It’s easier to understand a new concept when explained in familiar terms. For example, once when I was training a group of platers about Statistical Process Control (SPC), I began by offering an example. I began talking about a shop that was plating widgets. (A term often used to represent some unspecified product.) The trainees very soon appeared to be very confused and they kept asked me to clarify. After several unsuccessful attempts to explain my point, I changed my example and said the shop was plating fasteners instead. Now they got it!

The simple device of using a familiar example helped them understand. With this in mind, I am using another common example for this SPC study that might be familiar to my readers.

I realize that not everyone works in a manufacturing facility, so here is an example you might experience at home. When my son, Bob, turned 16, I decided to teach him to drive. The car I used was a dependable, older model Chevy. I had owned it for several years and I knew it so well I could tell when the oil was dirty by the way it sounded. Bob was a special cause of variation. He made a dramatic change in the mean time to failure of the vehicle. After a few months of his involvement, my delicate old Chevy gave up the ghost, and I ended up buying a new one. It offered several advantages over the old Chevy. For one thing, it would start. For another, it got great gas mileage.

After some time, I noticed that the car’s mileage had declined.  I have certain modest ambitions as an off-road driver, and at first I thought my gas mileage might have been adversely affected by those long, grinding runs down dry riverbeds. I guessed that spinning the tires while I wasn’t moving might degrade my miles per gallon. I decided to test my theory.

Here is where SPC comes in. Statistical process control is really just a set of tools that lets you test theories. My theory was that the bad mileage might have something to do with how Bob handled the car.

The first step was to devise a way to measure the fuel efficiency. I began by checking the mileage before and after each tank of gas. For those of you younger fellows, cars didn’t always have computers to just hand you that information. It had to be done manually.

I soon discovered that my wife was sometimes driving the car, so I couldn’t determine if she or Bob might be at fault. I had to improve my measurement system. I began to keep track of my mileage after each quarter-tank of gas. I had to ask my wife to not drive the car while the test was going on. This request was met with much groaning and moaning and it meant that I had to drive to the store, take her shopping, and otherwise become a chauffer.

Troublesome as this was, it gave me quicker feedback on how I was doing. It also taught me rather quickly that my gas gauge is nonlinear, as the first quarter of the gauge shows the older model Chevy getting considerably more miles to the gallon than the second.

I decided to fill up right after one road trip and stay on good roads until the tank was empty. This would give me a baseline for driving on good roads. I pulled into the first gas station I could find after I returned to civilization and tried to fill up my tank. Again—as frequently happens in SPC studies—I ran into trouble. The gas station was very busy and the only open pump didn’t work right. After every few ounces of gas flowed into my tank, the pump would shut itself off. I kept squeezing the stupid handle until the tank was full and was soon rewarded with evidence that my theory was correct: My gas mileage was over the top. I drove farther on that tank of gas than ever before. I was content.

However, the world has a habit of thumping overconfident saps in the head. I soon learned that my theory was wrong. On my next tank of gas, my mileage plummeted. It’s at this stage in a study that you want to slap, scratch and scream a bit just to let the gods of experimentation know that you really aren’t having fun.
Still, SPC is an iterative process. You develop a theory, and then you collect data. If the information does not support the theory, you have to revise it and then collect more data. Sometimes this process goes on for months.

The next time I filled up, the pump shut off right after I started to fill the tank. For an agonizing moment I thought I had another bad pump. But then I remembered that the best mileage I ever got was right after I had filled up with the defective pump; maybe the pump had affected my gas mileage. After I had put in the customary 13 gallons, I gave the pump another shot. I found that by giving it a few ounces at a time, I could put another two gallons in my tank. Topping off like this isn’t an especially good idea, but for the purposes of my study, it was a breakthrough. I had been computing the miles per gallon based on the amount of gas I added at fill-up. I had assumed that a “full tank” was a constant, but it wasn’t. The amount of gas it took to fill my tank was an important variable and had a dramatic effect on my mileage, a special cause of variation.

The object of an SPC study is to learn how the process works. I learned that my gas gauge was nonlinear, that my driving habits affect my mileage and that my feed tube can hold an extra two gallons of gas. That’s quite a bit to discover about something you thought you already understood.

There are a great many advantages of using SPC. However, many managerial efforts prove to be neither successful nor self-sustaining. The problem is not with the approach, but rather in trying to use the wrong tool. You cannot, or should not, try to drive a nail with a box wrench. Even when the SPC is used correctly, management and workers are often impatient, expecting instantaneous results. The Automotive Industry Action Group’s (AIAG) in there second edition of its SPC Manual discusses the need to understand the underlying way that natural (common) variation plays in any process. The manual does provide instructions for basic control charts. Of course, these basic control charts have wide applications in a variety of random processes, there are many more complicated processes that require the use of more advanced charts. SPC charts serve to help organizations increase their knowledge of the common causes and special causes affecting their processes. Then they will be able to replace SPC charts with robust strategies controlling processes.

When monitoring a process, anticipate variation in the data generated. Variation occurs even when nothing extraordinary takes place. Variation that is inherent and natural is called “common-cause variation,” which affects even a stable process that is in control. But if information constantly ebbs and flows, how do you determine is something unusual has happened? SPC answers that question.

Variation can be either predictable or unpredictable. To understand whether the information varies in a predicted or unexpected way, it is necessary to understand the expected degree of normal system variation. Once you know the likely level of variability, it is possible to recognize whether your observed data exceeds the expected amount.

The idea is to observe a stable process over a long enough period to understand the level of built-in variation. Using that accumulated data, it is possible to compute limits of expected variation, defined as “control limits.” To be worthwhile, the control limits must be computed and based upon data originating from a stable process.

Control limits, also called “natural process limits,” are horizontal lines drawn on a statistical process control chart, usually at a distance of ±3 standard deviations of the plotted statistic from the statistic’s mean.

Control limits are not tolerance limits or specifications, which are completely independent of the distribution of the plotted sample statistic. Control limits describe what a process is capable of producing (sometimes referred to as the “voice of the process”), while tolerances and specifications describe how the product should perform to meet the customer’s expectations (referred to as the “voice of the customer”).

If the process is not in statistical control, then capability is meaningless. Therefore, the process capability involves only common cause variation and not special cause variation. Special cause variation is variability that happens external to the process. They are not a basic part of the process—and when they occur, they lead to a statistically significant deviation from the norm. Common cause variation is the natural variation in a process, and it defines the statistical variability of a process.

I will go into more detail about SPC in future columns.

BIO
Leslie W. Flott, Ph.B., CQE, ASQ Fellow, is certified as an IDEM Wastewater Treatment Operator and Indiana Wastewater Treatment Operator. He received his Bachelor of Science Degree in Chemistry from Northwestern University and his Masters Degree in materials engineering from Notre Dame University. Most recently, Flott served as the environmental program director and instructor at Ivy Tech Community College. Prior to that, he was the health, environment, and safety manager at Wayne Metal Protection Company.