How can AI help make a production process more efficient—preferably without compromising reliability and predictability? Jonathan Berte offers answers.
As the founder of Robovision, Jonathan Berte is on the front lines of AI integration every day. Together with his team of 130 employees, he collaborates with some of the biggest players in the world to implement large data models in industry. That’s easier said than done, with many bottlenecks to navigate. At Advanced Engineering, Jonathan will lift the veil a bit—here’s a preview of what you can expect from his keynote.
Three Tips for Every Production Process
1. How do you deal with a lack of data?
“When it comes to data processing, it’s important to start by balancing the data being used. Otherwise, how can you detect anomalies? In a production process, everything starts with a well-thought-out setup of sensors and the corresponding vision engineering. You need to think about which type of sensor to use—such as 3D or hyperspectral—and what combination of data you want to gather to map out specific aspects. Always keep in mind: if you feed your data model junk, your output will be just as bad.
“However, in a production process, it’s not always easy to collect enough data to feed a model. Manufacturers have optimized their lines to be cost-efficient—tampering with that is a nightmare scenario. Don’t shoot the winning horse. That’s why it’s challenging to generate data that actually brings insights—while deep learning models are notoriously data-hungry.
“Fortunately, this doesn’t mean it’s impossible for smaller players to generate enough data. Most problems can be solved using generic anomaly detection. You store data from things that go wrong in a folder, and a human reviews it for verification.”
2. What is the applicability of LLMs in an industrial context?
“A Large Language Model (LLM) is, for many people today, the embodiment of ‘AI.’ The ChatGPTs of this world want us to believe that only the latest and greatest tools will help you move forward in your work. That may be true in a consumer-focused environment, but in industry, we look at least eight years ahead. That’s why we work with ‘smart tools’ that can still be maintained and upgraded eight years from now.
“These are two different worlds, two ecosystems, that we at Robovision are trying to connect. Tooling that lasts for years and remains relevant, and how we can link that to deep learning models. In this way, we’re bringing LLMs into the industrial world in pragmatic ways that can robustly support operators.”
3. Do you really need experts for this?
“Because those operators are the key to flexible production. Today, AI consultants are often brought in to deliver a project. If something goes wrong afterward, you have to call them again to fix it—and you end up at the back of the queue. That creates bottlenecks.
“By working with LLMs in an agentic way, you empower the operator to fix problems themselves. By processing data at the edge (via edge computing), with low latency, and giving the person next to the production line the necessary knowledge and tools, you make your process much more efficient. So yes, you need experts—but they don’t have to be external consultants.”
Learn more on Wednesday, May 21 at 12 PM during Jonathan Berte’s session: ‘Vision AI: Manufacturing’s Hidden Intelligence.’ Register now!
