Capability in sewing processes CP,CPk,PP,PPk

[Carlos52mx]

Hi community, I’m opening this discussion to talk about how to manage the capability studies in sewing processes.

For some processes like clothes industry, it is not a big issue but in case of some processes like airbags for cars, life jackets,etc it become critical to fill the specifications and assure the capability of the processes.
The question is, with the bellow values and information, is it possible to calculate a capability study?
Specification: 32 to 38 stitches in 100 mm
Machine: CNC sewing (it have very little variation)
Result of the measurement of 30 pieces:

35 35 35 35 35 35 35 35 35 35 36 35 35 35 35 35 35 35 35 35 35 35 35 35 35 35 35 35 35 35
Cp=?
Cpk=?

 

[optomist1]

Hi Carlos, welcome to the Cove, please share as much process related info as possible, this will illicit better and more responses....btw, this is a some what controversial subject within the Cove group, it usually promotes a good healthy dialogue... also what s/w are you using Minitab, excel,...

 

[Bev D]

First I would ask why you feel the need to calculate a process capability index? I would not expect a CNC sewing machine to have much variability in the number of stitches across some distance. How does a Cp or Cpk number improve your understanding of the process?


Things to think about:

The numbers you are reporting are a COUNT of the stitches. Therefore you have categorical (aka Attributes) data and not continuous (aka Variables) data. The formulas are different and you need to understand that. But with no variation in your result you will be dividing by zero no matter which formula you use.

Your sample size is incredibly low. Especially for count data.

What was your sampling scheme? Was it random? Did you sample across many set-ups? Different machines?

 

[Carlos52mx]

Hi Carlos, welcome to the Cove, please share as much process related info as possible, this will illicit better and more responses....btw, this is a some what controversial subject within the Cove group, it usually promotes a good healthy dialogue... also what s/w are you using Minitab, excel,...

I´m ussing minitab

 

[Carlos52mx]

First I would ask why you feel the need to calculate a process capability index? I would not expect a CNC sewing machine to have much variability in the number of stitches across some distance. How does a Cp or Cpk number improve your understanding of the process?


Things to think about:

The numbers you are reporting are a COUNT of the stitches. Therefore you have categorical (aka Attributes) data and not continuous (aka Variables) data. The formulas are different and you need to understand that. But with no variation in your result you will be dividing by zero no matter which formula you use.

Your sample size is incredibly low. Especially for count data.

What was your sampling scheme? Was it random? Did you sample across many set-ups? Different machines?

Hi thanks for your feedback

I need to calculate the cpk index because in the automotive industry is a requirement to show that the process and machines are capable to build pieces with same quality
In regards of the variation, yes, we have variations in the values collected (it was only a sample of the values, here the complete values)
we get the values in a continuous production of 150 pieces (see the complete values in the excel file)
Randon: no random values, it was collected in a continuous production
many set ups: No, it was used only a one set up after we adjust, we took the values.
Different machines: No, it was only one machines, one model.

See the attached excel file, that the way we calculate the cpk index, maybe it is wrong but I don´t know how to calculate the capability of the machine with this kind of values.

 

[japayson]

Hi thanks for your feedback

I need to calculate the cpk index because in the automotive industry is a requirement to show that the process and machines are capable to build pieces with same quality
In regards of the variation, yes, we have variations in the values collected (it was only a sample of the values, here the complete values)
we get the values in a continuous production of 150 pieces (see the complete values in the excel file)
Randon: no random values, it was collected in a continuous production
many set ups: No, it was used only a one set up after we adjust, we took the values.
Different machines: No, it was only one machines, one model.

See the attached excel file, that the way we calculate the cpk index, maybe it is wrong but I don´t know how to calculate the capability of the machine with this kind of values.

I am not familiar with CPK for attributes however I did find it has been discussed on the forum. Try to look here:
Method to compute Cpk for Attribute Data?

Sorry if I posted this incorrectly.

 

[Bev D]

OK - since I started this I will finish it.
jaypayson: what you posted is for converting defect rates to a Cpk value. (which is 'easy' to do, as Cpk has been so conflated with defect rates, unlike it's original intent which was to understand the process variation and location in relationship to the spec limits.)

for Carlos52MX: I don't use minitab but I can explain how to do this 'manually'. @Miner way weigh in on how this might be done in minitab...
you have count data - the count of stitches. The counts 'may' follow a Normal distribution. They may not. Count data often - but not always - follow a Poisson distribution. The data may also follow a geometric distribution - although rare, it has a long tail on the high side. The danger here is that if you pick the wrong distribution, your Cpk value may be horribly under reported as the standard deviation could be falsely large. The best thing to do here is to plot a histogram of the counts and LOOK at the distribution vs the spec limits.

Minitab probably has a distribution identifier.

You can do a rough estimate by calculating the standard deviation for the Poisson. SD = square root of the average count. then you use that SD in the normal Cpk formulas.

BUT here are the critical elements: you should span several set ups and not just look at one set up; the idea is to capture a representative set of samples that span the true variation of your process. (of course taking a single sample from a single set up will most likely have the smallest variation and it will satisfy your Customer because it will look good, it won't help you in understanding your real process capability.) You should look at each piece of equipment separately as they will likely have some systemic bias between them...

 

[Miner]

There isn't really a distribution of any kind here, continuous or Poisson. Due to the CNC process, it is almost entirely (93 - 97%) the stitch count it was programmed to provide with an occasional extra stitch. And I wonder whether that might be caused by inconsistent fabric tension while measuring. My only recommendation is to use the Nonparametric capability macro (ECAPA.mac), which can be downloaded from Minitab.

 

[Bev D]

Data never lies

 

[Welshwizard]

Hi Carlos52mx,

Could you flip the measurement issue and measure the length that x number of stitches occupies? You could for example choose the middle of your specification for x.

You could then plot these lengths on a ImR Chart, check for behaviour and use this chart to convince your customer of your quality.

Your current method is dominated by the increment which is a stitch, you could still publish this to bolster your argument.

Hope this helps