Thank you for sharing that (and sorry for the late reaction ☺️), I'm glad you liked the explanation. If you have any questions or suggested topics for future videos, don't hesitate to share those too - I love making videos from viewer questions (best way to ensure at least one person likes them...)
Hello, nice explanation, I have a question, machine capability study should be done during the machine qualification, specially the performance qualification, right?
Yes, I would definitely suggest that when purchasing/validating a new machine - this is probably the last chance you'll get to keep the machine manufacturer responsible for getting stable product quality at full running speed (be very strict on that the line has to be validated at full speed 😉) That said, it's not the only moment for doing Cmk/Cpk analysis - it is a tool you will also use for general process improvement projects and especially also when launching new products on an existing line. What I often see is that a line has been in a factory for years, but the organisation now wants to decrease process variation or reduce how actively operators have to steer settings. A combination of Capability studies and SPC is often used in such situations.
Hello, firstly thanks alot your answer on my question regarding Accuracy for another video! here I have a question, before doing the study of machine or short term capability, we should firstly check the stability of process and normality, is that true if my process is not normal, then no need to check the stability? are they depends each other?
They are connected, yes: If the process output is not normally distributed, you can't use normal statistics to predict future outcomes, and so you can't rely on the SPC rules. Now technically there are some other distributions too, and if your process follows them you could still say it's stable, but for 99% of processes we deal with, not normal means not stable. Especially where there's a strong break from normality (plot it in a graph, if it looks more or less normal, that's usually good enough; you don't need to do a Q:Q chart analysis or other specific test for normality). A 'secondary peak' in your histogram usually indicates that there's another distinct way your process behaves. Could be start-ups create different product, or each materials batch change does something to your process, or maybe there's an automatic cleaning process that kicks in from time to time but then also affects the products produced at that time,... Lots of possible reasons for this. A non-stable process can still generate normally distributed output. The most common example is when short term the process behaves normally, but the centreline moves up and down over time - if these moves are fluid you'll still end up with a normal distribution, if there are jumps up and down the total process with not show a normal distribution. These types of jumps are often from a change in settings (that then lasts until somebody changes them again), differences between materials batches (with the product made from the new batch all varying around one new centreline), or some other sudden but then stable change to a process parameter. Those fluid centreline shifts will also trigger SPC alarm, by the way, as they're also not stable. They just won't always show up as non-normal if the changes are fluid and often enough. I would always advise to check both normality of the distribution (visually, using a histogram) and stability of the process using a run chart. They are connected in their outcomes, but drawing both charts will help you to spot the underlying issues.
Thanks for your question. The sampling and how data points are taken into account for standard deviation determination is the same for Cp and Cpk, so the difference between sampling for Cp and Pp is the same as the difference for Cpk and Ppk. I made a video explaining most of your question: th-cam.com/video/uPfjDMjFLQE/w-d-xo.html Let me know which part remains a bit of a mystery and I’ll see how I can make that into a new video explanation.
@@TomMentink Thanks for reply.I understood perfectly. Is my one of conclusions correct that Pp can be equal or less than Cp but Pp can never be greater than Cp as Pp considers all variations?
@@Ajy77 yes, that is correct (except in some very bizarre cases, with high spread within sample groups, but no spread between sample groups - but don’t focus on that 😉)
Amazing explanation, Tom. Thanks a lot!
Thank you for sharing that (and sorry for the late reaction ☺️), I'm glad you liked the explanation.
If you have any questions or suggested topics for future videos, don't hesitate to share those too - I love making videos from viewer questions (best way to ensure at least one person likes them...)
I've watched a few of these types of videos trying to understand this. Yours is by far the best. Thank you very much. Subbed.
Thanks for your kind words, glad it was of value to you.
Hello! Really well explained. Will watch also other videos.
Thanks for your kind words. I hope you’ll find much value from my other videos too 👍
Hi Tom, thanks a lot for your great explanation!
Happy to hear that you liked the video, hope it will help you understand Process Capability better and allow you to effectively use it in practice.
Hello, nice explanation, I have a question, machine capability study should be done during the machine qualification, specially the performance qualification, right?
Yes, I would definitely suggest that when purchasing/validating a new machine - this is probably the last chance you'll get to keep the machine manufacturer responsible for getting stable product quality at full running speed (be very strict on that the line has to be validated at full speed 😉)
That said, it's not the only moment for doing Cmk/Cpk analysis - it is a tool you will also use for general process improvement projects and especially also when launching new products on an existing line.
What I often see is that a line has been in a factory for years, but the organisation now wants to decrease process variation or reduce how actively operators have to steer settings. A combination of Capability studies and SPC is often used in such situations.
Thank you for making this vedio.. and accepting my request.
Hope you liked it
@@TomMentink yes man. Absolutely it gave me a new way of seeing these things combined.
Hello, firstly thanks alot your answer on my question regarding Accuracy for another video! here I have a question, before doing the study of machine or short term capability, we should firstly check the stability of process and normality, is that true if my process is not normal, then no need to check the stability? are they depends each other?
They are connected, yes:
If the process output is not normally distributed, you can't use normal statistics to predict future outcomes, and so you can't rely on the SPC rules. Now technically there are some other distributions too, and if your process follows them you could still say it's stable, but for 99% of processes we deal with, not normal means not stable. Especially where there's a strong break from normality (plot it in a graph, if it looks more or less normal, that's usually good enough; you don't need to do a Q:Q chart analysis or other specific test for normality). A 'secondary peak' in your histogram usually indicates that there's another distinct way your process behaves. Could be start-ups create different product, or each materials batch change does something to your process, or maybe there's an automatic cleaning process that kicks in from time to time but then also affects the products produced at that time,... Lots of possible reasons for this.
A non-stable process can still generate normally distributed output. The most common example is when short term the process behaves normally, but the centreline moves up and down over time - if these moves are fluid you'll still end up with a normal distribution, if there are jumps up and down the total process with not show a normal distribution. These types of jumps are often from a change in settings (that then lasts until somebody changes them again), differences between materials batches (with the product made from the new batch all varying around one new centreline), or some other sudden but then stable change to a process parameter.
Those fluid centreline shifts will also trigger SPC alarm, by the way, as they're also not stable. They just won't always show up as non-normal if the changes are fluid and often enough.
I would always advise to check both normality of the distribution (visually, using a histogram) and stability of the process using a run chart. They are connected in their outcomes, but drawing both charts will help you to spot the underlying issues.
Thanks for video.pls explain more about cp vs Pp data capturing method and their standard deviation calculation methods.Thanks
Thanks for your question. The sampling and how data points are taken into account for standard deviation determination is the same for Cp and Cpk, so the difference between sampling for Cp and Pp is the same as the difference for Cpk and Ppk. I made a video explaining most of your question: th-cam.com/video/uPfjDMjFLQE/w-d-xo.html
Let me know which part remains a bit of a mystery and I’ll see how I can make that into a new video explanation.
@@TomMentink Thanks for reply.I understood perfectly.
Is my one of conclusions correct that Pp can be equal or less than Cp but Pp can never be greater than Cp as Pp considers all variations?
@@Ajy77 yes, that is correct (except in some very bizarre cases, with high spread within sample groups, but no spread between sample groups - but don’t focus on that 😉)