[Ari Antonovsky]

I have been studying failures in conveyor belt splices over a number of years. Almost invariably two distinct fault factors can be identified. That is, the splice would have survived to full-term if one or the other fault had been present, but the two faults combined to cause early failure. In effect, the faults are not additive, but multiply the unreliability of the system.

I agree with the statement in your chapter on "Competing Failure Modes" that there can only be one failure mode for each failed product. However, how do you analyse the effect of two interacting factors contributing to that failure mode?

The classic case is that of Corrosion Fatigue. The single failure mode is still fracture, but more importantly, the combined effect of corrosion and fatigue is far more detrimental than either of these operating independently.

I believe this will be an important area in future, as modern materials and systems are well-designed to withstand individual faults. Consequently, the failures that we see will increasingly result from combined effects.

[Pantelis]

In competing failure modes analysis one assumes independence of modes. In the case that the modes are dependent one needs to approach this differently.

One-way to achieve this is by using principles from “Accelerated life testing”. An article in the Reliability edge
“Using QALT Models to Analyze System Configurations with Load Sharing”
(see http://www.reliasoft.com/newsletter/4q2002/qalt.htm )
describes this using a load-sharing example (which is really dependency).

A reliability hotwire article expands upon this by describing an analysis with failure modes,
“Using Reliability Block Diagrams to Analyze Dependent and Independent Failure Modes”
see http://www.weibull.com/hotwire/issue27/hottopics27.htm.

I believe that the two above articles should answer your question.