Issue link: https://iconnect007.uberflip.com/i/1528798
14 SMT007 MAGAZINE I NOVEMBER 2024 cool dashboard. en the engineers and super- visors look at that dashboard, and magically they know what to do. ere are two challenges with that theory of where the value comes from. One, they didn't have data scientists, and it was hard to find them. Two, when the data scientists produced a really cool dashboard, their people looked at it and didn't immediately gain wisdom about what they were supposed to do from it. Finding the root cause can be challeng- ing, especially for complex defects involving multiple factors, and AI can have a role in that. But when we asked, "What's the big- gest lever to improve the factory?" the reply was, "I'm losing all my efficiency on the basic stuff. I've got four hours of downtime every day on my factory lines." It's not like those four hours need a PhD-level data scientist to explain. Typically, it's something like forget- ting to kit the right parts, and you've made that mistake a lot. You said you'd only take 10 minutes to do setup. e reality is it took you 20 minutes every single time, but you're still saying it will take you 10. It's those levels of things. You can do that analysis in the data, but because there are so many of them, you couldn't just hire a data scientist to find the one thing to fix and make a dashboard about it. You need something to address thousands of little things. at's where the AI agent makes it possible. Now, something's always watching the data and looking for these little things to improve your mean time to repair, or take your downtime from four to three hours a day. How is that happening? By determining that you're not, actually, doing a good job of consistently kitting on this line. So, let's do a better job of kitting the next day. People know how to make factories better, but one of the challenges is that you don't have enough of the, say, Six Sigma black belts or the lead manufacturing experts who know how to do these basic things really well and use data. But now these LLMs and SLMs can go through a similar thought process on top of your data and come to a similar conclusion.