Editor’s note: The first installment of Scott Cornish’s series on process improvement appeared in the July 2006 issue of Newspapers & Technology. In this, the eighth installment, Cornish discusses the role control charts play in defining process improvement.

In the last two articles, we reviewed the six classic quality tools: check sheets, Pareto diagrams, histograms, flowcharts, cause-and-effect diagrams and scatter diagrams.

This month, we’ll begin to cover the seventh tool, the control chart. There are many types of control charts. We will cover eight of them so this will be the primary topic of this series for the next few months.

Before we start though, I’m going to cover some preliminary groundwork. That will involve a little (but not too much) history along with a little (but again not too much) theory.

Both are important as a foundation to understand and effectively use this powerful tool. We are going to get a little technical during our discussion of control charts. There really isn’t any way around that, but I will make every attempt to explain all of this at a practical level.

At the beginning

The use of control charts to monitor process control dates back to the 1920s, pioneered by Dr. Walter Shewhart of Bell Labs.

Shewhart used the charts to link special and common causes of controlled and uncontrolled variation.

These are very important terms to understand so let’s define them below.

Special causes of variation, which occur about 15 percent of the time, can be detected by simple statistical techniques. These causes of variation are not common to all the operations involved. The discovery of a special cause of variation, and its removal, is usually the responsibility of someone who is directly connected with the operation, although management is sometimes in a better position to correct. Finally, resolving a special cause of variation usually requires local action.

Common causes

The extent of common causes of variation, on the other hand, can be indicated by simple statistical techniques, but the causes themselves need more detailed analysis to isolate.

Common causes of variation are usually the responsibility of management to correct, although other people directly connected with the operation sometimes are in a better position to identify these causes and pass them on to management for correction. Of variations that do occur, 85 percent are attributable to common causes. Resolving common causes of variation usually requires some form of action on the entire system.

How can you address these forms of variation? That’s where control charts help. First, they help managers direct attention toward special causes of variation when they appear. They also reflect the extent of common cause variation that must be reduced by management action.

We are now at a critical point where we have to select the appropriate chart for the type of data — either variable data or attributable data — we have or will gather (see Newspapers & Technology, January 2007).

Variable, or continuous, data results from measurements on some continuous scale. That’s because between any two values are an infinite number of other values.

A number of charts plot variable data. They include:

•Individuals and Moving Range control charts — An example is the ink density control chart with which many of you are familiar.

•X-bar and r charts — This type of chart would be used for data that is put together as rational subgroups.

•X-bar and s charts — This type of chart is the same as the X-bar and R chart except that it uses the standard deviation of the subgroup instead of the range.

•Median charts — This chart involves plotting the median (or middle) value of a set of data instead of the mean or average.

Black ink density - Moving range chart.





Black ink density - Individuals chart.

 

Attribute data

The second type of data is attribute or discrete data, which results from counting the occurrence of events. Here is a brief description of the control charts used to track this type of data:

•The p Chart for Proportion Nonconforming is used to chart binary data of defectives where each item is in one of two categories.

•The np Chart for Number Nonconforming is used where defectives are being counted and the sample size remains constant.

•The u Chart for Nonconformities per Unit is appropriate for use when defects, rather than defectives, are counted.

*The c Chart for Nonconformities is used when defects, rather than defectives, are counted and the sample size is constant.

Let’s revisit our example project, which is to examine the processes of production and distribution to ensure a home delivery time of no later than 6 a.m. For this stage, we will have only historical data upon which to rely. Using historical data comes with a caveat, however.  It should be used with caution unless information is available about the conditions under which it was collected.

Fortunately, for our small project, data should be current and the odds are that we know how the data were collected.

Given that, let’s assume the following scenario:

During a recent meeting, a discussion occurs regarding process monitoring practices in place.

As part of everyday production, a press operator pulls a copy every 1,000 papers, measures the ink density using a densitometer, marks when the paper is pulled and saves the papers with a data sheet.

The goal, or target spec, is 1.08, which reflects the SNAP offset target of 1.05 plus 0.03 for dryback.

Industry specifications are plus or minus 0.05 from the aim point value. That means our upper specification limit (USL) for our process is 1.13 and the lower specification limit (LSL) is 1.03. The pressroom manager has data going back for a number of weeks. Column one contains a table of the data for the press run of the April 17, 2007, issue.

 The pressroom supervisor is usually on press during these runs; he watches the operator take the readings and assures us that they are accurate. Since it passed the historical accuracy test, let’s look at the characteristics of this data to determine the type of chart to use.

What kind of data

First, is this attribute or variable data? The easiest way to determine this is ask this question: Is it from measurements on some continuous scale or is it from counting the occurrence of events?

Because ink density is measured with an instrument, this is variable data.

Now let’s further investigate the data. Is it individual or grouped data? It is individual, because the data reflects one measurement area during one press run. The correct charts to use? Individuals and Moving Range.

Let’s plot Individuals and Moving Range charts. To illustrate the charts for this article, I am using the software package QI Macros (www.qimacros.com). But there are other tools, such as Excel add-ins, that will also perform these functions (see page 20).

An important observation is in order. The CL (control limit), UCL (upper control limit), and LCL (lower control limit) are derived statistically from the data. It just happens that in this example the CL of 1.08 (rounded) matches our aim point value of 1.08. That’s ideal.

But the UCL value of 1.19 (rounded) exceeds our USL value of 1.13. The LCL value of 0.97 (rounded) is closer to the LSL of 0.98 but still slightly exceeds it. What does this mean? In the case of this one press run, the process variability exceeds the specifications limits. Some additional investigation is in order.
 

Scott Cornish has more than 20 years’ experience in production and quality assurance at newspapers large and small. He can be reached at scott@practicalprocessimprovement.com.