Issue link: https://iconnect007.uberflip.com/i/1534953
96 SMT007 MAGAZINE I MAY 2025 For Google and these large soware compa- nies, they apply something like this on smart- phones. When you type on your keyboard and get the next word suggested, this is fed- erated learning. To train these models, ide- ally knowledge from millions or billions of users is applied. However, user data must still be private and secure. at's where federated learning comes into play. Now we're trying this approach in electronics production. I can see how this approach would bring more confidence and data security to EMS companies when applying AI in a manufactur- ing space. What's next for you? In our paper, we examined whether this is applicable and offers advantages compared to a local learning approach, where each company trains its own model. is is only the first step, of course. When we have veri- fied different use cases, then for each of these use cases, it's necessary to create a suitable system so that all of this can be deployed in production and is scalable. For example, we will need something like the access rights and protocols for all the communications between client and server. is is something that has to be specifically designed for the use case and this would be the next step that we have. e paper only examines whether it would be advantageous to apply this approach. e next step is to ask how we can design a sys- tem using this approach. Thank you, Ben. SMT007 BOOK EXCERPT: The Printed Circuit Assembler's Guide to... Factory Analytics by Julie Cliche-Dubois, Cogiscan Factory analytics is all about gaining efficiency: re- gardless of the machine set, geographical location, number of sites, or production output, all manufac- turers really want to do is optimize the manufacturing ecosystem. With insights gained from what has hap- pened, is happening, and likely will happen on the production floor, these manufacturers are equipped with all the necessary intelligence to drive better de- cision making. And that's really the crux of it, getting all the data you possibly can from the factory floor to drive a more productive, and thus lucrative, op- eration. While the journey to a fruitful analytics strat- egy isn't an easy one—as discussed in earlier chap- ters, connectivity and data management challenges abound—the electronics manufacturers that priori- tize factory data and intel- ligence will be those that emerge as leaders in the coming years. The future of analytics Reimagining the Factory Analytics... and the Future in electronics manufacturing holds unprecedented potential. When we start integrating machine learn- ing (ML) and Artificial Intelligence (AI) tools into daily production processes, our expectations and hopes for "the factory of the future" will change. We're at the precipice of enormous change globally, both personally and professionally, and for manufacturers specifically, the realm of possibilities for embedding analytics with AI is absolutely incredible. Interestingly, the same challenges that plague fac- tory analytics, like data aggregation, normalization, and calculation, threaten the success of AI in manu- facturing, too. AI needs access to clean and digest- ible data to make the right recommendations to the different decision makers across the manufacturing operation. Garbage in, garbage out—AI is only sen- sical if the data feeding the system is meaningful. It's critically important that the decision makers re- lying on these intelligent systems fully understand how they are fed in order to raise a flag when some- thing seems erroneous or irregular. In the same way human error needs validation checks, an AI system requires similar checkpoints. Continue reading...