A New Generation of AI AOI

Estimated reading time: 4 minutes

Roberto Gatti is CEO of Delvitech, a Swiss company that has launched a new generation of AOI utilizing an AI neural network. As Roberto explains, their approach was not to bring another AOI machine to the market; instead, they focused on bringing a new technology to an AOI system.

Barry Matties: As you enter the crowded AOI market space with a new system, I am curious what inspired you to do this.

Roberto Gatti: To start, we did a complete study and assessment of the AOI market around the world, and we found areas that were not addressed by our competitors. Basically, we had two choices. We could enter the market with a machine and strategy like what was already out there and try to compete on price, or we could do something completely different. We chose to be different, with a new generation of AOI driven by AI. With that, we invested 30–40 million euro and developed a totally different system.

We are offering a single system with two different concepts. One has an optical head that is fixed and the board moving. The other is with an optical head that is moving, and the board is fixed. We developed a proprietary optical head for both machines that can be used for all processes: plated through-hole, SMT, SPI, and mechanical objects, as it can be used to identify whatever is available on a board.

Matties: Is this optical head something that you developed in-house?

Gatti: Yes, totally in-house, together with some universities in Switzerland. It's a project that took almost three years. It has several patents. Our system is the only one on the market with six cameras, all fully integrated with FPGA and Thunderbolt interface. The telecentric lens is another customized system with two arms. On one arm you get the polarized camera; on the other, you get a 12- or 25-megapixel top camera. The reason for such a complex optical head is the fact that we need to acquire a lot of information from the board because we do an inspection that is the opposite of everything our competitors are doing.

Our competitors—extremely good companies—are looking at what is wrong inside the window with a mathematical algorithm or an image comparison. We, on the contrary, check everything that is happening in each window, even what is good and why it is good. We apply AI to everything, but not an algorithm. The machine is AI-based, so whatever you touch in the machine has an impact on different neural networks that are doing different jobs. You can train the machine to inspect new components and new technologies. You take the component and tell the machine to learn it. The machine learns the component, and while it learns, it already understands the best way to inspect it.

This system is web-based and is totally available on the cloud which provides a big advantage, especially for those international customers who need to have different types of systems in their production line—SPI, pre-reflow, and post-reflow. With Delvitech, they don’t need to buy different machines, they can buy one model of machine and place it wherever they want in the line. The completed neural network training happens locally in the line, or if you have a large or a major account, everything happens on the cloud. If you have a plant in Germany, for example, with 20 of these machines on 10 different lines, the neural network is learning a lot of information and bringing detectability to the highest possible level. Then, say you decided to open a plant in India. You simply take the information from the neural network in Germany and apply it to the machine in India, which will instantly behave the same way it did in Germany.

Matties: When customers come to you to learn about your equipment, what are their greatest concerns?

Gatti: That it’s a new technology. If you go back to when Tesla arrived in the market with a new-concept electric car, everybody asked, “Will it work? Can we use it?” It was very difficult for Tesla to enter the market with this new-generation car. Then a lot of others decided to put their electric cars on the market. However, they simply took a very good gasoline or diesel car, removed the engine, put in an electric engine, and said, “Okay, this is an electric car.”

What's happening in the market now is something similar. We have not created a machine, but a concept, a new technology that is totally AI-based and matched with a very particular camera head to simplify the work of our customers. Some of our competitors understood that we were creating a different generation of AOI, so in some cases they have decided to bring AI inside one or two algorithms, data analysis, or data management, but it's not like having a real AI-based machine. It's not enough to do software. You also need to structure the computer hardware of the machine to support this huge amount of data flying in and out, providing a real-time result. We need to simplify the concept of the systems, and the concept of this technology, to let customers understand what's going on without using the difficult terms of AI—because AI is complex.

Matties: Bringing a new product to market can be slow and painful because people go back to what they know. What will motivate them to take that step?

Gatti: There are different reasons to take a new generation system like this. The first one is the return on investment...

Continue reading this article in the January 2023 issue of SMT007 Magazine.