Issue link: https://iconnect007.uberflip.com/i/1447212
20 SMT007 MAGAZINE I FEBRUARY 2022 when you do image recognition, natural lan- guage processing; this is the forefront right now, as well as very advanced algorithms that are dealing with B2C sales, marketing analyt- ics, self-driving, and so forth. is is where AI is focused because it is the biggest return. You can buy equipment that even has some of these features. You don't have to touch the concept of AI. But on the PCB and EMS side, it's more "AI light," or "more-than-Excel algorithms." Some- thing we called just a normal algorithm 20 years ago is now called AI. It's a sexy term. Robots aren't taking over the world any time soon. e fundamental things to do in a brownfield to tease out value from a planning perspective is just a step-by-step approach, one process at a time. It's probably a one- or two-person engineer- ing job. You give them subject matter exper- tise, availability, make sure they know how to do some basic coding, understand sensor options, and then they go process by process. What are some of the interpolations you can do? Well, just getting a time stamp. All you need is a photo sensor to know something went through the machine, and then you can also code at some point when it went through, when it came out. You could do this in one step or multiple steps. e sensor kit for most processes is gen- erally $5,000 to $10,000 per tool to get time and basic settings, and then you save their recorded data, which is just the serial number that you scanned, the times it went through the machine; you save it into a database. If you can pull the error codes by time out of your machine and put that in a data table, then you can correlate the two together and you can know something happened: "I heard an alarm while this panel was being processed." Mak- ing an API to your other systems adds further interpolations that can then lead to predictive recipes/decisions. e first step is understanding what is and is not important and being able to make some basic interpolation. Take AOI, for instance. is is a measurement site in PCB fab. e AOI machine has a lot of information built into it. It has all the false calls that it saw, and it says how long it took to start and finish the job. You can also code information in there on the real defects and then take all of this, put it in the database, and correlate it to your serial numbers. When you test it later, you can corre- late these serial numbers together and under- stand the root cause of escapes; a lot of escapes happen because of poor AOI setup, opera- tor errors, things like this, and it gives you an opportunity to improve your procedures in these areas to reduce scrap. is is an interesting area of focus to get a good return because in a layer, shorts and opens are typically the defects that we scrap boards for. You can reduce the frequency. Addition- ally, you have your use of regression in place of direct measurements. In PCB fab, for exam- ple, there are a lot of legacy specifications to cross-section everything. And when you have no controls in the process, cross-section is the only option. But if you add some very simple controls, you can avoid cross-sectioning just about Alex Stepinski