Destructive Test Methods: Validation of Continuous Measurements (Gage Reproducibility)

In our blog post about Gage R&R (with Minitab) aka Test Method Validation for continuous non-destructive measurements, we discussed things that are, of course, still applicable here, e.g., the  Installation Qualification (IQ) of the test equipment.

The main difference between non-destructive and destructive test methods is quite simple:

We cannot assess repeatability, as the part is already broken.

As the name already suggests, the part is broken after a destructive test method and cannot be used again.

In the nature of destructive testing, the measurement variation cannot be evaluated by repeated measurements on the same unit.

Furthermore, destructive testing cannot be used for 100% inspections, as one would not be able to sell even a single product. However, in some cases, it may be possible to obtain parts that are uniform enough to perform a gage R&R. In other cases, a gage reproducibility study is a preferable option [1].

This step is quite like in our blog post on Gage R&R (with Minitab).

Only this time can we skip the repetitions.

So, we already know that we have variable data (as opposed to attribute) and that we test in a destructive way (as opposed to a non-destructive way) – that's quite a lot.

The next step is to design the study and decide on the number of the following items:

With the number of operators (O), we can assume that the more operators we use, the better we understand possible weaknesses. At least ten operators must be selected from the pool of normal operators trained to the test method procedure. Each operator must then independently perform the entire test method, including any necessary preparatory steps (e.g., calibration or sample preparation) [1].

The number of parts (P) and the variation among them are slightly different from a gage R&R. For a gage reproducibility study, a minimum of five samples per operator is recommended, and the variation among them should be as small as possible. An increasing part-to-part variation makes it more challenging to observe the operator effects. If there is significant part-to-part variation, consider increasing the number of parts per operator [1].

Now that we know how many operators and samples we need, we can execute the actual measurements.

We recommend you organize the results in the following way:

After gathering the data, use statistical software to analyze it. We proceed using Minitab as our statistical software of choice.

When running an analysis in Minitab, we get quite a lot of tables.

Let's start with looking at Minitab's Analysis of Variance table (see Figure 1).

All p-values small or equal to 0.05 are considered statistically significant; thus, one can assume with 95% confidence that the operators are different.

NOTE: statistically significant only means that there is a detectable difference, but not that this difference has any technical meaning [1].

Figure 2 shows the size of the effects by showing the variance and standard deviation, which is almost the same, but the variance is the standard deviation squared.

The circled value represents the standard deviation of the reproducibility (SDreproducibility).

The last step is to verify the standard deviation of the reproducibility against the acceptance criteria.

First, we must decide whether we have a one- or two-sided specification limit.

Depending on that decision, we have the following acceptance criteria [1]:

If your analysis meets these acceptance criteria, you passed the gage reproducibility study. Otherwise, it’s best to improve the test method.