Good afternoon,

I have recently started work for a carton manufacturer supplying the food and drink market.

So that the cartons can be formed into their required shape they are run through a creasing tool which has male and female plates that indent the required creases into the paper board. We have a quality test that checks this part of the process called Relative Crease Strength (RCS). RCS is calculated as a percentage value of the average bending force of creased material (mN) divided by the average bending force of uncreased material. The tests are completed on both the Machine Direction (MD) and Cross Direction (CD) creases.

I have been tasked to improve the Cpk for this part of the process. I have just completed a Gage R&R study on this RCS test (just MD) and need some help interpretation the results. I would also appreciate comments on the assumptions I have made as to the homogeneity of the samples - as the test is a destructive test. I have assumed that every carton that is indented by the same creasing plates has the same properties (I haven't done anything to prove this). For the carton size we have used for this study the paper board fits 5 cartons across and the creasing tools circumference is 4 cartons, so 20 cartons are run through one revolution of the creasing tool. 10 of these 20 cartons were selected as our 10 samples for the R&R study and 9 revolutions worth of paper board was cut so that the 3 quality technician who currently complete the RCS tests could each do 3 repeat tests - hope this is making sense?

The results in Minitab indicate the we need to do a lot of work to reduce the measurement system error due to both repeatability and reproducibility - retraining for Quality Technicians etc.

I would appreciate any other observations or comments you may have on this data particularly linked to how I can increase the part to part percentage of variation - we are unable to select samples that cover the full range of RCS results. I am also interested in any other suggestions for improving the measurement system.

I couldn't attach the Minitab file so see S/Sheet data.

Other info:
We only have an upper spec which is 70%, although cracks will appear if the RCS value is under 40%
Each RCS result is the average of 4 RCS measurements across each carton
The Quality Techs also cut out the sample strips from each carton
The Quality Techs manually position the sample in the bend tester

Thanks in anticipation

I cannot completely follow your description of the process, so some of these comments may not be appropriate. Since the test is destructive, you definitely want homogeneity across one row of your data sheet. Row to row (part 1 to part 2) should not be homogeneous.

Can you cut 9 sample strips from the same crease of the same box? If you can do this, and randomize these amongst the operators, this would be the best. If not, can you collect these samples strips from the same crease of 2 (or 3) consecutive boxes? Each "Part" should be collected in a manner such as this.

When you collect the next Part (or row) of data, using the same method as above, the machine should be on a different revolution, and you may want to collect theses samples from a different crease. Repeat until you have your 10 Parts. If you can add additional time between the collection of Parts, this would help.

I evaluated your data in Minitab.
Your gage resolution is acceptable
There are distinct differences in Repeatability between operators. Operator 3 is significantly better than the others (see Range Chart). Study and standardize the technique used by this operator.
There are distinct differences in Reproducibility between Operators (see Xbar Chart). Standardization of Operator 3 technique should help this also.
There is no interaction between Operator and Part, so this is not an issue.