Hi Tom, a really interesting video. Theoretically it all makes sense, I'm concerned about an adverse effect that such adjustments may bring about. I mean, if you have artificially increased the tolerance from 75 to 150, the natural tendency of operators, people working at the shop floor may be to think: "ohh we have such a wide tolerance which we know we will easily satisfy. So, we don't really have to bother..." Consequently, the process is deteriorating and there is no way a higher sigma level will be accomplished. So, in a real world it may look like this: Tolerance 0.75 Sigma Level 1.0 Tolerance 1.5 Sigma Level 1.0. Thanks for all this good stuff, I'm following closely. Paweł
Thanks for sharing your experience. As you may be able to tell from the video, I’m also not convinced of the big advantages of setting higher tolerance limits, certainly not if it’s just to boost Cpk. My intuition is the same as yours - if you don’t first build a strong quality culture, widening tolerances will likely lead to higher variation.
Hi Tom, Your video was quite helpful in explaining my worries. This comes in handy when I'm negotiating with a consumer. I work in the SMT industry, and our machine was designed to have CPK>=1 @ 30um, but now I understand it equals CPK>=1.33 @ 40um. Our customers are sometimes so tricky that they lower the tolerance(ie CPK>=1.33@30um), but I have no idea why. Now they can't do it any longer. :))) Really appreciate your help.
Don't forget that your customers might actually want you to improve the capability of your process, in order to decrease variation in the products they receive from you. But yes - make sure they really want that (and are willing to pay for it) and not just putting a number down.
Hi, how does a supplier predict capability on dimension when he sees x +/- 100um on print with cpk of 1.33 requirement? Does he have to produce Y parts to calculate capability first before agreeing to the spec? How does he calculate his target tolerance which will be a % of 100um that guarantees he will meet the required cpk in order to win business?
Great questions Pritish. Often suppliers will know their own process and can predict the performance of a process based on previous production runs (even if that was for a slightly different part). If it's a totally new part or even a new line, you'd have to produce parts before getting a good idea of the capability: start with measuring 30+ consecutive parts to determine Cm -> this is the best Cpk will be, since Cm/Cmk is without any process, material or parameter changes and Cm isn't even centred per se (Cmk is). +/- 100um is the range of the specification (example: 2000 +/-100um = 1900-2100um). Cpk of 1.33 means that there will be 4 standard deviations of process variation between the average product and the closest specification limit, so the variation SD will be less than 25um (if the process is nicely centred, if it's not, the SD will have to be smaller). Cpk 1.33 means that 4 SD's are within specs, that's about 99.994%, so only 6 in 100.000 parts will be out of spec (although in practice it will be more, due to the delay in catching process shift when it occurs).
If you don't have USL or LSL but you have a hard limit (target). i.e., you cannot reach levels less than zero or above 100% like tests scores? How do you calculate Cp and Cpk or even the Sigma Level of the process?
Hi Vasilis, it sounds like you have a unilateral tolerance (only 1 limit you are not allowed to cross and it's generally best for your cost to creep as close to it as possible without crossing it) - I made this video specifically about that: ua-cam.com/video/pNIWbDOAq9Q/v-deo.html The short of it, is that you don't have a Cp, but you can calculate Cpk with only that one specification limit.
Hi Tom, Thanks for the clarity about tolerances. I from automotive industry, Is it possible Cpk= 10 or more than 10 or it is some error in deciding tolerances. Please help...
Well, technically it is possible: if your standard deviation is less than 1/30th of the specification tolerance, you’ve got a Cpk of >10. It is highly unusual though. In almost all situations, the specification tolerance limits would have been adjusted by now. Because when it’s possible to make the parts so precise and standardized, the customer will start to use this in their designs and will start to need that tighter tolerance (which is not a bad thing - the final product generally gets better from it).
Hi Tom. How can I predict defect rate in a metal work that is made by cnc machine and has 6 points in it where tolerance is measured? Do you collect minimum 30 metal parts and then minimum 30 measurements from those 6 locations, and then you calculate Cpk for it? Thank you
It depends a bit on how quality is defined in your case: If they are basically 6 separate quality parameters, then Cpk is best calculated separately for all 6 and combined into 1 table for good overview. If this is a sample subgroup (that is, if the defect/tolerance is measured from the average of these 6 measuring points), then use this average result as 1 point in the Cpk calculations.
Hi Tom, a really interesting video. Theoretically it all makes sense, I'm concerned about an adverse effect that such adjustments may bring about. I mean, if you have artificially increased the tolerance from 75 to 150, the natural tendency of operators, people working at the shop floor may be to think: "ohh we have such a wide tolerance which we know we will easily satisfy. So, we don't really have to bother..." Consequently, the process is deteriorating and there is no way a higher sigma level will be accomplished. So, in a real world it may look like this:
Tolerance 0.75 Sigma Level 1.0
Tolerance 1.5 Sigma Level 1.0.
Thanks for all this good stuff, I'm following closely.
Paweł
Thanks for sharing your experience.
As you may be able to tell from the video, I’m also not convinced of the big advantages of setting higher tolerance limits, certainly not if it’s just to boost Cpk. My intuition is the same as yours - if you don’t first build a strong quality culture, widening tolerances will likely lead to higher variation.
Thank you
That's a great lesson
Great to hear that my video was valuable to you. Please don't hesitate to share it with friends and colleagues 😉
Hi Tom,
Your video was quite helpful in explaining my worries. This comes in handy when I'm negotiating with a consumer.
I work in the SMT industry, and our machine was designed to have CPK>=1 @ 30um, but now I understand it equals CPK>=1.33 @ 40um. Our customers are sometimes so tricky that they lower the tolerance(ie CPK>=1.33@30um), but I have no idea why. Now they can't do it any longer. :)))
Really appreciate your help.
Don't forget that your customers might actually want you to improve the capability of your process, in order to decrease variation in the products they receive from you.
But yes - make sure they really want that (and are willing to pay for it) and not just putting a number down.
Very interesting. Thanks
Glad you enjoyed the video. Thanks for sharing that.
Hey Tom - I've recorded one of these! I'll load mine today and we'll see how they compare....
Great, can’t wait to see it
Hi, how does a supplier predict capability on dimension when he sees x +/- 100um on print with cpk of 1.33 requirement? Does he have to produce Y parts to calculate capability first before agreeing to the spec? How does he calculate his target tolerance which will be a % of 100um that guarantees he will meet the required cpk in order to win business?
Great questions Pritish.
Often suppliers will know their own process and can predict the performance of a process based on previous production runs (even if that was for a slightly different part). If it's a totally new part or even a new line, you'd have to produce parts before getting a good idea of the capability: start with measuring 30+ consecutive parts to determine Cm -> this is the best Cpk will be, since Cm/Cmk is without any process, material or parameter changes and Cm isn't even centred per se (Cmk is).
+/- 100um is the range of the specification (example: 2000 +/-100um = 1900-2100um). Cpk of 1.33 means that there will be 4 standard deviations of process variation between the average product and the closest specification limit, so the variation SD will be less than 25um (if the process is nicely centred, if it's not, the SD will have to be smaller).
Cpk 1.33 means that 4 SD's are within specs, that's about 99.994%, so only 6 in 100.000 parts will be out of spec (although in practice it will be more, due to the delay in catching process shift when it occurs).
If you don't have USL or LSL but you have a hard limit (target). i.e., you cannot reach levels less than zero or above 100% like tests scores? How do you calculate Cp and Cpk or even the Sigma Level of the process?
Hi Vasilis, it sounds like you have a unilateral tolerance (only 1 limit you are not allowed to cross and it's generally best for your cost to creep as close to it as possible without crossing it) - I made this video specifically about that: ua-cam.com/video/pNIWbDOAq9Q/v-deo.html
The short of it, is that you don't have a Cp, but you can calculate Cpk with only that one specification limit.
Hi Tom,
Thanks for the clarity about tolerances.
I from automotive industry, Is it possible Cpk= 10 or more than 10
or it is some error in deciding tolerances.
Please help...
Well, technically it is possible: if your standard deviation is less than 1/30th of the specification tolerance, you’ve got a Cpk of >10. It is highly unusual though.
In almost all situations, the specification tolerance limits would have been adjusted by now. Because when it’s possible to make the parts so precise and standardized, the customer will start to use this in their designs and will start to need that tighter tolerance (which is not a bad thing - the final product generally gets better from it).
@@TomMentink
Thanks a lot for the clarity👍
Hi Tom. How can I predict defect rate in a metal work that is made by cnc machine and has 6 points in it where tolerance is measured? Do you collect minimum 30 metal parts and then minimum 30 measurements from those 6 locations, and then you calculate Cpk for it? Thank you
It depends a bit on how quality is defined in your case:
If they are basically 6 separate quality parameters, then Cpk is best calculated separately for all 6 and combined into 1 table for good overview.
If this is a sample subgroup (that is, if the defect/tolerance is measured from the average of these 6 measuring points), then use this average result as 1 point in the Cpk calculations.
Paul Allen just released a video answering this very same question, you can find it here: ua-cam.com/video/z72AitbrXI0/v-deo.html