from our weekly newsletter, The Visual Thinker

2016
Posted by & filed under Quality.

Mistake-Proofing Series Part 2 of 8. Read Parts: 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8

Print This Article Print This Article
This eight-part series was co-written by Drs. Hinckley and Galsworth, under Dr. Hinckley’s signature, based on the training system they jointly developed: The SMS Method for Perfect Quality.

The Roots of SPC

Statistical Process Control (SPC) was developed in the 1920’s when companies needed a way to get better quality without the high cost of inspecting every part. SPC was touted as the way to control variation in the production process (e.g., wear, vibration, temperature, etc.). After establishing proper control, it did this by spotting and correcting drifts in the process which could result, for example, in out-of-tolerance parts (defects). Way back then, when variation was the biggest problem in manufacturing, SPC helped—and substantially. Regrettably, no one has ever validated that out-of-tolerance events are accurately predicted—or corrected—by SPC. This shortcoming means that we cannot achieve world-class quality through SPC. Let me explain.

Why the Results of SPC Are Not Predictable

graph

Figure 1: Why SPC Results Are Not Predictable

SPC is only useful if it accurately predicts and controls defects. Using a widely accepted normal distribution (bell curve) as an anchor point, SPC is supposed to tell us how many events fall outside of the control limits. For example, consider Figure 1: a hole-drilling operation where the spread of diameters is normally distributed. Note that the diameters just outside the control limits (in red) rarely cause product failures. But look at the blue square—that represents a hole that never got drilled. It was accidentally omitted—the equivalent of a hole having a diameter of zero. SPC does not and cannot predict it—because it is not a variation event. It is a mistake. And because of its distance from the control limit, we know it is a severe outcome—and from the height of the event, it happens a lot more often than SPC can ever predict.

And there are a host of other drilling errors that SPC cannot and does not predict: using the wrong drill, drilling the hole in the wrong location, not drilling the hole completely through the part, or having the drill break during drilling are but a few examples. Because all of these are mistakes, rare and random events that cannot be predicted or controlled by SPC methods, none of the SPC methods, including Six Sigma, predict and control the most severe defects.

The Evolution of SPC Methods

To be clear, Six Sigma, Total Quality Management/TQM, and SPC are all derivatives of the same statistical approach to quality—and with similar limitations. Even before Six Sigma was developed, studies showed that 95% of the companies that implemented TQM saw less than a 5% reduction in defect rates—a negligible impact on quality despite the major investments these represent. And we know why: the inability of SPC to predict and control mistakes and their resulting defects.

In Six Sigma, the predicted number of events outside the control limits (in the tails) is 3.4 events per million repetitions. The Six Sigma approach, however, has the same fundamental flaw. While claiming the ability to achieve incredibly low defects, actual defect rates are 200 to 400 times greater than predicted because, by definition, this method fails to capture the real defect culprit: those rare and random events called mistakes which are far and away the most dominant cause of modern defects. Just try to predict your next unintentional mistake, when it will occur, and what the outcome will be with any method. It cannot be done!

Historically, the precision, accuracy, and durability of manufacturing equipment has improved dramatically since the invention of SPC, while the control of the production environment has also made great strides. As a result, the role of variation in defect rates has dropped dramatically. Just as lowering the level of a pond reveals rocks that were previously hidden, reducing variation has exposed the problem of mistakes, which is the dominant quality problem today.

Confusing Theoretical Predictions with Actual Results

One medical equipment company in Utah boasts that it is a Six Sigma enterprise. In the same company, the documents used to record and track process execution have an 8% defect rate—or 80,000 defects per million documents. Assuming 10 recording operations per document, the defects per recording operation would be 8,000 parts per million—or 2,300 times the predicted defect rate calculated for Six Sigma. If your defect rate is thousands of times greater than the Six Sigma standard, then calling your organization a Six Sigma company is misleading. It is similarly misleading to assume that a hypothesis is correct without validation. In the absence of a demonstrated cause-and-effect relationship, we will never be able to predict the result of applying SPC. Using SPC is like setting the temperature on an oven, but never knowing what the temperature will actually be when the oven is used.