Remember that old Creedence Clearwater Revival song,
Who’ll Stop the Rain? In it, John Fogerty states, “As long as I remember / The rain's been comin' down / Clouds of mystery pourin' / Confusion on the ground.” What’s the point here? Too much of a good thing brings confusion, lack of clarity, and misunderstanding.
Reminds me of data in the manufacturing world. Specifically, it reminds me of control charts. And too much data going into those control charts. I’ve seen it time and again. We can collect all this data, so why shouldn’t we? Right? If collecting data is a good thing, then how can collecting a lot of data be bad? Well, like anything involving human endeavors, it’s easy to overdo it when you try too hard.
Gully Washer
Never mind stopping the rain. Did you ever take the time to stop and
watch the rain? I’m not talking about a sprinkle here, I mean a real
downpour—or as we say in the South, a “gully washer.” Ok, I’ll translate that. A “gully washer” means that it rained so hard that the ditches and storm drains overflowed. This is a serious storm, wherein the rain falls in sheets so thick that it’s hard to see through. Driving a vehicle in a gully washer is downright hazardous.
The next time you see a real downpour, take a few seconds and try to focus on a single drop of rain. It’s nearly impossible. And if you were to see a single drop and track it as it hits an object, it would be hidden by other drops within a few seconds and you would lose focus on it.
Gully Washers and Control Charts
What exactly does heavy rain have to do with statistical process control (SPC) control charts? Well, I’m glad you asked! Before I drown in metaphors here, allow me to explain. Implementing SPC and collecting data too frequently can also create a “gully washer” that renders the control chart unusable. And then you’ve defeated your purpose for collecting the data in the first place.
As we all know, the purpose of control charts is to study our manufacturing processes over time. They are useful tools that have been in the hands of operators and quality pros for decades. As my colleague, Eric Weisbrod, VP of Product Management here at InfinityQS so aptly put it in his
control chart blog earlier this year, “The reason control charts were created was to give a visual indication of trends and abnormal patterns, to let you know when something may have changed from an expected baseline.”
What you’re looking to do is answer the all-important question, “Do I need to take action—Yes or No?” When you discover an anomaly in your production process, what do you do? So, naturally, when we have questions, we try to prepare as best we can to answer them. One way we prepare in the manufacturing industry is by collecting data.
Data Collection Overload
Collecting data too frequently is one of the main mistakes we make when implementing SPC control charts. There are many reasons that collecting data too frequently will cause problems; however, one key issue is that it will create process behavior limits that are too sensitive and generate false alarms. Every time we generate false alarms we tell the operator that the system doesn’t really work and therefore alarms don’t matter. This can’t happen. Operators need to trust the process and trust the control charts.
The following chart was created from data collected at the rate of one sample per minute. Please note how sensitive or tight the process behavior limits are with the majority of samples being “out of control.”
By adjusting the sampling frequency to once per hour we can see a true representation of the process’s behavior.
Ensuring that the sampling frequency is correct presents process behavior limits that give the user true signals and allow for true process control and improvement. This makes all the difference in the world to an operator. This is a process that the operator can trust. This is a system that the operator can trust.
Ensuring that the Sampling Frequency is Correct
Sampling frequency is based on how fast the process is changing. Samples must be taken often enough to catch any expected changes yet have sufficient time between samples to display the variation. The key to setting a rational sampling frequency is to understand the process. Every process has variation or normal shifting, and an understanding of this behavior enables us to more accurately sample the process.
Here are a few tips to help with setting a sampling frequency:
- Conduct a process study to help understand normal patterns of the process
- Collect data as frequently as possible during the process study to ensure the common behavior of the process is understood—don’t worry, frequent data is preferred for a process study, since the data will not be used for a control chart
- Evaluate the process study data trend to determine the amount of time or number of products produced between process shifts
- Set an SPC sampling frequency to collect two (2) subgroups between the process shifts—for example, a process that shifts every three (3) hours should be sampled every hour
Next time you are tasked with creating a sampling strategy, take a moment to think about the gully washer. Remember that the goal is to sample the process flow at a rational rate and not to overflow the storm drain!
Let’s close with some more Fogerty: “Still the rain kept pourin' / Fallin' on my ears / And I wonder, still I wonder / Who'll stop the rain?” Answer: You will…if you pay attention to how much data you’re collecting and control your sampling frequency. No gully washers!
Please feel free to read Part 2 to this blog:
Rational Subgrouping.
Take advantage of the technology at your fingertips today: contact one of our account managers (1.800.772.7978 or
via our website) for more information about InfinityQS products and services.