Gen I™: A new generation of inspection freedom

The market need for the new solution, according to Product Manager Alexia Vey, has long been apparent: “Producers with prototyping and NPI lines simply don’t have time for AOI programming. Delivery time is critical in this business, and many rely solely on visual inspection. We wanted to give them full-coverage inline metrology for these jobs in minutes – not hours or days.”

We sat down with Romain Roux, R&D Director for AI, and Nicolas Guillot, Software Systems Architect, to look behind the scenes of the development of a game-changing technology for AOI programming.

In a few words, what is GenI, and what value does it deliver?

Romain Roux: GenI is an automated solution that drastically reduces the time and specialized skills required for AOI programming. Traditional AOI programming can take weeks of training and hours a day of programming for those who have, say, ten batches per day. GenI cuts that down to one or two hours of training, and just minutes of eparation per product. This means operators can channel their expertise into meaningful SMT skills rather than specialized AOI programming.

Nicolas Guillot: Exactly — it’s the fastest, easiest AOI programming solution on the market. AOI programming has often been one of the toughest line steps, usually just behind the pick-and-place programming. With GenI, it’s a matter of a few clicks and a few minutes, or just under 10 minutes for bigger products. The system only asks operators what they know best: Which features do you want to recognize? And is this component good or not? That’s it!

What inspired the development of GenI?

Romain Roux: Years ago, when we began working on MYWizard, we had a concept very early on. We knew it would be a huge advantage to have a system that extended beyond recognizing fixed packages to cover more general characteristics. Because, you know, a lead is a lead, and a more generalized system would be able to capture this. Then, during the launch of DeepReview, we learned a lot more about where customers faced the most time-consuming tasks, so we had a solid idea of what we would tackle next.

Nicolas Guillot: I would say there was a convergence of factors that inspired us: the rise of AI enabling fully automated solutions, our focus on high-mix manufacturing, and a major evolution of our software development capabilities. It’s part of a longer arc at Mycronic, where we’ve transitioned from hardware-focused to software-focused, and now AI-driven development. In the simplest sense, this gives us access to far more high-quality data. We knew that older image comparison methods — most often based on 2D imaging – were simply too low quality for our customers’ expectations. So, we challenged ourselves take this to the next level, to build a generative inspection system using 3D metrology.

What scale of data was used to train the AI models?

Romain Roux: We trained GenI with roughly three million images from real production data. The samples represent a wide variety of production types, serving a range of industries, to ensure a diversified, representative dataset. To make the neural network more robust and more sensitive, we also generated ten times more synthetic data. Of course,

Nicolas Guillot: When a customer installs GenI, all inferences are done on the AOI machine itself – so everything remains secure on the customer’s own sub-system. Many Mycronic customers work in high-security sectors like defense, so we knew from the start that GenI would need to operate locally on the customer’s site, with the highest confidentiality and data security constraints.

How does GenI differ from other AI-based AOI solutions using golden board comparison?

Romain Roux: Golden board-based inspection assumes a perfect board, which is unrealistic, especially during prototyping or varied production. GenI performs metrology — it measures component dimensions and orientations, comparing that data to CAD data from the placement machine. Instead of just image matching, it performs geometric descriptions, checking leads, joints, and positional tolerances. SMT knowledge – not AOI programming skills – is what guides the inspection criteria.

What were the main challenges during development?

Romain Roux: The heart of the challenge was scaling data preparation. Training the neural network meant handling and labeling millions of images. We had to solve annotation issues and efficiently manage datasets of three million real production images, plus synthetic data. Data transfer alone could mean potentially weeks of delay. So, it was crucial to plan everything with this sheer scale of data in mind.

Nicolas Guillot: Coordination was also critical — everyone on the team was working simultaneously and had to stay perfectly aligned. Development at this level of complexity means we had to optimize every second of our software development process.

The pace of development for GenI was clearly quite intense. If you had more time, would you have done anything differently?

Romain Roux: Looking back, of course there were dead ends. But working closely for over a decade as a team helped us focus on delivering real customer value instead of just code or features. A working mockup came together quickly and was taken to customers within a few months for feedback. This would not have been possible if we had the luxury of too much time, which can often distance you from the actual day-to-day needs of customers.

Nicolas Guillot: I think Darwin said it best: It’s not the strongest nor the fittest who survive, but the most adaptable. Our ambitious targets fostered creativity and engagement with sales and other stakeholders early on, allowing us to quickly co-develop a solution that could be trained and refined in actual production environments. Then it’s a matter of rapid iterations run through the gauntlet of real production scenarios – and this is when you find out how adaptable the system really is.

Who is the ideal customer for GenI?

Nicolas Guillot: In the current version, it’s a Mycronic pick-and-place customer. There are those without an AOI who rely only on visual inspection, and those with an AOI whose programming burden is too heavy – especially those with many NPIs or no dedicated AOI programmer.

What have you learned from early customer testing?

Romain Roux: It’s like designing a chair right in the customer’s living room – you’re building something that’s fit for purpose in real time. We’re AI software engineers, but we were there on the factory floor, loading the customer’s boards onto their machines. This means we really live and breathe their pain points, and it results in simple, pragmatic, and easy-to-use solutions.

Nicolas Guillot: We spent several intensive months on site learning customers’ workflows and showing real-time prototypes. By the end of it, more than half our R&D team had worked onsite with the customer. Even though these customers were already happy with their existing AOI solutions, found both the project and the platform exciting and energizing. Everyone took pride in the process, and in the platform we were building together. The system runs on hardware compatible with Mycronic’s MYWizard software, which means we designed the original platform in a way that continues to prove its value and relevance. It’s really satisfying to see this, and to share in the customers’ enthusiasm firsthand.

What’s next for Mycronic AI in PCB assembly?

Nicolas Guillot: The AI journey continues. We build for the future, developing products that will be in use months or years from now. Based on where we are today, and the fact that even more of our R&D is co-developed from the start with customers, I’m convinced that our solutions will continue to be more relevant and impactful.

Romain Roux: One lasting change is in the way we work. The pace and scale of an AI-driven R&D program, together with this intensive co-development with customers, means that we can see and test our systems much faster and with better data than ever before. It’s more engaging for us and gives deeper meaning to our work. Because it’s not just a series of technical tasks. It’s not just programming behind a screen. It’s something we build and shape together where it matters: on the factory floor.

Related Content You Might Like

GenI Generative AOI Programming debuts: Mycronic sets a new standard in AOI automation - press release