14.03.2014 Views

Quality and Reliability Methods - SAS

Quality and Reliability Methods - SAS

Quality and Reliability Methods - SAS

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Chapter 19 <strong>Reliability</strong> <strong>and</strong> Survival Analysis 367<br />

Estimation of Competing Causes<br />

To use this feature, hold down the Shift key, click on the red triangle of the Product-Limit Survival Fit<br />

menu, <strong>and</strong> then click on the desired fit. You may then enter a constrained value for the parameter as<br />

prompted: theta for the exponential fit; beta for the Weibull fit; <strong>and</strong> sigma for the lognormal fit.<br />

Estimation of Competing Causes<br />

Sometimes there are multiple causes of failure in a system. For example, suppose that a manufacturing<br />

process has several stages <strong>and</strong> the failure of any stage causes a failure of the whole system. If the different<br />

causes are independent, the failure times can be modeled by an estimation of the survival distribution for<br />

each cause. A censored estimation is undertaken for a given cause by treating all the event times that are not<br />

from that cause as censored observations.<br />

Nelson (1982) discusses the failure times of a small electrical appliance that has a number of causes of<br />

failure. One group (Group 2) of the data is in the JMP data table Appliance.jmp sample data, in the<br />

<strong>Reliability</strong> subfolder.<br />

To specify the analysis you only need to enter the time variable (Time Cycles) in the Survival dialog. Then<br />

use the Competing Causes menu comm<strong>and</strong>, which prompts you to choose a column in the data table to<br />

label the causes of failure. For this example choose Cause Code as the label variable.<br />

Figure 19.20 Competing Causes Window<br />

The survival distribution for the whole system is just the product of the survival probabilities. The<br />

Competing Causes table gives the Weibull estimates of Alpha <strong>and</strong> Beta for each failure cause. It is shown<br />

with the hazard plot in Figure 19.21.<br />

In this example, most of the failures were due to cause 9. Cause 1 occurred only once <strong>and</strong> couldn’t produce<br />

good Weibull estimates. Cause 15 happened for very short times <strong>and</strong> resulted in a small beta <strong>and</strong> large<br />

alpha. Recall that alpha is the estimate of the 63.2% quantile of failure time, which means that causes with<br />

early failures often have very large alphas; if these causes do not result in early failures, then these causes do<br />

not usually cause later failures.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!