Determine subgroup size-newbie

[Amalina]

Hello,

I'm new here looking for help and opinion regarding the subgroup size in SPC.

May i know if there any reference can be refer to decide how many sample should be include in every subsgroup.

For example, at our production, we produce products 5k per lot and we decide to get the sample total 20pieces for every subgroup as per our customer.

currently we just follow our customer to get the data for 20 pieces every subgroup.

so, i would like to know if there any reference that I can refer to decide our own subgroup size.

thank you.

 

[Bev D]

welcome Amalina!
Your question is a really great one. I would first give some advice: SPC is not a simple or easy method. You cannot just copy paste. I suggest that you begin reading the works of Donald Wheeler. You can go to his website (spcpress.com, reading room tab) to start and then proceed to Quality Digest to read his many columns.

Subgroup size is best kept small. 1 to 8 samples is usually sufficient. The sample size is driven by the nature of the process variation. Research “rational subgrouping” to understand this better. Be aware that subgroup frequency is also very important. SPC is not acceptance sampling. Teh basic statistics for each is very different.

can you tell us a bit more about what you are trying to do? It sounds like your Customer has required you to inspect 20 parts to accept or reject a lot. And now you want to apply SPC to the results?

 

[Amalina]

Hi Bev D,

Thank you for your reply… really help me.
Currently my customer required us to do SPC with subgroup size is 20pcs.

As far as i know from my customer, they are using AQL table so they would like us to implement in SPC. It is bit tough for me since i did not have any reference regarding to determine subgroup size.

At least with a good reference, i will have my own decision and reason to implement how many sample should be for subgroup size.

 

[Bev D]

AQL tables are used to determine the sample size for acceptance sampling. The samples are taken after the lot is made and are supposed to be randomly selected from the entire lot. This is not SPC at all.

Why is your Customer asking you to do SPC? Are they asking you to plot your inspection results on a control chart or are they asking you to add SPC to your manufacturing process?
Are you taking measurement data? do you record the values or only the number of acceptable parts? Do you have a occasional failures? Or do all units usually pass?

SPC uses small subgroups that are not random. They are of a specific small size taken at very specific time intervals throughout the manufacture of parts - it is not lot based.

 

[Amalina]

Actually, we had to control certain dimension at product based on customer requirement inside drawing and Cpk value should be more that 2.00. Currently, we do not have a big issue with the dimensions but my customer request to plot control chart for 20pcs since they refer to AQL.

The dimension I stated is 4.5mm + 0.1.

 

[Bev D]

This isn’t really SPC.
As I understand it your Customer is asking you to plot the 20 values on a control chart. Given the tolerance I suspect that a traditional Xbar R chart of the 20 values would exhibit a lot of out of control points. You could try a traditional Xbar, R chart and see what happens. Or you could try a “3 way chart” (Donald wheeler has a great article on this - look it up. Someone will no doubt post the link but I think you will benefit from a search).

If you could post about 20 lots worth of data in excel we could take a look for you

 

[Amalina]

Hai Bev D, sorry for late reply due to I'm not in good condition.

In the mean time, I'm still struggling to understand slowly all articles of Donald J.Wheeler for SPC. However, it is really gives me a lot of good information.

I did use traditional Xbar R chart and it is totally a lot of point is out of control.