How You can Use Statistics in the Food Industry - Part 4

Gauge R&R

Gauge R&R (Reliability and Reproducibility) is a method to look at personnel, protocols and equipment when studying your quality control checks. 

1. Are your operators doing the checks the same way?
2. Are our protocols / procedures being followed?
3. Is the equipment (gauges) acceptable for the task that is being performed?

The statistics on this are fairly heavy and yes, I use Excel to do the calculations.  There are some free programs that will do the heavy lifting for you.  Here is what you need to do for a gauge R&R study:

1. Make sure that your measurement device is functioning correctly, calibrated, and the same device is used for all tests.
2. Three operators is optimal - you can use two but the results will not be as accurate.
3. Ten samples that each operator will test three times (mix the samples up so the operators will not know which sample is which).  These samples should be representative of your process.
4. Run the test and plug in the results into your program.
5. You will need to capture the average and ranges for each sample on each test as well as the average for each operator on each of the three tests.

What will come out of the program is the variability of the samples vs. the operators in both numerical and graphic terms.  What you want to see is a total variability below 30%.  Let us take our Jelly Donut example where we tested ten donut weights - the information from an Anova test could look like:

% Gauge R&R = 36.8%
-Reproducibility=-0.4%
-Repeatability=13.9%
-Donut to Donut=86.5%
So, the total number of 36.8% is much higher than the needed 30% - OUCH! Reproducibility is the measure of the variation from the operators - Great! if we had a higher variation for the operators we could look at how the operators are performing the weight checks. The repeatability is variation from the equipment - OOPS - at 13.9% this needs to be corrected.  Either the equipment is faulty or maybe not sensitive / accurate enough.  Looks like we need a "more accurate scale".  The Donut to Donut variation is 86.5% which is good.  You want most of the variability to be found in the product.  As the variation in the equipment and operators decreases, more variation will be accredited to the product.

What about graphs?  Everyone likes to see a visual that summarizes and makes the results very clear.  From the above example we could look at the variation between the operators based on the averages of the samples.

From this Operator 1 and 3 are close to each other but operator three has more variation in the average weights of their samples.  Operator two is quite different from the others.  Visualizing this definitely pin points the need of retraining operator 2.

 

 

The last item to discuss is the type of samples you need to perform this analysis.  The above works really well on non-destructive testing.  If the test is destructive, you will need to do one of two things;

1. Find a method to simulate the destructive test non-destructively.
2. Find enough "identical" samples to run the test