Why the tight control limits? Histogram of the data is slightly skewed to the left

D

DJCRAI

I am not a statistical expert by any means, so I need some help. I have attached data from our process. Right now we use Measurlink for data collection and SPC. What I don't understand is why the control limits for this data are so tight.
The spec is .110" +/-.005". If you look at a histogram of this data it is slightly skewed to the left, but the mean is slightly right of the target. I am trying to get my manager to understand SPC a little better, but it is like the visually challenged leading the completely blind. He insists that the control limits are too tight. The control limits as calculated by Measurlink are LCL=.111003 & UCL=.11169 for a range of .000687. I try to explain that we need to look at the reducing any variation, yet he insists that the process is already stable. What do I do? Can someone please crunch this data and help in any way?
Thanks....
Craig H.:frust:
 

Attachments

  • spc data.txt
    1 KB · Views: 407
C

Craig H.

Craig:

What type of process does this data come from? Is there anything that would indicate rational subgroups for Xbar&R charts, or are you using an IMR? If you can use subgroups the effect of the skewness will be lowered.

Maybe you can explain to your manager that the data dictates the control limits. That is one thing that I would change if I could. Instead of "control" limits, how about "process" limits or "statistical" limits?
 
D

Darius

DJCRAI said:
He insists that the control limits are too tight. The control limits as calculated by Measurlink are LCL=.111003 & UCL=.11169 for a range of .000687.

It's a clare example of change of mean (two process are involved) or autocorrelation. Look up for autocorrelated process and rational sampling.

I still added autocorrelation to the control limits but IMHO, you are mixing 2 process setups on the same chart and it's against rational sampling.:caution:
 
D

DJCRAI

More Info....

Okay..this is a machining process on a 16 station rotary transer machine. The data is subgrouped according to the way it is listed from left to right. Each data entry in the row represents 1 part out of the possible 16 consecutive, so 10 parts per subgroup. The parts are collected every 2 hours (maximum). It is the same machine, all data within a day and a half run time. :confused:
 
C

Craig H.

Darius said:
It's a clare example of change of mean (two process are involved) or autocorrelation. Look up for autocorrelated process and rational sampling.

I still added autocorrelation to the control limits but IMHO, you are mixing 2 process setups on the same chart and it's against rational sampling.:caution:


I asked my questions before charting the data, and Darius, as usual, is quite right. This data is bimodal. From your explanation, I am not quite sure why that would be. When I have time I will group the data in the proper order and maybe that will offer a clue.
 

Tim Folkerts

Trusted Information Resource
You say you have LCL = 0.111003 and UCL = 0.11169, which would be and average of 0.111347. (First of all, are there really 6 digits in the LCL and 5 in the UCL, or is it perhap 0.11103?)

I checked the overall average and stdev of the data: x-bar = .111408 and 0.000771. My average is lightly higher than yours, but similar.

What subgroup size did you use? I'm having a hard time getting my estimate of sigma to correlate with your control limits. Also, only 31/140 points are within your limits, so you presumably are grouping your data and running an x-bar chart rather than just a I-MR chart. Perhaps that has something to do with the control limits looking too tight.


Tim F
 
D

DJCRAI

Trying again....

Sorry, the control limits as calculated by Measurlink on an Xbar & R chart are LCL=.111127, UCL=.111688 . I was lookin at a different data set the first time. Yes, Tim, The limits are exactly as shown on the chart. BTW, thanks you guys, for looking at this. The subgroup size is 10. Could the Bimodal distribution be that we don't collect the same 10 pieces? It is a 16 station machine, but due to time issue we only check 10 pieces, and I know it is not the same stations every time.
 

Jim Wynne

Leader
Admin
DJCRAI said:
Could the Bimodal distribution be that we don't collect the same 10 pieces? It is a 16 station machine, but due to time issue we only check 10 pieces, and I know it is not the same stations every time.

Yes. Each station is essentially a separate process.
 
D

Darius

Thanks Craig.

But still, I charted the data and it looks like the first 4 subgroups are from one setup coondition and the others are from other, so something changed and don't change radomly (if it changes radomly the variation would be across all samples not just a change of mean the beginning of the chart), you have to check if the first 4 subgroups are from one batch or different material, different setup conditions, etc.
 

Miner

Forum Moderator
Leader
Admin
JSW05 said:
Yes. Each station is essentially a separate process.

Your best course of action would be to construct a control chart for each of the 16 stations and evaluate it for control. I plotted the data, and saw two definite, and a possible third process mean. From your description, station to station difference in mean is a likely suspect to evaluate.
 
Top Bottom