Time to replace Shewhart Control charts? Computer algorithms show the way, says Dr. Juergen Ude. (Part 1)

Dr Juergen Ude Dr Juergen Ude

Historically process control has been performed with the Shewhart Control chart, such as the Individuals chart. Shewhart control charts are often hailed as the most important quality improvement tool. The popular objective of their use is to identify assignable causes, to then remove these, thereby improving the process.

The Shewhart control chart was first introduced practically 100 years ago. It is ironical that the quality profession encourages continuous improvement and yet the tools it uses are almost a century old. The time has come to ask, are the tools really timeless?Is the Shewhart chart really as effective as we are told?

Real World

I have used the Shewhart Control chart for many years in the real world. I have researched it in an academic environment and as a consultant I have seen and studied applications with hundreds of small and large organizations throughout the world. These organizations include Alcoa, Black and Decker, BHP, Coca-Cola, Hoechst, Philips, Sipca, Texas, Instruments, the second largest oil company in the United Arabic Emirates, Singapore Aerospace Technologies. The countries include Africa, Arabia, Australia, Canada, Peoples Republic of China, India, Indonesia, Malaysia, Singapore, Philippines, Mexico, New Zealand, Germany, Sri Lanka, Thailand, the USA and more. I feel that I am therefore in the unique position of knowing how successful the Shewhart Control chart really is in the practical world.

The conclusion is that contrary to the hype, the Shewhart control chart (and indeed other charts) are not as effective as alleged. There are of course exceptions and no doubt readers can report successes. I myself have helped save $3.1m for one organization using the Shewhart Chart. However, overall I have not seen many successes, instead considerable expenditure in their application with little real return. I understand that some will disagree, but that is what I have found. Often upon close scrutinization, reported successes were not due to the Shewhart Chart but the mere fact that the process was taken seriously. A simple run chart would have yielded the same results.

There are many reasons for the noted lack of success. Notably a lack of commitment, lack of follow up, incorrect applications. A disturbing tendency is to use control charts as wall paper, not as an improvement tool that requires action.

There are also serious theoretical short comings that have not been highlighted to the quality profession. Although some have challenged the Shewhart Chart, the theoretical basis of the Shewhart Chart has not been challenged extensively. One reason is that it is difficult to challenge a tool that with considerable dogmatic support is on a crest of popularity. However, if we in the quality profession, wish to also continuously improve our tools we must challenge old methods and offer better alternatives. The latter is important. We should not challenge a concept unless we can offer better alternatives or unless it will lead to a better alternative. Destructive criticism helps no one.

There are many shortcomings with the Shewhart Chart, (some are already documented) for example:

  • Shewhart Charts are insensitive. Solutions such as run test and increases is sample size, or tightening of limits have their own problems.
  • In a world, where mother nature does not practice SPC, where there is batch to batch variation, set up variation and other inherent causes for time to time variation, computed control limits are often too tight.

Series Problems

Although important, I consider these problems minor and there are ways around these problems. The most serious problem is that the Shewhart Chart was not engineered to detect the onset and duration of a change.

It only advises that a process is out of control, no more. Since surveys have concluded that 90% or more processes are out of control, no matter how good the form of process control, to be advised the process is out of control is not a significant benefit. The Shewhart chart only concludes what we know anyway, so why use the Shewhart Chart?

The Answer?

The answer is to identify causes, but this is where the Shewhart Chart falls down! Most practitioners (perhaps not skilled statisticians) assume that when the Shewhart Control Chart signals an out of control point, then this is where the problem has occurred and hence will search for an assignable cause around that point. Most of the time then causes will not be identified. The reason is that what caused the change in the process, in all likelihood, manifested itself much earlier and the search for an assignable cause should not have occurred at the location of the out-of-control signal, but earlier. ( If a cause is due to a batch change, that occurred at 10 am and we notice the effect at 1pm, searching for causes on the assumption they manifested themselves at 1pm will not identify the batch change as the cause). The fact that many practitioners, through incorrect training (and therefore no fault of their own), assume that a problem occurred only at the point where the control chart signaled a problem is a serious problem for the quality movement.

Another problem is that the Shewhart chart and indeed other popular control charts only test for changes relative to the center line or reference value, not previous process changes. If the objective is to improve the process we need to be aware of all new problems. If the process has increased and there is a new increase we need to know. A Doctor would need to know if there are signs of a new blood pressure increase to prevent a possible heart attack.

There are several other theoretical shortcomings.

As quality improvement is vital for survival in an increasingly internationally competitive environment, and recognizing the fact that the Shewhart Chart is out-of-date, I am promoting a new alternative to the Shewhart chart. The alternative is the Manhattan Control chart which is engineered for the very application that Shewhart Charts are promoted for, but not appropriate for, ie - To identify assignable causes, so that they can be removed and hence the process improved. They were engineered to detect the onset and duration of a change and to detect relative changes. They are capable of doing this because they use the number crunching power of the computer, previously not available.

Part 2 of this blog can be read here

200 Analytics Tools For Quality

Download 200 statistical analytics tools for quality improvement - Mobile Version

Improve product quality through smart data insights using new-age, machine-powered driven analytics, available to download through a suite of Apps!

  • Apps for SPC, MSA, Process Performance, Inferences, Visualization, and much more
  • No licencing or subscriptions! Pay ONLY per analysis billed monthly - don't use, don't pay!
  • Always up-to-date with the latest features