SMT007 Magazine

SMT007-Mar2023

Issue link: https://iconnect007.uberflip.com/i/1493797

Contents of this Issue

Navigation

Page 43 of 89

44 SMT007 MAGAZINE I MARCH 2023 boards (PCB). We thought that created an opportunity for us. When we looked at PCB manufacturing, it became apparent that while the SMT place- ment workflow was highly automated, there was a need in automating back-end production and final assembly. is laborious part of the process was where EMS companies and OEMs were still employing manual inspection. ese inspection tasks are tough to crack from a tra- ditional machine vision perspective, and we wondered, "Can AI bring anything to bear on this problem?" We spent about a year develop- ing a hardware and soware solution which fits into the typical assembly line for PCB manu- facturing. It also does post-assembly analysis, and what's really fascinating is how quickly an operator can program our product. We oen hear that AOIs are good at what they do but are laborious to program and maintain. With our system, there's not a lot of manual work. You give the system a good (i.e., golden) board—or a couple of good boards if there's a union of different components—and our AI system creates a map of where compo- nents should be in less than a minute; away you go. You can tweak it aer that, and it's striking how quickly you can configure the product. We brought our mini system to IPC APEX EXPO, and the response was fascinating. So many companies were intrigued by finally automating back-end production, and we're really excited about becoming a part of the community. There's a lot of hype about ChatGPT, and what it can and can't do. How does AI work in an industrial environment compared to AIs like ChatGPT? When ChatGPT came to the forefront, I had been invited to appear on "e National," which is the Canadian equivalent of "60 Min- utes." I commented that ChatGPT is remark- able in what it has achieved in terms of nat- ural language understanding. I saw it as an important inflection point in AI that we—or our children—will look back and say, "at was the moment when something notewor- thy happened." ChatGPT is very good at gen- erating language that human beings find con- vincing, but it doesn't always contain useful or correct information. It uses a technology that is broadly termed "generative" AI; which is essentially, when AI generates artifacts for us. In this case, the artifact is language. (ere's also DALL-E, a technology that creates images from text.) For example, if you ask ChatGPT for information on an esoteric topic it will pro- vide what sounds like a completely plausible answer. If you didn't know the correct answer, you might be convinced that it's the truth. is shortcoming is an extremely important caveat with the technology. at said, it is quite an accomplishment from the natural language perspective, and that will infiltrate many ele- ments of work. So, how does generative AI apply to an industrial context? Remember that AI requires data to learn. In our industry, it needs exam- ples of defects that are valuable to detect in a manufacturing context (missing components, bad orientations, tombstones, etc.). Given Sheldon Fernandez

Articles in this issue

Archives of this issue

view archives of SMT007 Magazine - SMT007-Mar2023