Adding a Little Context to your Data Analysis Diet
Data is Good. Actionable Information is Better.
20 December, 2020by
G5 Consulting & Engineering Services, David Schultz
| No comments yet
Context in Baseball
Growing up in St. Louis during the 1980s, it was hard not to be a Cardinals fan. They made their first appearance of the decade in 1982 and appeared again in 1985 and 1987. The 1982 Series against the Brewers was exceptionally memorable, not only because they won, but because it had been such a long time since they were in one (1968). The series went to 7 games and finished in St. Louis with Bruce Sutter throwing a strike to Darrell Porter. I remember the celebration well.
Since that time the game of baseball has changed. Most notably, as told by Michael Lewisin his book Moneyball, is that baseball is much more data driven. What used to be simple statistics, like RBIs and ERAs, has evolved into a complex data business. Knowing a batting average is pretty straight forward: divide the number of hits by the number of at bats. But what if you want to know the batting average after the 7th inning at an away game versus a left-handed pitcher. Or a right-handed pitcher? What if the opposing team is ahead? As a manager, having this information is critical when deciding what relief pitcher you are going to send to the mound. And this, is what context is all about.
Context in Manufacturing
There is a great analogy to manufacturing. Simple statistics, like time-series trend data are fairly common. When there is a manufacturing issue, people can readily check the history of critical process values during the time of the event. This is the main reason you will have a process historian. But what if you want to know the performance of a line relative to a shift or a specific product? How about the performance of two lines on the same product? What about the performance of a line per shift per batch? This will help determine what product should be made on which particular line or to investigate improving the performance of a line or shirt. This is context you don’t readily have.
Consider the picture below. If I were to ask you to find as many numbers from 1 to 100 in order in a minute, how many can you find? How long does it take you to find all of them?
Now repeat the same exercise but I will tell you that 1 is in the first quadrant, 2 is in the second quadrant, and so forth, all the way to 100. Now how many can you find? How long does it take to find them all. This is context.
In order to add context to your manufacturing information, you may need to expand your current technology stack. If you have a SCADA, a process historian and some clients tools, that is a great start. This will provide the capability to evaluate time-series data. Once you have mastered these capabilities, the next piece is system to capture events, like downtime and batch changes. This is commonly know as a Manufacturing Execution Systems (MES) or an event database. Coupled with a process historian, you can start to look at process data in context.
What you want to avoid is trying to capture this information without an event database. It is common for manufacturers to get an analytics tool to provide context. However, every query consumes network bandwidth and database processor capacity. The additional of an event engine dramatically improves response time and reduces load.
As I have previously discussed, adding context is like any other manufacturing intelligence effort. It is a good idea to keep this effort simple and built capabilities over time. Identify the desired information, like efficiency and performance and correlate it to line, batch or shift. Once you accomplish simple analyses, you can add complex analyses. You might not win the world series, but you will improve your manufacturing process.