Issue link: https://iconnect007.uberflip.com/i/1172746
68 SMT007 MAGAZINE I OCTOBER 2019 Artificial Reliability Over Intelligence As the industry begins to shift from stan- dard design tools to artificial intelligence (AI), reliability might be overlooked in an effort to build "smarter." Over the last few years, the desire to manufacture anything and everything for less has included removing humans from as many positions as possible. There are a couple of viewpoints, and I can see positives in both. When you remove the human error rate from inspecting things like solder depo - sition or part placement accuracy, pro- ductivity and throughput can greatly increase. We have been using equipment with machine perception (think AOI, or automated optical inspection) with great success for years. Where judgment calls are important, auto - mation isn't always the answer; actual intelli- gence and experience are required to determine if things like flux and other processing res- idues are present. Even more impor- tant is the decision of what to do with those residues to determine if they pose a risk to your product's reli - ability. I have seen solder paste inspection equipment at contract manufacturers (CMs) that look at millions of pads each day that will catalog any locations that it can't fully determine if the paste is sufficient or not; then, a human needs to accept or reject that PCBA based on visual inspection experience. Paste printing is a starting point for reliability that has not been completely taken over by AI yet, and that is a good thing. The same can be said for AOI after SMT reflow. Keep in mind this is all contingent on the operators being properly trained for the pass/fail visual criteria making that call. At this time, I am not familiar with any AI equipment that is pur - posely built for reli- ability as a stand- alone; instead, there are sev- eral pieces that, when working in conjunction, can give you a clear pic - ture of your products' reliability. This requires humans to determine the full scope of a proper design of experiments (DOE) that will yield useful infor - mation outside of what AI is cur- rently capable of doing. Humans can detect small changes in the operat- ing parameters of equipment they have been using to produce hundreds of thousands of pieces of product, and that experience is what leads to better reliability—not an AI program that is Quest for Reliability by Eric Camden, FORESITE INC.