Manufacturing and AI: Promises and Pitfalls
Manufacturing Engineering Magazine July 2019
By Bruce Morey
We all know the buzzwords circulating around digital data and the factory. You have heard them—Industry 4.0, smart factories, data analytics, and artificial intelligence (AI). The question we all have is how will this impact workers in the long term? What do these terms really mean? Nevertheless, both traditional software suppliers and makers of advanced manufacturing equipment are offering digital solutions.
Take, for example, the United Grinding Group. In 2017 it started offering a Digital Solutions package as an addition to its high-end, high-precision line of grinders. It makes sense—data collection, handling, and analytics are as important to grinding as they are to any other process. Digital Solutions was a featured topic at the company’s May 2019 Grinding Symposium in Switzerland.
I came away from the Symposium with more clarity. The technology behind those buzzwords will be useful—but only so long as we human beings intertwine ourselves in using it. Because, as a speaker pointed out, artificial intelligence algorithms are anything but “intelligent” outside of the narrow scope of operations they train for. Humans will need to be in the data loop.
How easily confused even sophisticated AI algorithms can get was one of the core messages from Dr. Sebastian Risi, an associate professor at the IT University of Copenhagen, speaking at the Symposium. Still, make no mistake: AI technologies, especially the class of algorithms known as deep learning neural nets, are useful.
Why are they emerging now? “Three trends have converged to make them useful today,” said Risi. “The availability of big data sets, better algorithms than we have had in the past, and the availability of GPU acceleration [to provide enough computing power to run them].”
Machines can now outperform humans in many domains, such as games. “However, these systems still pale in comparison to even simple biological intelligence, which can learn, evolve and adapt to unforeseen experiences,” he said.
The issue is that current machine learning systems can only deal with situations they have trained for in advance. “They are unable to adapt during execution to unexpected situations,” he said. “[This] greatly limits their autonomy.”
Current AI technology is static and specialized to only perform a limited and fixed set of functions. There is a term for it, catastrophic forgetting, where AI algorithms forget what they have learned when learning new tasks. He noted that his group is researching learning and evolving machines that can adapt like biological intelligence.
Until the next wave emerges, we can still get value from AI. That is what companies like United Grinding are doing, setting up data collection via connectivity and using that data to monitor machines remotely, predict future events, and enhance productivity. This is where the blend between data and humans will come into play.
That also explains United Grinding’s approach to product development. United Grinding’s CEO Stephen Nell explained during a talk that until recently United Grinding would develop a new product and announce it to the market. “Now, with our emerging digital solutions, it is different. We need more feedback from our customers in shaping our future,” he said.