Every company that believes in "smart manufacturing" has great hope for artificial intelligence.
This unknown new world requires a large investment in high-cost artificial intelligence systems, as well as the need to build “learning” platforms and cloud service providers. This ambitious plan begins with big data collection so that the machine can learn and find out what is unknown.
However, this is only theory.
In the real world, many companies find AI difficult to implement. Some have accused them of lacking internal data to prove that humans can make full use of artificial intelligence. Others complain that they can't even build the concept of an AI system. Manufacturers are beginning to realize that artificial intelligence is not a deal that “you make it, they will appear”.
So, how does Renesas Electronics do it?
The Japanese chip company is a leader in the global factory automation market. It is proposing "real-time continuous AI" for operation and maintenance technology (hereinafter referred to as OT). This approach is in stark contrast to "statistical artificial intelligence," which is often driven by big data companies to facilitate automation in the information technology (IT) space.
Yoshikazu Yokota, executive vice president and general manager of Renesas Industrial Solutions, pointed out that embedded AI is critical for fault detection and predictive maintenance in OT. When an exception occurs in any particular system or process, the embedded AI can "make decisions locally in real time," he explains. Renesas proposed the idea of “endpoint AI” three years ago and began experimenting at his own Naka semiconductor factory.
"Our plan is to implement real-time reasoning in OT while gradually increasing the AI capabilities of the endpoints," Yokota said.
Renesas Executive Vice President Yokota Kenichi
By bringing artificial intelligence to the factory floor, Renesas hopes to help customers who are currently working on the AI concept and to make a return on their investment in AI.
When to apply artificial intelligence to OT
Mitsuo Baba, senior director of strategy and planning at Renesas Industrial Solutions, said that when specific issues (such as production lines) have been identified, it is the best time for AI to be applied to OT.
For example, suppose AI is an experienced operations manager whose experience can be used to discover certain anomalies in the plant, rather than having the manager check each stage of the manufacturing process one by one. "We can use artificial intelligence to draw Line to detect the time and place of anomalies during production defects," Sanfu said. AI can continuously monitor the production line to prevent defective products from entering the next stage of production.
In such a factory automation example, the AI needs to be trained only on a predetermined basis. AI reasoning runs in real time on endpoint devices without going back to the cloud. Sanfu said that 30K bytes of data is usually sufficient for endpoint reasoning compared to statistical AI learning and reasoning, which usually requires processing up to 300 megabytes of data in the cloud.
In short, Renesas is advocating AI reasoning that can be done on the MCU.
Renesas' "AI Unit Solutions" kit can be connected to existing production equipment instead of replacing existing production lines with new AI machines, which are costly.
Sanfu said that Renesas has no plans to challenge AI chip companies like Nvidia. “Our goal is to lead the new market segment of embedded AI, where the data required for reasoning is very small and can even be run on existing MCUs/MPUs,” said Sanfu.
