SPC (Statistical Process Control) Overview

Steve Prevette said:

I am going to attach here "The Life Cycle of a Trend". That will help explain what is happening.

The idea is the EITHER a circle, or a shift in the baseline is indication of a shift in the process.

If the center line is proven guilty, your prediction of the future based upon that center line and control limits is no longer useful. You must revise the center line and control limits so that your prediction of the future is more accurate.

Click to expand...

I see, we are using our charts for in line quality measures, real time, less for prediction going forward. Excellent paper I will use it as a reference the next time we do an SPC audit. The trouble with being an auditor is trying to maintain some level of proficiency in subjects we might only audit once a year. This is an excellent reference, thanks!

No Link.........

Last edited by Craig H. : 1 Hour Ago at 10:58 AM. Reason: Add link to the moved article

Craig H.:

I cannot get the link to work. Suggestions?

qualitygoddess said:

Last edited by Craig H. : 1 Hour Ago at 10:58 AM. Reason: Add link to the moved article

Craig H.:

I cannot get the link to work. Suggestions?

Click to expand...

Yeah, wait a few minutes 'til Marc has a chance to fix it. He just PMed me that he got the message I sent him when I realized that my handiwork was not quite working as planned. Sorry for the confusion.

Don't worry, I am not going to quit my day job....


ADDED LATER: Marc has fixed the link. It is worth a look. Thanks, Marc, and thanks, Steve!!!

qualityboi said:

I understand that the center line average is recalculated from the process data, however, how would you know if your process was out of control if you keep on readjusting the center line? We had an SPC specialist that had set up a control chart for our out of box quality defects. The center line and control limits kept changing with the data, the chart was useless to me. Maybe I missed something? Is it here that was posted that the center line is innocent until proven guilty?

Click to expand...

Steve's description of trend's life cycles is excellent (I couldn't get the link to work but I saved a version of it the last time). The only problem is that the SPC specialist may be recalculating the control limits every time new data is added. If the process is stable, there is no reason to do that unless there is a cause as described by Steve in his paper. Many software packages allow the control limits to be recalculated each time unless you fix the control limits. Some allow data to be removed from the calculation even though it still appears on the chart as an outlier.

It probably doesn't change the limits that much in the short term but I can see why it would confuse an auditor who is not sufficiently trained in SPC. A good question for the auditor to ask is what is the purpose for changing the control limits each time and what review or corrective action is being taken for the out of control points.

Bill Pflanz

Good things you do! thanks for sharing!

Steve Prevette said:

I am going to attach here "The Life Cycle of a Trend". That will help explain what is happening.

The idea is the EITHER a circle, or a shift in the baseline is indication of a shift in the process.

If the center line is proven guilty, your prediction of the future based upon that center line and control limits is no longer useful. You must revise the center line and control limits so that your prediction of the future is more accurate.

The Life Cycle of a Trend

Click to expand...

Steve,

In your previous .ppt presentation (which is great btw), why were the center lines repeatedly re-calculated in your Hanford examples? Was there a trend in each case? I could not discern that from the slides and it was a bit confusing. If you have a minute, would you mind explaining that piece for me?

Thanks,
Scott

sbickley said:

Steve,

In your previous .ppt presentation (which is great btw), why were the center lines repeatedly re-calculated in your Hanford examples? Was there a trend in each case? I could not discern that from the slides and it was a bit confusing. If you have a minute, would you mind explaining that piece for me?

Thanks,
Scott

Click to expand...

In those cases, if you look back at the baseline and UCL/LCL in effect prior to the shift, you should be able to see the significant pattern (usually 7 in a row below average was most common) at the beginning of the new baseline time interval.

There is a bit of a problem that can occur in injury trending - sometimes an injury that starts out as non-reportable later becomes reportable due to the symptoms worsen. Carpal Tunnel is a good example. It may be several years from onset of initial symptoms until surgery is needed. But, the injury is charted on the first date of reporting. So sometimes past data may change aand may remove the decreasing trend. If data are affected withing the baseline time interval, I do recalculate the baseline. On very rare occasions, I have had to remove the baseline shift and go back to the old baseline.

Origin of recalculating baseline and control limits?

Hello

I think it sounds correct to keep the same baseline until a significant change has been detected. We are currently looking at reprogramming our SPC software, and I would like to propose that we do not recalculate control limits/baseline periodically. I know, however that there are people in the company who disagree with me on this, I don't know why though.

I would like to know why literature (at least the books I have read) on this subject recommends recalculation. What is the origin of this approach?

Qaware said:

Hello

I think it sounds correct to keep the same baseline until a significant change has been detected. We are currently looking at reprogramming our SPC software, and I would like to propose that we do not recalculate control limits/baseline periodically. I know, however that there are people in the company who disagree with me on this, I don't know why though.

I would like to know why literature (at least the books I have read) on this subject recommends recalculation. What is the origin of this approach?

Click to expand...

If you recalculate the control limits despite nothing significant having changed, the control limits won't change significantly. In other words, it's mostly a waste of time but otherwise does no harm. I know that's not a justification for doing it, but it might be a justification for just letting them do it and not worrying about it.

Rebaselining Pros and Cons

But first - Do not calculate new average and control limits unless the existing one has been proven guilty by the data! Make sure your baseline has not become some sort of moving average - it needs to stay fixed in time, even if that was three years ago.

Con - People can get over anxious to rebaseline. And, in the timeframe just following a trend, you don't have much data to make a new baseline from. This may also be a production line where you want to take corrective action to get the data back to the old baseline. Dr. Wheeler tends to default to don't change the baseline unless you have both a data shift, and you know why the data shift occurred.

Pro - "The job of management is prediction" (Deming). The baseline average and control limits provide prediction. If the current baseline has been proven no good, it is no longer a good prediction. We want to detect when the data stabilize out again, and set up new predictions.

In either case, SPC is a remarkable self-healing, self-correcting process. If you shift the baseline too early, on a false alarm, the new baseline will be proven guilty and you will be back on the old baseline.