The aim of this article is to try and clarify what the prerequisites are for building Quality (with a capital Q).  The first step is to understand what variation is. Once we have an understanding, we can recognize how the two approaches to variation – conformance to specifications versus consistency of  processes – are irreconcilable.
 

Quality means (1)  “On Target with Minimum Variance.”

Walter Shewhart

At the end of the eighteenth century, the fundamental problem the industrial revolution had to deal with was producing identical parts. This problem was (and still is) the source  of enormous difficulties. We all know that no two things are exactly alike, so manufacturers had to settle for making things that were “similar enough”. The concept of “specification” thus arose from the need to  define how similar two things had to be so they could be categorized as “good” (within specs) and “bad” (out of specs and therefore to be scrapped.)

This distinction did not help manufacturers to make more parts within specs, nor did it help them  understand the reason for non-conformities.

The only thing this method allowed them to do was separate the wheat from the chaff at the end of the production line. The production process was more or less what you can see in figure 1.

The inevitable consequence of this concept of variation, which we will call an “engineering type” approach, was that manufacturers perceived quality (with a small q) as something that adds complexity to the production process and so raises costs. There was nothing you could do about it, you just had to live with it!

Figure 1:  a typical process of a western manufacturing company

With the passing of time. Industry got used to the process described in figure 1,  and the production process became a competition to satisfy specs by eliminating non conforming pieces.

As a consequence, with increasing volumes manufacturers tried to widen the specs in order to sell as many pieces as possible, instead of using specs to ensure that client requests were satisfied. Engineers found themselves  caught in the middle of this conflict between the clients’ requests and the manufacturers’ desire for wider tollerances.

The conflict eclipsed the real problem, which is how to produce parts with as little variation as possible.

Manufacturers lost sight of the fact that if they produced parts that were virtually identical, there would be no need to distinguish between good and bad parts. Scrap and reworking would disappear, as well as all the expenses connected with managing  quality checks.

When you have processes that produce virtually identical pieces, i.e. with a very limited variation, the entire production cycle can be described as in figure 2.
 

Figure 2

Uniform products can only be achieved by carefully studying the process that generates them to discover sources of variation. Management’s job is to eliminate or reduce the sources of extraneous or excessive variation.

The idea of control as applied to variation was developed in the 1920s by Walter Shewhart in the Bell Telephone Laboratories. Shewhart’s work became the foundation on which W. Edwards Deming built the modern theory (and practice) of Quality Management.

In his words:

“While every process displays variation, some processes display controlled variation, while others display uncontrolled variation.” (2)

Controlled variation is characterized by a pattern of variation which is stable and consistent in time. This kind of variation is due to normal causes, i.e. that are intrinsic to the process.
 

Uncontrolled variation is characterized by a pattern of variation that changes in time. These changes are due to special causes, i.e. that are external to the process.
 

Let’s imagine a manufacturing process that produces a series of discrete pieces.  Each of these parts has a size or feature that can be measured. A periodical check is made of some of these parts and they are measured. These measurements vary as  a result of the fact that machines, operators, raw materials, working methods and the surrounding environment interact with each other, and this produces variation.

If we wanted to study this manufacturing process in depth, it would be essential to understand what type of variation it is affected by: controlled or uncontrolled. Uncontrolled variation will not only have a strong impact on the trend of data, but  will also reduce their predictability. In Shewhart’s words:

“A phenomenon will be said to be controlled when, through the use of past experience, we can predict, at least within limits, how the phenomenon may be expected to vary in the future.” (3)

The way towards Quality (with a capital Q) consists in constantly reducing the sources of variation that threaten the predictability of  processes. Predictability is thus the result of careful and constant attention to the uniformity, consistency and reliability of processes.
(This concept is not limited to the production of goods, but is also applicable to services. Quality in services is achieved in exactly the same way.)

In summary:

1. in order for data to be of some use, they must be presented in such a way that allows us to analyze them and make decisions.
2. The first problem to deal with when creating Quality in a company is understanding the type of variation the processes are affected by.
3. We can start talking about Quality only when we have predictable processes.

In conceptual terms, the “engineering type” specs approach and Shewhart’s approach have nothing in common. They have different objectives and results.

The former aims to satisfy specifications. The result of this approach is products that have uncontrolled variation which is accepted as long as it is within specifications. Shewhart’s approach, instead, aims at consistency and therefore seeks to continuously reduce  variation.

Management has to decide which of these two objectives it wants to pursue: conformance to specifications or continuous improvement of processes.

(An example of company management inspired by conformance to specifications that has become very popular in recent years is ISO 9000.)

As a conclusion to this section, we would like to illustrate the various states a process can be in.
 
 

The Ideal State

In this state the process is in statistical control and produces 100% of conforming products. What are the characteristics of a process in this state? What do we have to do to achieve it?

•  The process must be stable over time
•  The natural spread of the process must be inferior to the tolerance specified for the product.
•   We must act on the process in a stable and consistent way
•   Conditions cannot be changed arbitrarily
•   The average of the process must be set and maintained

The Threshold State

In this state the process will display a reasonable degree of statistical control but it will produce some non-conforming products; the fact that the process is in control means that the number of non-conforming products will remain more or less regular over time.
The only two ways to guarantee 100% of conforming products in this case are to
 

•  Change the specifications
•  Act on the process to reduce the variation

The Brink of Chaos

In this state the process is out of control even though it is producing 100% of conforming products. This is a particular situation. Indeed everything seems to be going fine, but the process is affected by special causes of variation that undermine the stability of the process and make its development unpredictable.
In other words, a process of this type can degenerate any moment, thus altering the quality and conformity of its output.

Chaos

Here the process is contemporaneously out of statistical control and produces non-conformity. It’s impossible to determine the percentage of non-conformity produced by this process over time. The only way to come out of this state of chaos is to first remove the special causes of variation. (4)

As  can be seen, we can give the customer 100% of what he wants operating a process which is unstable. Conversely we can consistently produce something out of specs while operating in a predictable manner.
Which is the best? In general, we can’t say. What we can say is that a state of control is not a natural state for a process and entropy does exist.
In other words, a process that presently has an outcome that satisfies the customer but  is highly unstable is very likely to degenerate into a chaotic state without much warning. This is the risk that we run if we operate our processes in a state of brink of chaos.

   1. See Donald J. Wheeler, David S. Chambers, Understanding Statistical Process Control, SPC Press, 1992, p. xix.
   2. Ibid., p. 4.
   3. Ibid., p. 6.
   4. See Wheeler & Chambers,  pp. 14-16.
 
 
 
 
 
  copyright 2000 mst-usa LLC