The Road to Six Sigma: Applying Statistical Process Control Tools

Bizmanualz Solutions

Part 1 of a 3 part series

In the manufacturing world, statistical process control (SPC) has been effectively used for decades. Walter A. Shewhart at Western Electric in the 1920’s introduced the concepts of “Assignable Cause�? and “Chance Cause�? each of which could be responsible for variability in a process or product. Chance cause, often referred to as common cause, produces a random and predictable variability; while assignable cause produced unstable and unpredictable variability.

Common Cause and Assignable Cause

A system in which only common causes of variability is present will produce a stable and predictable output. If this variability in the output is within the limits required by the product or customer, then the number of problems or defects produced by the system will be vanishingly small. Common cause is present in all systems and processes. It is possible to reduce, but not eliminate, common cause by redesigning the system.

Assignable cause variability is different in that it is possible to find and eliminate the causes of that variability from a system. If assignable cause is present in the system, the output will be unstable and unpredictable. Such a system is unable to meet product or customer requirements. In order to control the output of such a system, we must determine when assignable cause variability is occurring, and eliminate or compensate for it. Fortunately, Doctor Shewhart gave us not just the concepts, but a simple tool for separating assignable cause from common cause.

Origins of the Control Chart

On May 16, 1924, Shewhart prepared a memo, little more than a page in length, including a diagram which we now recognize as a control chart. The control chart solved the problem of deciding when to take action to correct the system, and when to leave the system alone. If only common cause variability is present in a system, adjusting or correcting the system will produce greater variability, not less. When assignable cause variability is introduced to the system, we must know about it as soon as possible in order to determine the cause and correct for it.

Shewhart’s control charts were adopted by the American Society for Testing and Materials in 1933, and advocated during World War II for improving quality and productivity. After the war, Shewhart’s methods were largely ignored in the U.S., but a Shewhart disciple, W. Edwards Deming, took those methods to Japan and implemented them with remarkable success. As a result, U.S. manufacturing quality steadily deteriorated, while Japan’s manufacturing quality steadily improved.

During the early 1980’s, U.S. auto manufacturing woke up to the fact that Japan was a serious competitor with a significant quality and cost advantage. This led to wholesale efforts to improve U.S. quality by installing new quality management systems and quality standards using the work of Shewhart, Deming and others.

Road to Six Sigma

Today many organizations have implemented a rigorous application of statistical methods called Six Sigma. Six Sigma is not for everyone because it requires extensive, specialized training in statistical methodology, and a thorough understanding of the processes involved. For service industries, especially small to medium organizations, this is often not immediately possible or practical. Still, much of the value of Six Sigma can be achieved with a few simple statistical tools.

The four tools we will introduce are the:

Pareto Diagram
Cause-Effect Diagram
Scatter Diagram
Control Chart

Pareto Diagrams

The Pareto Diagram is based on the empirical observation that a large majority of the results of a system are attributable to a few actors within the system. Often called the 80-20 rule, the Pareto’s Principle is a way of separating the “vital few from the trivial many�?.

Silicon Wafer Example

In a large electronic materials manufacturing plant, the reject rate for large diameter silicon wafers increased over a period of a week from near zero to about 17% resulting in waste. The defects were scars on the surface of the wafers, found during inspection after a final polishing step and just prior to packing and shipping. Incoming wafers to the polishing step showed no evidence of scarring.

Clearly the polishing operation was causing the damage, but how? And why was the reject rate relatively constant. A number of actions were taken to try to resolve the problem, including cleaning of the polishing wheels, closer inspection of the abrasive polishing medium, and others. Some improvement was noted immediately after making these changes, but performance deteriorated almost immediately thereafter. Finally, someone got the bright idea of sampling the output of each of the six polishing machines individually.

Guess what? The reject rate for the number four polisher was about 100% and the rejects from the other five machines were near zero! Without going into great detail, it was found that rust from an unpainted structural steel beam over the number four polisher was flaking off and dropping onto the polishing wheel. It was noted that the high reject rates began several days after an unusually heavy rainstorm.

Sewing Machine Example

In another example, a large discount retailer was disappointed in the sales volume of a new line of sewing machines. When first introduced, the machines sold well, but the anticipated sales growth goals were not being met. There were six models in the new line and management felt, not having reviewed the sales data in detail, that the problem was common to the entire line.¹

To test this assumption, sales figures were broken out for each model in the new line. Table 1 shows the results of this analysis.

Table 1 – Quarterly Sales vs. Target

	Model
	1	2	3	4	5	6
Quarterly Sales Goal	26,000	20,000	20,000	18,000	18,000	18,000
Actual Sales	15,000	18,000	18,000	19,000	22,000	21,000
Revenue Shortfall	11,000	2,000	2,000	-1,000	-4,000	-3,000

Pareto Diagram

The Pareto Diagram on the right illustrates graphically that the problem with sewing machine sales is clearly not due to models four, five and six. The problem is most severe with model one and, to a lesser degree, with models two and three.

In looking at monthly sales figures from introduction to the present, management found that all models had sold well initially, but after two months, sales of model one had begun to drop significantly. Sales of models two and three also dropped but not to the degree of model one. Sales of models four, five and six showed steady growth through the period examined.

Why would sales of model one, the expected value leader for the line show such poor sales performance? It had the lowest price and most of the same features as the higher priced models four, five and six. And why would sales of models one, two and three start out well then decline?

Without going into the analysis details, what the managers found was that because of the high volume expected for model one, the commission rate for that model was set significantly below that of models four, five and six. The commission rate for models two and three were somewhat below that of the higher priced models, but not as much lower than model one. The managers concluded that the sales force did not recognize the differences until they began receiving their commission checks, one month after the introduction. Subsequently, their corrective action focused on selling the machines that provided the higher commission rates. A simple change in the commission structure solved the problem.

Few practical problems are solved as easily using Pareto analysis. But Pareto gives us a start in separating a complex problem into manageable pieces and setting priorities. Next week we will talk about Cause-Effect diagrams and how they can be used with scatter diagrams to establish the most likely cause of a problem.

Learn more about developing policies, procedures and processes, or improving your organization by attending the next Statistical Process Control, Implementing Lane Thinking or How to create well-defined processes classes. To address other training needs, please visit the Bizmanualz website.

Sphere: Related Content