Is SPC (Statistical Process Control) useful?

David DeLong said:

Steve:

I am a bit confused on your answer. Do you mean to say that a small number of out-of-control product close together will cause a signal rather than "defects".

Click to expand...

The time between defects can be much more sensitive (and effective) in detecting a change in defect rate rather than simply counting defects per time interval. See https://www.hanford.gov/rl/uploadfiles/VPP_TrendLow-Rate.pdf for an example which I took from Dr. Wheeler's book "Understanding Variation - the Key to Managing Chaos" (a very good introductory book on SPC, I recommend it highly).

Steve Prevette said:

The time between defects can be much more sensitive (and effective) in detecting a change in defect rate rather than simply counting defects per time interval. See https://www.hanford.gov/rl/uploadfiles/VPP_TrendLow-Rate.pdf for an example which I took from Dr. Wheeler's book "Understanding Variation - the Key to Managing Chaos" (a very good introductory book on SPC, I recommend it highly).

Click to expand...

Steve:

Sorry I misunderstood you. You are talking about attribute data rather than variable. In the automotive supplier base, attribute charts are rarely used.

David DeLong said:

Steve:

Sorry I misunderstood you. You are talking about attribute data rather than variable. In the automotive supplier base, attribute charts are rarely used.

Click to expand...

Definitely - if the defect is defined by a measurement, you are much better off plotting the measurement, rather than just analyzing the defects.

A bit more input: rare defect rates where the average count is less than one will produce an out-of control signal when the one rare event occurs. This is obviously a false alarm. Additinally many rare event conditions tend to 'cluster'. this is when 2-4 defects happen within a relatively short time (single defects with a small number of good events in between.) This is typically followed by a relatively large number of good events. (you know how "celebrities die in 3s", people have "Winning streaks" and accidents seem to come in bursts? It's not always true of course but if you plot these things out in time sequence you'll see that clusters definitely happen. Now either there is an assignable cause for the clustering or it's just the nature of random events. In any event, a better way to chart rare events is to use the number of good runs between a defect and plot that number. This distribution can be modeled by the geometric distribution (mode near zero with a long tail to the large side...)

I use the following formulas:

calculate the average RUN (it's the inverse of the defect rate by the way)
The center line of the chart is .7 * average RUN.
The UCL = (alpha/2)*average RUN
The LCL = -Average RUN*LN(Alpha/2)

This works really well - I've used it successfully for many years.

Reference: Goh, T. N., “A Statistical Procedure for Defect Control in High Quality Manufacturing”, Sensors Controls and Quality Issues in Manufacturing, ASME 1991, pp395-401.

Yes It Is A Great Tool But Only If Used For The Long Run. It Can Infact Bring Some Degradations As Well If The Results You Expect From A Shorter Period.

That is a great question!

SPC is useful for process optimization (Design of Experiments). After going through the SPC steps, the process is characterized well enough to set parameters and have them stick. SPC simplifies DOE randomization (long story).

I would encourage any reader that has a solid SPC program to cash in their chips and use DOE.

I have found that process engineers use SPC. They call it calculating "deadband".

Hope this helps,

Tom

Thanks first as this topics I tried to find since long time ago!
However I'm quite doubt of the balance for:
1. The time and expenses, that those manufacturing companies has spent for the training of SPC, data collection and made their employee correctly understand how to use the SPC - control chart ... then use the SPC with
2. The actual benefit they got.
As a TS 16949 auditor and from most of those companies I visited - the SPC actually bring to them "nothing" except some demonstration chart to the auditor! No actual process variation reduction, no special cause found for the correction! - In aonther words, without those control charts, their company still doing OK!

As you know that the condition for apply the control charts are quite specific and not easy to understand for applying, so that is the reasons now, most of the company having not-correctly apply the SPC - but they still run and operate! That is why we can re-consider the contribution of SPC. In my opinion, it take more expense than brought to the business!

Effocus said:

As a TS 16949 auditor and from most of those companies I visited - the SPC actually bring to them "nothing" except some demonstration chart to the auditor!

Click to expand...

Yes, that is one of the down sides of doing SPC to satisfy an auditor or a customer. Neither of them know your process, so it is unlikely that they are going to accurately demand SPC to be applied for those characteristics that are actually going to control the process. Nor does doing SPC to satisfy an auditor or customer encourage deep enough understanding to actually use it correctly. Generally, people do not understand it, hear horror stories like this, and do a shallow implementation to keep auditors and customers at bay. I agree, that will yield an abject waste of time and money.

But, that does not negate its value if actually applied correctly.

Effocus said:

In other words, without those control charts, their company still doing OK!

Click to expand...

Well, they probably would do "OK" without the vast overhead of meeting international quality standards, also. If "OK" is good enough, then let's just scrap all of it for an international Quality Level TCE.

minudil2000 said:

as quality engineere, could you please tell me whether statistical process control is useful or not?

Click to expand...

I think SPC is one of the easiest and most useful statistical tools. It's been around for a long time. I studied it 40+ years ago. Amazingly, when I got into industry, it seemed that nobody was using it.

I have implemented it everywhere there is any volume production. What I like is that you get process control and information about the product / process. With statistical sampling you get a pass / fail and no real data.

It's true that some companies don't use the data being generated, but that's a management problem not an SPC problem.

I do not think SPC is the easiest tool! For you may be but just image for the production floor, where they have to face with many other thing during his working day. And I think, the tool is quite difficult in the applying conditions for each chart (that in many real case are not "practicable")

As you can see in the practice, why the tool is "usefull" but very little application, and the value-added process is even less? In posting this, I do not intend to say the control chart is not correct, but I think the current way now of applying the control chart is too expensive meanwhile the value its brought is too little. So, as SPC experts - we might think of some more efficient tool, that are statistical based but applicable to daily production fields.

May be I'm not experience enough, but could you please to share with me some real example, when the production can use the control charts for the quality / productivity improvement and got actual benefit in long run (said at least 1 - 2 years)