Benefits of AI and machine learning for automation safety systems
The thought of using artificial intelligence (AI) and machine learning for safety seems like an oxymoron. Presenter Tina Hull, product engineer for Omron Automation, joked, “I’m an expert in safety, and no one I know thinks I’m sane for using AI for safety applications,” at the start of her presentation, “The Use of Machine Learning in Automation Safety Systems,” at Automate 2019 on April 9 at McCormick Place in Chicago. (Automate is organized by the Association for Advancing Automation, A3.)
Using AI for safety is no more absurd, perhaps, than people landing on the moon, which Hull used as a parallel to start the presentation. What might have seemed fanciful prior to space exploration wasn’t to people like Robert Goddard. The same was true of John McCartney, the father of AI. In 1956, he wanted to teach machines how to think. And this was back in the days when computers or their precursors were gigantic machines that took up an entire room to perform very primitive tasks.
Fast-forward to 2019, and we aren’t quite there yet, but it’s getting closer. Hull said, “We have a lot of information. And with that info, there is so much we need to. When it comes to AI, we need to organize data for safety applications to determine good and bad. Using this information helps us better understand our systems.”
And with machine learning, engineers have a little bit of an advantage because collaborative robots have become common on the manufacturing floor. There is a lot of safety and precaution built in through different standards and learning from experience.
“Everything applied today in safety is based on things we’ve learned from the past,” Hull said. “We have to understand based on the machines we’ve learned from and apply that.”
Teaching robots how to learn
Hull used an analogy involving learning how to sail and knowing what kind of boat she and her husband wanted to learn on. After a lot of reading and YouTube videos she came to a surprising conclusion based on her initial learning experiences.
“What we’re learning is we know nothing, but we’re making connections,” she said. “We’re using bits and pieces from past experiences and applying them to new experiences. We learn what works, what doesn’t, and use them to make future decisions.”
The same, Hull said, is true for a robot. Robots are designed to work for the 95th percentile so they can function properly with a person regardless of height. The robot is a taught what it’s like to work with a person and vice versa. Humans have the advantage of adapting whereas the robot, in its initial stages, is fairly dumb by comparison because it doesn’t have that human element.
“We want to get to the point where the robot can adapt to the person it’s working with,” Hull said.
Hull cited a few more analogies or potential capabilities through machine learning. Can the robot learn to respond to someone whether they’re right- or left-handed? Are ergonomics, which are all the rage from a safety standpoint, really needed if the robot is taught the basics about the person they’re working with?
What about emotions? By definition, robots have no emotion, but they might be able to adjust their programming through technology such as facial recognition. Facial recognition is already used to turn on computers and unlock smartphones. Hull took the idea a step farther and suggested a day where the robot can sense if a human is having a bad day through facial recognition and adapt its programming accordingly. Emotional sense? Robots have no emotion. But facial recognition in robots is not far off.
Processing information to predict, prevent incidents
Hull said there is so much information out there, and it’s impossible for workers, especially younger ones, to learn and retain it all. However, companies can close that gap by teaching the machines and computer systems how to read and synthesize all that information for us.
An AI, through machine learning, also can interpret data and tell the worker what is needed to build a safety system.
“Machine learning gives you the risk reduction tools,” Hull said. “It’ll give you short- and long-term data of how things change.”
It also can predict when something is about to break or is in danger of breaking. Like a car’s brakes, Hull said mechanical parts in a robot tend to wear down. A robot’s stop time, she said, is slower as the robot gets older. Safety standards exist to give workers an idea of when something might break, but there’s still risk involved.
What if, Hull asked, if an AI could predict when something happened by learning how a system operates and prevent it from happening. “The technology is available to make these calculations and help troubleshoot if there’s a problem,” she said.
Machine learning and AI in a new world
Manufacturing has changed from a static enterprise to a mobile one for many companies. A robot used to be designed to perform the same task for 20 years. Now, Hull said, “We’re gonna be lucky if robots do the same thing for two years. Consumer changes are changing product demand, and it’s become more customized, and they have to adapt to those changes.”
Robots with abilities to adapt and learn from experiences, once the stuff of science fiction, are getting a lot closer to science fact. With the rest of the Industrial Internet of Things (IIoT) and Big Data, every last bit of information is going to be scraped and analyzed. It won’t be long, Hull said, before the machines start thinking with us because they’re already pretty close to doing it.
Chris Vavra, production editor, Control Engineering, CFE Media, firstname.lastname@example.org.
For more from Hull, see articles highlighted and also see related content from Automate below.