If your process is behaving in a very non-normal way even when you don't have any setups or material changes (within those first 30-50 consecutive data points), then you've got some work to do in understanding (and possibly improving) your process: where is that variation coming from in such short time-span? If, for instance, your machine has a few moulds that produce the same product, but they have slight differences, that can be the reason for non-normal distribution in a short time-span. In this case, you either fix this difference (which would improve you process capability) or accept a lower final Cmk/Cpk. To check if this mould variation is the only source of non-normality, you can make sample subgroups taking the average of one whole cycle of products (one form each of the moulds) and check if those averages are normally distributed. Do note that if you continue to use sample subgroups for your Cpk calculations later on, you'll artificially be reducing the variation, so you're no longer showing the same number (so if you were planning Cmk on single sample data points, which I think is the best way, do return to that after improving the process).
Thank you for your vid. One question i have about the scenarios: Why should I check the several factors (in the ishikawa) if the originally selected 50pcs of parts were consequtive, and where there was no material change, person change, setting change, method change...etc..? Thank you
When your Cmk is low, you’ve got a structural instability that occurs even within short spans of time - this can still be caused by the operator, when they’re not skilled in operating the machine consistently for instance; or the material, if there are significant deviations even within material batches; or settings, when settings shift during machine runtime. More often than not, the machine cannot produce a stable product quality with the available materials, but I wanted to explicitly point out that low Cmk isn’t only due to machine causes. High Cmk + low Ppk points you in the direction of machine+operator+material are able to find stable conditions for each single combination of those three, but each operator (or each material batch) produced a different stable result, so your total process is still unstable.
Hi. In regards to this would you suggest also doing CP and CPK after you do CM and CMK even if the results are good? Do you think CM and CMK is enough for PPAP submission also (i am guessing what the customer wants). Also lets say you take consecutive 50 pcs doing CM and CMK study, would this be classed as 1 subgroup size?
Yes, Cmk is just the start - if it's good, then you continue with Cpk (and maybe Ppk). Cmk is not enough for PPAP, because it doesn't tell you much about the final variation the customer is getting. For this, you'd need at least Cpk, which is a good indication of variation within each of your batches, or Ppk, which describes the longer term variation also when your customer gets different batches from you (any probably assumes that they will all be the same). 50 consecutive pieces is a subgroup size of 50, but that doesn't figure into Cmk calculations - just calculate standard deviation directly over sample sizes larger than 8 (no need to use SD estimation tables based on subgroup size).
@@TomMentink Hi Tom, thanks for the detailed explanation. In response to what you have said above. i thought if you collected x50 pcs off a run (1 part after another), then this would be classed as a subgroup size of 1 and then if you done the same an hour later (collected another x50 pcs) that would then be subgroup 2 or have i misunderstood would you have said
@@deanopenn these 50 pieces would make up 1 subgroup of size 50. Subgroup size means how many pieces/parts are within 1 sample. And again - when calculating Cmk, don't worry about subgroups, since there is just one large sample (of 30+ pieces), this has nothing to do with how many pieces you'll put into each sample subgroup for later tests.
Yes, that’s a nice idea - let me think about a good way to explain what it precisely means in practice, and keep watching the channel for your update sometime soon 😉
I’ll have to see what I’m allowed to share - I’ll probably have to strip more than just a few labels. The thing you have to look out for the most in practice is operators reacting to the process while you’re trying to set the baseline. I’ve had a measurement systems analysis where the laboratory staff could see the results of our tests, and because the first results showed quite a deviation, they already started changing the settings before we did our full sampling plan - the second half of the results did show less deviation, but they (1) overreacted and introduced a structural deviation in the opposite direction and (2) we could toss away the results, because the spread was way bigger than that system would normally have (without the lab operator touching the settings). I also had a project to reduce off cuts, where the operators were used to change the cutting height based on the off cut percentage over the last hour or so - they sort of reduced the overall off cut rate, but because we couldn’t get a clean measurement it took us a long time to notice what was really happening: when the cheese size went under a certain weight, the machine would cut one less piece, but that ‘not so sweet spot’ caused it cut a big end piece off. Once we forced ourselves to just let the machine run at one setting for a whole shift, that problem became very clear and we could set up the cutter based on the size of the incoming cheese, in stead of guessing it from the last hour’s results.
@@jb769 for a preventive maintenance plan you don’t really need Cmk. Cmk will help you to stabilize your processes and reduce variations - a very good thing by itself, but it tells you more about your process than the state of the machinery. To make an effective PM plan, the foundation is a good ABC analysis determining which equipment is vital, important or redundant - make sure you focus your PM on the vital machines first.
This is good video. My question is that how to deal with 30 or 50 data points distribution during calculation of Cmk if is not a normal distribution.
If your process is behaving in a very non-normal way even when you don't have any setups or material changes (within those first 30-50 consecutive data points), then you've got some work to do in understanding (and possibly improving) your process: where is that variation coming from in such short time-span?
If, for instance, your machine has a few moulds that produce the same product, but they have slight differences, that can be the reason for non-normal distribution in a short time-span. In this case, you either fix this difference (which would improve you process capability) or accept a lower final Cmk/Cpk. To check if this mould variation is the only source of non-normality, you can make sample subgroups taking the average of one whole cycle of products (one form each of the moulds) and check if those averages are normally distributed. Do note that if you continue to use sample subgroups for your Cpk calculations later on, you'll artificially be reducing the variation, so you're no longer showing the same number (so if you were planning Cmk on single sample data points, which I think is the best way, do return to that after improving the process).
Thank you for your vid.
One question i have about the scenarios:
Why should I check the several factors (in the ishikawa) if the originally selected 50pcs of parts were consequtive, and where there was no material change, person change, setting change, method change...etc..?
Thank you
When your Cmk is low, you’ve got a structural instability that occurs even within short spans of time - this can still be caused by the operator, when they’re not skilled in operating the machine consistently for instance; or the material, if there are significant deviations even within material batches; or settings, when settings shift during machine runtime. More often than not, the machine cannot produce a stable product quality with the available materials, but I wanted to explicitly point out that low Cmk isn’t only due to machine causes.
High Cmk + low Ppk points you in the direction of machine+operator+material are able to find stable conditions for each single combination of those three, but each operator (or each material batch) produced a different stable result, so your total process is still unstable.
Hi. In regards to this would you suggest also doing CP and CPK after you do CM and CMK even if the results are good? Do you think CM and CMK is enough for PPAP submission also (i am guessing what the customer wants). Also lets say you take consecutive 50 pcs doing CM and CMK study, would this be classed as 1 subgroup size?
Yes, Cmk is just the start - if it's good, then you continue with Cpk (and maybe Ppk).
Cmk is not enough for PPAP, because it doesn't tell you much about the final variation the customer is getting. For this, you'd need at least Cpk, which is a good indication of variation within each of your batches, or Ppk, which describes the longer term variation also when your customer gets different batches from you (any probably assumes that they will all be the same).
50 consecutive pieces is a subgroup size of 50, but that doesn't figure into Cmk calculations - just calculate standard deviation directly over sample sizes larger than 8 (no need to use SD estimation tables based on subgroup size).
@@TomMentink Hi Tom, thanks for the detailed explanation. In response to what you have said above. i thought if you collected x50 pcs off a run (1 part after another), then this would be classed as a subgroup size of 1 and then if you done the same an hour later (collected another x50 pcs) that would then be subgroup 2 or have i misunderstood would you have said
@@deanopenn these 50 pieces would make up 1 subgroup of size 50. Subgroup size means how many pieces/parts are within 1 sample.
And again - when calculating Cmk, don't worry about subgroups, since there is just one large sample (of 30+ pieces), this has nothing to do with how many pieces you'll put into each sample subgroup for later tests.
WOW! This is good, thank you!
Thanks!
Hi tom, you are doing good job,please listen your video using earphone, when you are writing something there is a irritating noise.
Thanks for the heads up - I’ll check it out.
Could you describe the noise? Is it likely the sound the marker makes on paper, or some different noise?
@@TomMentink yes exactly, it seems that either one is rough surface so its making a painful noise, you can experience it. 😊
Please make a vedio explaining the physical meaning of standard deviation formule of Cmk,Cpk and Ppk
Yes, that’s a nice idea - let me think about a good way to explain what it precisely means in practice, and keep watching the channel for your update sometime soon 😉
Here's that video I promised (decided to bump it up my time-line, because it is a very nice topic): ua-cam.com/video/e13chnE5Odc/v-deo.html
Do you have any applicable samples? Can you share with us, please?
I’ll have to see what I’m allowed to share - I’ll probably have to strip more than just a few labels.
The thing you have to look out for the most in practice is operators reacting to the process while you’re trying to set the baseline.
I’ve had a measurement systems analysis where the laboratory staff could see the results of our tests, and because the first results showed quite a deviation, they already started changing the settings before we did our full sampling plan - the second half of the results did show less deviation, but they (1) overreacted and introduced a structural deviation in the opposite direction and (2) we could toss away the results, because the spread was way bigger than that system would normally have (without the lab operator touching the settings).
I also had a project to reduce off cuts, where the operators were used to change the cutting height based on the off cut percentage over the last hour or so - they sort of reduced the overall off cut rate, but because we couldn’t get a clean measurement it took us a long time to notice what was really happening: when the cheese size went under a certain weight, the machine would cut one less piece, but that ‘not so sweet spot’ caused it cut a big end piece off. Once we forced ourselves to just let the machine run at one setting for a whole shift, that problem became very clear and we could set up the cutter based on the size of the incoming cheese, in stead of guessing it from the last hour’s results.
Do you have a webinar?
No, not yet. But I am thinking about it. Doubting between a minicourse on practical problem solving or PCS.
What were you looking for/thinking about?
@@TomMentink Im proposing to implement PM Optimisation Plan, but i think we also need Cmk data
@@jb769 for a preventive maintenance plan you don’t really need Cmk. Cmk will help you to stabilize your processes and reduce variations - a very good thing by itself, but it tells you more about your process than the state of the machinery.
To make an effective PM plan, the foundation is a good ABC analysis determining which equipment is vital, important or redundant - make sure you focus your PM on the vital machines first.