Is SPC (Statistical Process Control) useful?

GoKats78 said:

Reading all these there is a common over-riding theme..."if applied properly"

And my experience is that SPC is rarely applied properly...it becomes an exercise unto itself...in which case it is an utter waste of time and resources.

Click to expand...

This is true with anything of course.

SPC - while fairly simple - is not necessarrily easy or straightfoward. There are many hack teachers as well as persons who try to implement it without really understanding it.

A quote from Ishikawa is very appropriate: "Control charts are easy to construct so are widely used, But there are surprisingly few really useful charts.?

Control charts require a deep understanding of the process and what can and shourd be done to correct out-of-control conditions and improve the stable performance. This is not easy. This is hard work. This is science and engineering.

Bev D said:

...

SPC - while fairly simple - is not necessarrily easy or straightfoward. There are many hack teachers as well as persons who try to implement it without really understanding it.

A quote from Ishikawa is very appropriate: "Control charts are easy to construct so are widely used, But there are surprisingly few really useful charts.?

Control charts require a deep understanding of the process and what can and shourd be done to correct out-of-control conditions and improve the stable performance. This is not easy. This is hard work. This is science and engineering.

Click to expand...

My point exactly! and few want to do the hard work.

Quote from Bobdoering"
"Calculating Control Limits
They should be approximately 75% of the tolerance, centered within the tolerance."

I have very simple question that why you can get into 75% of tolerance as to build the UCL / LCL lines?

Because to determine this line is very important: This is the border of the common cause and special cause - so it also to define the local action OR the action on whole system. The confusing of those - could bring to the big lost as the actors are different: the workers shop floor OR the director!
So in fact, reading your article, I could not fully understand why the control limits can calculate like this? Following your judgement I may conclude that: better we use the "traffic chart" instead of control charts!

In that case my argumentative point stands: to spend too much time/expenses for the control charts, when it bring too little! Better to use more simple charts

JoeDM said:

I wouldn't expect shop floor personnel to be able to set up SPC. That should be done by someone trained to do it. But once set up, it is easy to use. Add up 5 numbers, move the decimal point. Subtract the high from the low. Plot the two calculations. I have been very successful training shop floor people to do this.

Click to expand...

I said that the conditions, that you have to consider when applying the charts are more difficult (not the way to build it as now simple excel file can help or even more professional like Minitab)
Can you please check back to those people - that you trained them successfully - (I guess that you use the Xbar-R chart?) and ask them to share with you the actual benefit after they apply those chart. And I will be very grateful if you can share with me such success (as I mentioned, that could bring to stable process and reduce the variation!)

TRshepherd said:

In 1972 - NBC did a White Paper on the ever increasing quality level of Japans Auto Industry, they sent a plane full of reporters and industry experts to Japan to find the answer.

One of the things they found is the massive use of SPC methods to gain understanding of their process variation.

...

My opinion of SPC has not changed in over 45 years - with the proper understanding and application it's hard to beat.

Tom

Click to expand...

At that time, the manufacturing tool is not so precise, there were not much digital measuring device ... so I think the Steward or E. Deming's control charts could be more useful!

I think with the big improvement of the machine and equipment - SPC (like other things) should be changed to help the industry in reality - not theory or only on the desk with some tutor's creative case studies and is far to the practice.

Effocus said:

I said that the conditions, that you have to consider when applying the charts are more difficult (not the way to build it as now simple excel file can help or even more professional like Minitab)
Can you please check back to those people - that you trained them successfully - (I guess that you use the Xbar-R chart?) and ask them to share with you the actual benefit after they apply those chart. And I will be very grateful if you can share with me such success (as I mentioned, that could bring to stable process and reduce the variation!)

Click to expand...

I was asked to work on a process of manufacturing close tolerance shafts (+0 -.00005 in) because the process was making parts not even close to tolerance - so bad that the parts didn't even clean up in some cases. There was a rough turning, finish turning and final grind steps.

I used SPC on the first two process steps. The first process tolerance was +/- .015. SPC showed the process to be capable of-0 / +.005, so we used SPC to control the tool. The second process had a tolerance of -0/+.002. SPC showed the process to be capable of .0005. This gave the best possible parts to the grinding process which was shown to be capable of .00008 and unable to hold .00005. To do better required better environmental controls. This was much better than the .0002 that the process produced prior to SPC.

In another case I applied SPC to control concentricity of an assembly. We were able to reduce inspection from about 150 pieces a day to 15 pieces a day with significant savings in inspection time. When the customer tightened the tolerance we were able to show with SPC data that we were already meeting the tighter tolerance.

Hope it helps.

Thanks JoeDM,

As far as I know for SPC - the UCL & LCL is defined by +/- 3sigma of the normal distribution. So to define the Control Limits based on the tolerance - somehow for me is not correct way. You properly sure that the tolerance is defined by the client and the distribution is based on the different sources of variation. We can not base on the tolerance to define the Control limits as it will bring not statistic control chart at all but the "traffic charts".
When you said:
"I used SPC on the first two process steps. The first process tolerance was +/- .015. SPC showed the process to be capable of-0 / +.005, so we used SPC to control the tool. The second process had a tolerance of -0/+.002. SPC showed the process to be capable of .0005. This gave the best possible parts to the grinding process which was shown to be capable of .00008 and unable to hold .00005. To do better required better environmental controls. This was much better than the .0002 that the process produced prior to SPC."

I really do not understand how SPC can help you when you do not calculate the UCL & LCL - so then you also can not identify what is the special causes and what is common? so what kind of actions could be taken to improve the process?

If possible - please explain me the raw data that you used in that period and I think I could clearly understand how you get into this result

Hi Effocus

Sorry to mislead you, but control limits were not calculated based on the tolerance. A couple of items of clarification:

- SPC is based on the theory that the mean of a normal distribution is normally distributed with mean equal to the process mean and standard deviation of the mean equal to process standard deviation divided by the square root of 2. Control charts control the mean and standard deviation of the process by controlling the mean and range of small samples.

-It turns out that even the means of non-normally distributed data are normally distributed so SPC is a very robust technique.

-In the example I gave, we started the first process and took some process data. From that data we calculated control limits and ran the process. From the control chart data we estimated the process standard deviation by calculating RBAR/2.326 (sample size of 5). Then we were able to show that process capability of the lathe was .005 while the drawing tolerance was +/-.015. We adjusted the tolerance to the process capability.

-In the second process the data showed that the machine tool was not capable of meeting the second process tolerance, so we changed to a CNC lathe which had much better process capability of .0005 (using the same technique as above).

-In the thrid process the grinder showed that it was not capable of meeting the design tolerance, but there was no change we could make to improve.

I find that some process variation is caused by too much machine adjustment. SPC tells the operator when to make adjustments, so there tend to be fewer adjustments. If you run the process for a while and re-calculate the control limits often you find that the control limits tighten indicating that the process in better control.

Sorry to mislead. Hope this helps.

JoeDM said:

Sorry to mislead you, but control limits were not calculated based on the tolerance.

Click to expand...

I believe he was referring to the X hi/lo-R technique, which does use the tolerance to develop the control limits. This is because standard Shewart charts may be robust, but are not as robust as a charting methodology based on the uniform distribution, which should be found in precision machining. X hi/lo-R which: captures within-variation, prevents measurement error from masking the process variation, avoids overcontrol, and provides more specific feedback to the operator as to when to make an adjustment.

This makes the X hi/lo-R far more robust than Shewhart charts when dealing with precision machining and tool wear.

Using the correct uniform distribution to understand process capability also is a more realisitic approach.

JoeDM said:

I find that some process variation is caused by too much machine adjustment. SPC tells the operator when to make adjustments, so there tend to be fewer adjustments.

Click to expand...

Absolutely, and the X hi/lo-R charting methodology does a better job of providing that feedback to the operator. It also can give leading indicators of the need to change tools prior to tool breakage, which X bar-R can never do.

JoeDM said:

If you run the process for a while and re-calculate the control limits often you find that the control limits tighten indicating that the process in better control.

Click to expand...

Actually, it will continue to tighten the control limits, causing overcontrol. It is based on using the wrong distribution and therefore wrong calculations. I prefer to use the correct, uniform distribution and capability calculations.

JoeDM said:

SPC is based on the theory that the mean of a normal distribution is normally distributed ...

Click to expand...

To clarify - this is a common misconception. Shewhart charts are NOT based on a theory of the Normal Distribution - either for the individual values or the subgroup means (Central Theorem). (They are based on the existance of a homogenous process stream or at least a rational subgrouping schema that properly allocates the components of variation.)


Effocus: you seem to have taken some experiences with poor execution of SPC as proof that SPC does not work? That's unfortunate as SPC can be a powerful tool as many of us here have experienced. But like any tool it is not a miracle application that solves everything, nor is it foolproof or easy in it's deployment.

How it is applied can vary greatly depending the science of the process that is being monitored. In some cases the operator has the direct ability to measure and correct the process without intervention from managers or engineers. In other cases, the operator has no ability to correct the process and engineers must be called in. In some cases the chart is best used on input characteristics, in others it's the output and in some it's high level 'management' data.

There are many valid chart types available for use - not all processes can or should be monitored using the Xbar R chart.

Perhaps a few specific examples of your experiences would provide an informative starting point for discussion?