Correlation Between two OGP optical comparators

I am working on a project with many small stamped and molded components. We developed the product and processes in the US and are now running production in China. My customer requires me to pass a correlation between my measurement system in the US and my measurement system in China. Both are OGP optical comparators.

We selected and numbered 30 pieces of each component and measured all the critical to function dimensions in the US and shipped the parts to China where they were measured again. I am running correlation analysis and a fitted line regression plot to compare my data sets.

My customer has the following three requirements to pass correlation:
1. R-squared >0.75
2. Slope SE to 1
3. Bias SE to 0

As a bit of back story, I have only been a quality engineer for 9 months so I am still learning the statistics side of my job. My background is 27 years in manufacturing process engineering with an emphasis on continuous improvement. I am Greenbelt certified twice over.

So to get to my question, the results of my analyses are showing very low R-squared values and low slope values. However, when the total part to part variation is .002 to .01mm acros a group of 30 pieces (min to max) the data looks more like a cloud than a line. Will I ever be able to pass a correlation between these two machines without getting my measurement error to something far less than my part to part variation? I really believe that based on the high accuracy of these parts that I will never be able to pass correlation.

I appreciate any input you may have and would be happy to post a couple of sample data sets if it would be helpful.

 

So is there a way to calculate technical significance to pass correlation?

 

As I mentioned, I'm still still learning the quality engineering role so I will do my best to explain what I mean by technical significance.

Since the part to part variation is too low to pass correlation on the basis of statistical significance, I would like to define a range of technical significance that shows that my data is at least as accurate as it needs to be to pass an MSA. Maybe it would be defining a window equal to 30% of my tolerance range that is centered about the sample mean and making sure my data falls within that range. Just a thought. Somehow I need an alternative to present to my customer that proves both of my OGP machines are capable. I'm open to any recommendations you may have.

 

Here is a sample of the data. DGN and HBG are the two machine locations. Each two column set of data is a different feature location on the parts.

 

• What type of feature are you measuring?
  (ie: radius measurements of small degrees of arc will never correlate)
• Are you using edge detection on both machines? The exact methodology used can cause correlation issues.
• Is this metric data you sent?
• Looking at the first two locations, there is a flier in sample #6. Do you have the option to re-measure obvious fliers? But even without the flier, the data looks more like technique not equipment.

 

1. we are measuring the width of a square hole in an injection molded part.
2. We are using edge detection on both machines.
3. The data is metric.
4. We do have the option to remeasure parts but for the purpose of this study will probably just remove the fliers. I would agree that the data does indicate an issue with operator technique. We have higher turnover in China and therefore our operators have less experience.

 

• Are you checking the hole at plus or minus draft? I'm assuming plus draft if its a thru hole.
• Have you verified that the methodology is the same at both locations? Hopefully, if it's run from a program, the programs are identical. If not, a precise work instruction should help.
  
  • Number of points taken & where
  • How is the number calculated (corner to corner, mid-pt to mid-pt, line to mid-pt, etc)?
  • Are parts are at the same temperature (parts straight out of the press will be a lot different after cooling)
• How is the part fixtured at each location?

 

Good day Ken Johnson;
Have you considered gage R&R studies at both locations? You may determine that the part and/or other variations are beyond what your customer is stipulating. This information would be helpful if indeed you need to approach the customer refuting/negotiating their requirements. If this scenario proves to be true, then your data supported response will allow you (your organization and the customer) to determine accurate criteria based on the product and equipment being utilized.

Hope this helps.