Statistically Significant Data for Test Engineer

A common engineering experiment is to test something until failure to determine its ultimate break strength. Sample sizes can range from 5 to 100 pieces.

I would like to know what statistical tests are used to make sure the data is good.

Statistical tests that I am aware of are used to compare data sets. For these engineering tests, there are no data sets to compare. You do the test, average the break strength, and that's it. However, sometimes the data is all over the place and the variances are large.

Is there a test that concludes, "I am 95% confident that the break strength is X."?

And if it is lower than 95% we should increase the sample size?

Thank you!


TS Contributor
By good, do you mean that it passes the requirements? There are several approaches depending on your goal.
  • If your main concern is to quantify the mean ultimate break strength use the confidence interval for the mean. You determine the % confidence you desire, which determines the sample size.
  • If your concern is to determine whether the mean ultimate break strength exceeds the specification, use a one-sided, 1-sample t-test.
  • If your concern is that the process will always yield an ultimate break break strength greater than specification, verify that the process is stable (stationary time series) using a control chart, then perform a capability study.
Note: ultimate break strength will probably not be from a Normal distribution, so keep that in mind. you may need to use the non-parametric equivalent tests. It will probably be right-skewed unless you have a very large component of measurement variation.