The future of Industry 4.0 obviously includes artificial intelligence (AI). But how can companies use it to its fullest potential? The hard truth is, the industry isn’t yet to the point where that’s possible. Even though it’s been around in some form for more than a decade, AI is still young; and just like a child, it has a lot of growing to do.
At Hannover Messe, a panel discussed how AI systems builders can create regional ecosystems of AI fusion learning and reinforcement learning that can enable us to use AI in its best form.
“The manufacturing industry, mechanical engineering, plant engineering, etc.—this is where AI is going to be injected,” said Prof. Dr. Sepp Hochreiter, the head of the institute for machine learning at JKU Linz. “AI systems need to get millions of input data, millions of images, thousands of words that it needs to learn first…but in industry, you can’t wait that long.”
The solution, then, would rely on skipping some of the data collection to make up time that would enable the system to learn new things. “You want to build a basic knowledge with your database, and from this basic foundation, you try to learn and adapt to a new situation,” Hochreiter explained.
This is where control systems can come into play.
“Our control systems already have all the data,” said Dr. Fabian Bause, product manager at Beckhoff Automation. “In the manufacturing industry, data sit on the shop floor for the machine builders. They don’t necessarily have those data. Why? Because machine builders are designing, developing and building a new piece of equipment.”
This means the AI system builder won’t have all the data available, will have to apply the AI routine on top of that and the system will have to learn more data during the commissioning stage.
Hochreiter suggested we take a cue from consumer electronics giants and apply it to manufacturing AI systems. “People at Google or Facebook give us a device—a mobile phone—and they’re mining data from there and they know exactly what we want,” he said. “So what did we do so far? We build superb machines, but we’re selling them and that’s it…You let go of your product and don’t get your data back.”
If several companies teamed up in the early stages of AI integration (companies at the end of the cycle wouldn’t benefit from a data-based learning partnership), they could share data and help AI systems learn faster. The problem is that no one wants to share all their data.
“The idea would be telling the AI system how it can learn, but I keep my data—the data could be anywhere,” said Hochreiter. “You only have a centralized AI system without disclosing your data, then the AI system draws on different vendors, different manufacturers and learns by an underlying process.”
Regardless, data privacy is a concern for many companies across various industry verticals. Peter Seeberg, owner of Asimovero.AI, posed a scenario in which data could be shared and private. A client would have the option to use a centralized solution that distills knowledge from data where the data remained with the client’s machine builders.
He ended the scenario with a hypothetical question to the large vendors like Bosch and Siemens: Can’t we all just team up?