We All Need Control
©2021, George J. Irwin. All rights reserved.

Oh, no, it's the dreaded Control Chart topic. We all need control, though, specifically process control. I'll be as gentle as possible here. There are two things we learn from control charts. First, is the process "in control" meaning is the variability random or because of something else - like an improvement that we want to positively impact the process. Second, is the process performing in accordance to the customer requirements. Simple, right? So why does a control chart look so complicated to the non-statistician?

Let's break it down. There are five horizontal lines in your typical control chart. The one smack in the middle is the process mean. (I know I've talked about using caution with means, but it's generally OK this time.) Then there are what are called Upper and Lower Control Limits, also called the "Voice of the Process." I'll skip the math hereó there are lots of places to find itó but they are set up so that they are three sigma above and below the mean line. And finally there are the Upper and Lower Specification Limits, also called the "Voice of the Customer." These represent what the expectations of the process are, and they are not mathematics based but requirements based.

Here's a simple and tasty example. Thrifty Bakery has a contract with Happy Dollars Discount Store to supply chocolate chip cookies, which states that each chocolate chip cookie must have ten chips, plus or minus one. So nine chips in the cookie is the Lower Specification Limit and if it "goes to 11," that's OK too as it's the Upper Specification Limit. Anything else and Happy Dollars won't be happy.

As the Black Belt, you collect a sample of cookies and count the number of chips in each. The more cookies, the more accurate the control chart will be; thirty is about the lowest reasonable number. After calculating the Upper and Lower Control Limits, create the graph. From there, there are a series of tests to check whether the process is "in control" statistically speaking. The most obvious one is finding that any data point (number of chips in a cookie) is below the Lower Control Limit or above the Upper Control Limit, but there are others. Note that this does not directly connect with the Specification Limits! So there are, at a high level, these possible outcomes from the reading of a control chart:

The process is in control and within the Specification Limits: Good.

The process is not in control but within the Specification Limits: Possible, but for cookies, this probably won't happen.

The process is in control but outside the Specification Limits: Bad. For example, Thrifty Bakery consistently puts seven chips in a cookie, but Happy Dollars wants nine, ten or eleven. Time for a conversation with the bakers.

The process is out of control and outside the Specification Limits: Really Bad. Looks like there really needs to be some Serious Process Improvement.

More from the Notebook...