nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Reliability Analysis - Fit Competing Risk Mixture vs. Fit Mixture

Joseph_Reese

Hello,

Interested in an example of when you would choose the Fit Competing Risk Mixture vs. Fit Mixture in the Life Distribution platform. Fit Competing Risk Mixture makes sense to think about multiple failure modes competing throughout the life of the product but I am uncertain as to what the use cases would be for the Fit Mixture. Is this a case where both should be fitted and AICc/BIC compared or is there only ever one of these tools that makes sense to apply?

Thank you!

Accepted answers

DonMcCormack

Sorry I didn't remember this data set earlier. Take a look at the sample data table Mixture Demo (in the Reliability directory of the sample data). The first column, Y1, is a mixture of Weibulls. Y2 is a competing cause mixture of Weibulls. The formulas give a bit of insight as to the generating mechanism. The difference between the two types of data (if I am remembering correctly) is that the competing cause data comes from a single population with multiple failure modes whereas the mixture data comes from multiple populations each with a single failure mode. Hope this make sense.

All comments

Phil_Kay

Sounds like a job for global reliability expert, @DonMcCormack .

DonMcCormack

Thanks @Phil_Kay. Try the attached attached ignoring the Cause column. It is from Meeker & Escobar, p 383, Table 15.1. We've used it for one of the recorded sessions we did with Dr. Meeker. Unfortunately, I don't know if the recordings are still available. The data set won't necessarily give you insight into why you should choose one option over the other (I'll need to make one up for that), but it's a start since you have an explicit indication as to the failure cause.

DeviceGJMP.jmp

Phil_Kay

Those recordings with Bill Meeker were great. Really good motivational stuff as well as the technical how-to.

DonMcCormack

Maybe @gail_massari knows if they are still available and where they can be found.

Mark_Bailey

I am not sure if you got an answer yet. Basically, the mixture model is an investigative approach when you do not have assignable failure modes but you believe that more than one of them are active in the single set of lifetime data. You can explore a mixture of models to determine the number of failure modes and begin to understand their hazards.

This case is similar to unsupervised learning in predictive modeling.

DonMcCormack

gail_massari

Don @DonMcCormack and Phil @Phil_Kay , Joseph beat me to providing the Explorers link Best Practices in Reliability Data Analysis | On-Demand Webinars | JMP and here are a few more from Billl Meeker here, too:

https://www.jmp.com/en_us/search/jmpcom.html?q=bill%20meeker

Joseph_Reese

Thank you @DonMcCormack! I overlooked that dataset and it is quite helpful for picturing the differences. Any perspective on what scenario might be more common in industry? To me it seems like the competing risk mixture could be more common (although maybe harder to detect if the failure modes are similar enough?) but would welcome your opinion. Thank you to @Mark_Bailey as well!

DonMcCormack

@Joseph_Reese I think it would be process/product dependent. I could imagine that in something like protein fermentation you may get both, particularly in cases where accelerated stability is involved. With high purity manufacturing, competing risk may predominate. Definitely a case where the science is needed to drive the decisions.