AI and the Future of Maintenance (Video)
Sensors can extend human reach in terms of distance, simultaneity, types, and volume of data but challenges still abound because real world scenarios affect machines differently and the ability of machine learning and sensors to detect and understand different environments may not be as acute as that of a worker with decades-long experience.
“I did some work with an industrial company recently and they are in the testing space. And one of the things that they want to do was narrow the window for maintenance, right? So, you have these big industrial machines and you have these factors of production, quality, labor, and inventory. And if you take a machine down for too long, or at the wrong time, it interrupts the production line and you don't get as much stuff out the back,” Rasmus explained.
Alternatively, if the machine is interrupted at the wrong time time and a new variable is introduced, and the quality may be decreased. “You also have to think about who can repair the machine and when are they available? There's all this stuff that has to happen. And of course, the answer to all of that is that we're going to use AI to figure out that complex coordination of all of these resources, right?
And typically today, most companies, and they still do this, have a guy with a clipboard and sometimes maybe a manual tester walk out to a machine and see how it’s doing, right? They have these things called ‘walking paths’ and they literally ... And this is knowledge transfer from one generation to the next. They go, ‘We're going to walk around to the machine. Okay. Let's see. How are we doing over here?’ We take some notes and get some feedback. And we go, ‘Oh, there's feedback in that one.’
And then we take those paper notes and we go upload them some place. And maybe, if you're really sophisticated, you use your iPad, or a mobile device, and you upload it in real time. And maybe you do something where you analyze that information and you say, ‘Well, there's something that we got to go fix on that machine.’”
What people are finding is that machines live in the real world, said Rasmus. “We've kind of always known that and the AI people and the IoT people are going, "Let's just outfit them all with sensors," right? ‘Let's put all these sensors on this machine.’ Sound, vibration, stress, pressure, level sensors, is it still level on the ground? Smoke and gas sensors in the environment, right? Because you can have all this stuff that happens. And what they're finding is that, if I take a machine and I go, it should be serviced every 6 months, and I stick it in another environment with higher humidity, or different heat signatures, maybe I have to do it every 3 months.”
Beyond the real-world technical issues around the transition to AI and sensors is the people problem, he pointed out. Industrial test maintenance personnel can often just walk up to a machine and identify a problem, such as that it is vibrating too much. “And they can sense that because they've been doing that for 20 years. And now we can put these sensors on that. So, there's also this whole transition issue of the, are we going to have displaced workers, because the guy who senses it can now be replaced by a sensor, right?”
Many speakers have made their presentations available at www.kmworld.com/Conference/2018/Presentations.aspx.
Learn more at KMWorld 2019, coming to Washington, DC, Nov. 5-7.
Watch the complete video of this presentation, Rethinking KM for an Age of AI & IoT, in the KMWorld Conference 2018 Video Portal.