Informational Control Chart Interpretation - General "Rules"

Geoff Withnell said:

Unfortunately, the sampling, measurement and/or gage error is part of the process! It's one of the sources of process variation. There are various methods for determining how much variation your measurement method introduces, and that variation may be one of your targets for reduction, if the process variation is unacceptable. But measurement variation is an inherent part of the the process and needs to be considered when doing statistical process control.

Click to expand...

That's the problem, it is part of the total variation, but not part of the process. The generic interpretation of the term process used in auditing only confuses the problem. Here, the process is the specific change agent - cutting, molding, stamping, etc. In order to keep the other variances from masking the process variation, you need to identify those variations and make them statistically insignificant so that only the process is being analyzed. Otherwise, you are making decisions based on error rather than process variation. Not such a good idea.

OK I have attached my data (I think) so let me know what your thoughts are. I only have 18 data points and will be collecting more tomorrow but I thought that 18 data points would be a start.

For those that are keeping score this part is being turned on an Okuma lathe using Cobalt Chrome bar stock.

Notes were made on the couple of adjustments that were made over the course of these parts.

There is a setup change over that occurs at the end of each shift/day which caused several dimensions to spike. Still trying to account for this variation.

I had time to stick the first dimension in an X hi/lo R Chart. See attached.

bobdoering said:

That's the problem, it is part of the total variation, but not part of the process. The generic interpretation of the term process used in auditing only confuses the problem. Here, the process is the specific change agent - cutting, molding, stamping, etc. In order to keep the other variances from masking the process variation, you need to identify those variations and make them statistically insignificant so that only the process is being analyzed. Otherwise, you are making decisions based on error rather than process variation. Not such a good idea.

Click to expand...

Normally, a control chart isn't used for auditing. It is used to control the process during production. And in that environment, measurement, and hence measurement error, is definitely a part of the process. The variation in the output of the process will be affected just as much by changes in the measurement variability as in the machine process. The operator HAS to "make decisisons on based on error". We don't know the source of the signal on the chart until we investigate. It may well be gage error that caused the signal. And as far as making measurement error "statistically insignificant"? Normally NOT going to happen. The rule of thumb of measurement error being ten percent or less of total variation still results in statistical significance for the measurement error.

Geoff Withnell said:

Normally, a control chart isn't used for auditing. It is used to control the process during production. And in that environment, measurement, and hence measurement error, is definitely a part of the process. The variation in the output of the process will be affected just as much by changes in the measurement variability as in the machine process. The operator HAS to "make decisisons on based on error". We don't know the source of the signal on the chart until we investigate. It may well be gage error that caused the signal. And as far as making measurement error "statistically insignificant"? Normally NOT going to happen. The rule of thumb of measurement error being ten percent or less of total variation still results in statistical significance for the measurement error.

Click to expand...

The 10% number comes from the voodoo math of the AIAG GR&R criteria and shouldn't be considered particularly useful. If you want to effectively eliminate measurement error from the equation, it first has to be quantified, and then manufacturing tolerances should be adjusted accordingly--guardbanding, in other words. This doesn't mean that unexpected measurement error won't happen; it just means that measurement error as a potential source of variation has been conscientiously accounted for.

Jim Wynne said:

The 10% number comes from the voodoo math of the AIAG GR&R criteria and shouldn't be considered particularly useful. If you want to effectively eliminate measurement error from the equation, it first has to be quantified, and then manufacturing tolerances should be adjusted accordingly--guardbanding, in other words. This doesn't mean that unexpected measurement error won't happen; it just means that measurement error as a potential source of variation has been conscientiously accounted for.

Click to expand...

It has been conscientiously accounted for, and is a part of the statistical process control limits, shown on the control chart. The control chart should not have the manufacturing tolerances. These are what have been determined to be the limits necessary for the manufacturing process to run as designed. The "voice of the industrial engineer" if you will. The specification limits also do not belong on the control chart, these are the "voice of the design engineer" and hopefully the customer. What belongs on the statistical process control chart is the calculated limits of the process, as it is defined, when in control, as seen by the operator controlling the process. It is both impossible and undesirable to eliminate measurement variability from these limits, since this error is a part of what the operator sees and is in the data he is using to control the process.

As far as the 10% being "voodoo", I said it was a rule of thumb. If I was dealing with a process where measurement variability was about 10% of total variability, I would likely start elsewhere to reduce the variation.

Geoff Withnell said:

Normally, a control chart isn't used for auditing. It is used to control the process during production.

Click to expand...

I understand it is not for auditing, I was comparing the use of the term "process" in the all-encompassing usage found in auditing and the usage in "process" control, which is the actual transformation of raw materials, not all of the peripheral items.

Geoff Withnell said:

It is used to control the process during production. And in that environment, measurement, and hence measurement error, is definitely a part of the process.

Click to expand...

I disagree - measurement is used to describe the process, and if it is becomes a part of the process, it masks the true process that is to be controlled, and makes the control unreliable.

Geoff Withnell said:

The variation in the output of the process will be affected just as much by changes in the measurement variability as in the machine process. The operator HAS to "make decisions on based on error". We don't know the source of the signal on the chart until we investigate. It may well be gage error that caused the signal. And as far as making measurement error "statistically insignificant"? Normally NOT going to happen. The rule of thumb of measurement error being ten percent or less of total variation still results in statistical significance for the measurement error.

Click to expand...

The variation of the process is not affected by the measurement, it just becomes poorly described by the measurement, hence making the measurement inadequate. To control the process, you must be able to segregate the process variation from all of the other variations by making them statistically insignificant. This is the point of Gage R&R, ndc, etc. - when properly applied.

Geoff Withnell said:

Normally NOT going to happen. The rule of thumb of measurement error being ten percent or less of total variation still results in statistical significance for the measurement error.

Click to expand...

It better happen, or you are wasting your time. If you accept an ndc < 5 for SPC, you are wasting your time. Accepting measurement systems that are not adequate is not an excuse for rolling them into the process, it is a reason to change them to adequate measurement systems. You will even mask the process distribution with measurement and gage error, and that is a terminal failure, and totally inacceptable.

Geoff Withnell said:

It is both impossible and undesirable to eliminate measurement variability from these limits, since this error is a part of what the operator sees and is in the data he is using to control the process.

Click to expand...

It is impossible to eliminate measurement error, which is why all gages are bad - but it is not impossible to have adequate gaging to ensure that the variation is statistically insignificant. If you do not, your charting quickly becomes voodoo, for sure.

Geoff Withnell said:

The control chart should not have the manufacturing tolerances. These are what have been determined to be the limits necessary for the manufacturing process to run as designed. The "voice of the industrial engineer" if you will. The specification limits also do not belong on the control chart, these are the "voice of the design engineer" and hopefully the customer.

Click to expand...

For Shewhart charts, and those subsets of processes whose distributions apply, yes. But, that is not always applicable, and therefore not hard and fast rule. We have discussed those exceptions in detail elsewhere in the forum. Precision machining is a classic example of such an exception.

bobdoering said:

I understand it is not for auditing, I was comparing the use of the term "process" in the all-encompassing usage found in auditing and the usage in "process" control, which is the actual transformation of raw materials, not all of the peripheral items.

Click to expand...

bobdoering said:

I disagree - measurement is used to describe the process, and if it is becomes a part of the process, it masks the true process that is to be controlled, and makes the control unreliable.

Click to expand...

This is an ivory tower viewpoint. The process operator measures. When indicated by the SPC and the results of the measurement, he adjusts. There was some degree of variation inherent in the measurement, and this is part of the control loop of the process. You can say it is not a part of the process all you want, but if you adjust the process input based on the measurements, then measurement variability is going to be a portion of process output variability. If you can describe to me ANY scenario of a process which is adjusted based on measurement, where the measurement variability does not end up as a portion of the process output variability, I will concede your point.

bobdoering said:

The variation of the process is not affected by the measurement, it just becomes poorly described by the measurement, hence making the measurement inadequate. To control the process, you must be able to segregate the process variation from all of the other variations by making them statistically insignificant. This is the point of Gage R&R, ndc, etc. - when properly applied.

Click to expand...

Gage R&R quantifies the gage variation. It does NOT segregate it. It is still there when the operator takes his measurement.

bobdoering said:

It better happen, or you are wasting your time. If you accept an ndc < 5 for SPC, you are wasting your time. Accepting measurement systems that are not adequate is not an excuse for rolling them into the process, it is a reason to change them to adequate measurement systems. You will even mask the process distribution with measurement and gage error, and that is a terminal failure, and totally inacceptable.

Click to expand...

Never said you should accept inadequate measuring systems. Just that, in the real world, production application of SPC, the measurement error is in the process, it's part of the process control loop, and it is in the process output.