Dr. Robert J. Shillman - Cognex Corp.
Management
Sure, Rick. Thanks for the question. And it's an important one. Up until now, the vast majority of our bookings and revenue came from companies who had problems that we call the something somewhere problem. In other words, well, let me back up. And machine vision can be broken into – every machine vision problem can be broken into two types of problems. They're both difficult. One is the something somewhere problem. And the other is the anything anywhere problem. Our expertise from the founding of the company was in this something somewhere problem. And an example of that is when a razor blade is made, the customer, the manufacturer wants to know if the edge of the blade is in the correct position. Our systems are trained by our engineers to know what an edge – what the cutting edge looks like. And we train the system. We write software when the customer configures a system, to tell it where approximately that cutting edge should be. Then the camera captures the image and the software goes in, finds the edge, and measures the angle of the edge and the position of the edge very accurately and determines if it's within spec or not. So that's the something somewhere. We know what it is we are supposed to be looking at. It's the cutting blade. And we know approximately where it's supposed to be, something somewhere. Most machine vision problems, until this time, on discrete items, items that you can pick up, are the something somewhere problem in manufacturing. Are the buttons on the cellphone in the right place? The system is trained to know what the buttons look like and where the button's supposed to go. But very recently – most recently, in variety of factory automation applications, in particular consumer electronics, the manufacturers are very much concerned with the aesthetics of the product. Not just are all the parts in the right place and is the assembly correct and the screen correct and the screen brightness right, but are there flaws or defects on the case, either inside the case before the items are being put into the case, because that's very important if there are flaws in the surfaces of the inside of the aluminum or plastic or whatever is the case. They can make very big changes and the phone won't operate. They could perhaps damage the battery if there are bumps some place. So, they've come to us. Our customers have come to us, many customers, and asked for the ability to detect anything anywhere. Now anything anywhere means a scratch or a flaw. If you look in the dictionary of the definition of flaw, it can be anything. It's something that isn't supposed to be there. In our cases, the manufacturing, it's typically a surface defect on a discrete item, such as the cellphone case or any kind of consumer item that – it could be a watch face, and it could be a scratch or a defect, extra paint or something that's not supposed to be there. And until now, we didn't have very good technology or tools for defining those defects to tell the computer what to look for. So, fortunately, we found these very, very smart guys in Europe called ViDi and there are many applications of deep learning and AI. But most of them or the ones that we have, the most of the ones we've seen, have nothing to do with really machine vision. They have to do with recognizing faces and photographs. Is this really my dog, not your dog? Is this a cat? And those are very interesting, but they don't apply to factories. ViDi, on the other hand, we found specialized in the application of AI, deep learning, self-learning systems, specifically to the kinds of things our customers want. So in the way this operates is, the customer, it shows various examples of what they consider defects, what they consider defects. And the system then trains on these and by itself through this very, very intensive software, determines what it means. Is it a scratch of this length? Is it a whip? All sorts of things that our engineers used to pull their hair out to actually write code for and now they don't have to. The code exists in this deep learning system, self-learning system, to automatically look at this range of defects that the customer's shown and by itself come up with the descriptions for what it is. And then it uses those – that's called the training phase and then it uses those in the run mode of the system. So, once it's been trained by the factory engineers – not by Cognex. We're not going to be doing that. We ship standard products, as you know. Once the system has been trained by the manufacturer on what those defects are, the system then is set to run mode and operates and we believe in our tests far better than any human could do and far better than our own engineers could previously do writing their own software. So, it's – this is truly a breakthrough for us because it adds this kind of capability. And it opens new opportunities for us where we previously couldn't go. So now, we have a full suite of tools, I shouldn't say full suite, we're always expanding them. We have tools that cover the something somewhere problem which we've been selling since 1981. And now, we will be able to offer the anything anywhere for discrete manufacturing items around the world. Does that answer your question? It's a long answer, but it's an educational one. I used to be a professor, so sorry for the long-winded response. Did you get it?