Elsmar Forum Sponsor
R
Narfeldt,
This topic should bring some interesting comments.
From the Automotive Industry Action Group publication on Statistical Process Control page 31. "Subgroups should be collected often enough, and at appropriate times, that they can reflect the potential opportunities for change...Subgroup frequencies for ongoing production monitoring could be twice per shift, hourly, or some other feasible rate." Interesting, but not very useful.
From Statistical Quality Control by Grant and Leavenworth, page 42. "No general rules may be laid down for frequency of subgroups. Each case must be decided on its own merits, considering both the cost of taking and analyzing measurements, and the benefits to be derived from the action based on the control charts." Better, but still not much help.
In Dale Besterfield's book Quality Control he suggests that it is a cost-benefit analysis but if you absolutely need a rule of thumb use the Normal Inspection, Level IV sample size table from the old US military standard MIL-STD-414. As an example, if you produce 4000 pieces a day you should look at 60 pieces. With a subgroup size of four that would be 15 subgroups or about one every 30 minutes across an eight hour shift. If the subgroup size is 5 you would have 12 samples or one every 40 minutes.
Regards,
Rick
This topic should bring some interesting comments.
From the Automotive Industry Action Group publication on Statistical Process Control page 31. "Subgroups should be collected often enough, and at appropriate times, that they can reflect the potential opportunities for change...Subgroup frequencies for ongoing production monitoring could be twice per shift, hourly, or some other feasible rate." Interesting, but not very useful.
From Statistical Quality Control by Grant and Leavenworth, page 42. "No general rules may be laid down for frequency of subgroups. Each case must be decided on its own merits, considering both the cost of taking and analyzing measurements, and the benefits to be derived from the action based on the control charts." Better, but still not much help.
In Dale Besterfield's book Quality Control he suggests that it is a cost-benefit analysis but if you absolutely need a rule of thumb use the Normal Inspection, Level IV sample size table from the old US military standard MIL-STD-414. As an example, if you produce 4000 pieces a day you should look at 60 pieces. With a subgroup size of four that would be 15 subgroups or about one every 30 minutes across an eight hour shift. If the subgroup size is 5 you would have 12 samples or one every 40 minutes.
Regards,
Rick
C
Please let me rise some other questions.
Definitely, there is no rules for a topic like "sampling".
I suggest you to take a "progressive" approach in order to reach some optimal sampling plan. Technically:
1) Make a general view of the process through a "pre-sample". Don't worry with sample sizes and frequencies because you are not making decisions that will affect the process right now. Be careful to take a good sample that represents the majority of the conditions that rules the process.
2)Calculate the Standard Deviation of that sample.
3) Use any software that allows sample size calculations (e.g. Minitab). You must inform the "power" and the "confidence level" for that. Both of them refers to the probabilities of making wrong decisions. Declare several "high values" for them (minimum 0.7 for power and 0.9 for CL) The software will list a series of sample sizes each one associated with the probability of making wrong decisions.
4)Choose the better one that fits your budget and time restrictions (unfortunately all these calculations do not take into account such "terrenal" stuff!)
5)Start de SPC with that sample. With this sample you WILL make decisions. A bit of time will tell you if it is ok. If no, at least you will have a better estimate of the standard deviation to re-start these steps!
Is it too complicated?
Definitely, there is no rules for a topic like "sampling".
I suggest you to take a "progressive" approach in order to reach some optimal sampling plan. Technically:
1) Make a general view of the process through a "pre-sample". Don't worry with sample sizes and frequencies because you are not making decisions that will affect the process right now. Be careful to take a good sample that represents the majority of the conditions that rules the process.
2)Calculate the Standard Deviation of that sample.
3) Use any software that allows sample size calculations (e.g. Minitab). You must inform the "power" and the "confidence level" for that. Both of them refers to the probabilities of making wrong decisions. Declare several "high values" for them (minimum 0.7 for power and 0.9 for CL) The software will list a series of sample sizes each one associated with the probability of making wrong decisions.
4)Choose the better one that fits your budget and time restrictions (unfortunately all these calculations do not take into account such "terrenal" stuff!)
5)Start de SPC with that sample. With this sample you WILL make decisions. A bit of time will tell you if it is ok. If no, at least you will have a better estimate of the standard deviation to re-start these steps!
Is it too complicated?
G
A short and sassy answer is, "how much bad product can you afford to ship?" 
Really, as other members have suggested, it is difficult to say how often to take samples if you are just starting. I agree with Rick's suggestion of using MIL-STD-414 as a guide. It has been replaced by ANSI/ASQC Z1.4-1993, and can be purchased in electronic or print form from ASQ ( www.asq.org ). Once you start making measurements, examine the data to understand what it is telling you. However, be careful not to fall into the twin traps of too many samples or too few samples - either one can lead you astray.
Graeme
Really, as other members have suggested, it is difficult to say how often to take samples if you are just starting. I agree with Rick's suggestion of using MIL-STD-414 as a guide. It has been replaced by ANSI/ASQC Z1.4-1993, and can be purchased in electronic or print form from ASQ ( www.asq.org ). Once you start making measurements, examine the data to understand what it is telling you. However, be careful not to fall into the twin traps of too many samples or too few samples - either one can lead you astray.
Graeme
