# Help with Cpk procedure

#### Golfman25

Trusted Information Resource
We have a print requirement to maintain a cpk of 1.67 on a certain dimension. It is very difficult to achieve. We need to figure out the best way to set up and monitor the measurement.

The part is a drawn stamped cup and the dimension is the outside diameter - lets say 2.500 p/m .002. We'll run maybe 5,000 to 10,000 pcs. each batch. Right now, we set a target dimension of p/m .0012. So we set the die up and adjust until we reach inside that target -- easier said than done. Generally, once it is set, it stays at that dimension with minimal variation.

So I am trying to figure out the best way to create the spc chart and calculate the cpk for each batch. If we wait until the end of the run we may have a batch that doesn't meet the cpk -- and creates a whole mess of other problems. So I am thinking of taking 20 samples from the first 100 pcs. to start the initial calculation. Then adding like 5 samples per hour or something until the end of the run to monitor. So we will probably end up with 30 to 50 sample measurements total.

#### Bev D

##### Heretical Statistician
Super Moderator
that's a reasonable approach...I would suggest adding a standard deviation control chart for the first 20 pieces. while 20 is sufficient for the mean, its a bit small for the SD. So I'd plot the SD of the first 20 of each set-up on an SD chart. If the SD of the first 20 pieces is out of control high you will have a decent alarm that you won't make the 1.67 Cpk...
Since this type of process tends to be homogenous within a run, this should work.

#### Golfman25

Trusted Information Resource
that's a reasonable approach...I would suggest adding a standard deviation control chart for the first 20 pieces. while 20 is sufficient for the mean, its a bit small for the SD. So I'd plot the SD of the first 20 of each set-up on an SD chart. If the SD of the first 20 pieces is out of control high you will have a decent alarm that you won't make the 1.67 Cpk...
Since this type of process tends to be homogenous within a run, this should work.

Great. Thanks

#### Golfman25

Trusted Information Resource
Ok, continuing on. Let's say we are running along without issue. Then we have a "blip" -- something cause the part to go out of control (but not out of spec.). It is corrected and we finish the run. Before and after the "blip" we have a good cpk. However, if we include the blip data cpk is no longer good. How should we handle this while running?