Do all non-random patterns apply to attributes and variable charts including R chart?

[luis.armendariz]

Sorry for doing this but I'm trying to open another discussion in regards with "non random patterns" but it's been impossible, there is an error stating "you must select a preffix" or something and don't know how to fix it, anyways, <Split post off into new thread. - Miner, Forum Moderator>

Do all non random patters apply to attributes charts and variable charts including R charts?
I was told that attributes charts follow a normal distribution so that's why the same rules apply, but R charts supposedly follow a Chi-square distribution which is a variance distribution, not means distribution, so I was told that the only non random pattern that apply to R charts is the point ouf of control, no runs, no trends, etc.
Is that correct?

 

[Miner]

The answer is no.

Xbar/R (or s) chart: Nelson's or Western Electric Rules apply to all three types. Btw: Nelson recommended that only the first four rules be used in all cases. Rules 5 & 6 were optional rules if earlier warning was desired. Rules 7 & 8 were only intended as diagnostic tests when the chart were first set up. They provide alerts if the rational subgroups were incorrectly set up. If multiple process streams were included within all subgroups, rule 7 will alert, and if multiple process streams are sampled sequentially between subgroups, rule 8 will alert.

XMR chart: Nelson's or Western Electric rules apply to the X chart only. Points on the MR chart are not random (e.g., one point out on the X chart means two points out on the MR chart), so the one point outside the control limit is the only rule that should be used.

Attribute chart: Typically, only the first four rules apply from Nelson or Western Electric.

Final note: The more rules that you employ, the greater the chance of a false alarm. Use only those rules that are relevant to your process. For example, if your process does not trend out, do not use the rule.

 

[luis.armendariz]

Do all non-random patterns apply to attributes and variable charts including R charts

Thanks for your response, now, the following question is why?
Why it is considered rules 5 & 6 a warnings and 7 & 8 only at the beginning?

On the other hand, I realize now why Minitab 16 has only the first 4 rules for attributes charts, however, I run a I-MR chart with a run and trend intentionally on the MR chart, and Minitab showed these non random patterns, see attachement. Do you know why?

Another question that I have is why AIAG SPC second edition mentions a run as 7 points in a row, while Minitab shows 9 points, I have downloaded an article from internet and mentions also 9 points in a row, what is correct?

 

[Steve Prevette]

Do all non-random patterns apply to attributes and variable charts including R charts

I do disagree with the previous posts.

If you follow Shewhart's original reasoning for control charts - that they work regardless of the underlying distribution, and work due to Tchebychev's Inequality, then once you select a set of rules (and yes, they vary from author to author), you would use these rules consistently on EVERY SPC chart, regardless of distribution type.

 

[Miner]

Do all non-random patterns apply to attributes and variable charts including R charts

Thanks for your response, now, the following question is why?
Why it is considered rules 5 & 6 a warnings and 7 & 8 only at the beginning?

On the other hand, I realize now why Minitab 16 has only the first 4 rules for attributes charts, however, I run a I-MR chart with a run and trend intentionally on the MR chart, and Minitab showed these non random patterns, see attachement. Do you know why?

Another question that I have is why AIAG SPC second edition mentions a run as 7 points in a row, while Minitab shows 9 points, I have downloaded an article from internet and mentions also 9 points in a row, what is correct?

As to the first question, that is from the words of Lloyd S. Nelson himself in a Technical Aid printed in ASQ's Journal of Quality Technology, October 1984 entitled The Shewhart Control Chart-Tests for Special Causes. The link may require membership to ASQ.

To the second question, it appears that your process is undergoing oscillations of increasing amplitude. This could be caused by some type of positive feedback whether automated or manually induced such as process tampering. To the other, Minitab applies the rules to both charts in error. See Donald Wheeler's article What Makes the XmR Chart Work?

I believe that AIAG is using the Western Electric rules, which were empirically developed. Nelson modified these rules to provide equal probability of giving a false positive alarm. Minitab uses Nelson's Rules.

 

[Darius]

Re: Do all non-random patterns apply to attributes and variable charts including R ch

This file is an addemdum to the patterns on SPC, is an article written by Donald Wheeler.

Which rules you need to use?

Ths conclusion on that article is; "use the rules that make sense to your process".