ON HARNESSING THE POTENTIAL OF MACHINE LEARNING…
Instead of producing a part and then fine-tuning it through trial and error testing—wind tunnel, track testing—we can use virtual data, put the drivers into the simulator and let them run. Then, if we are able to extract performance, meaning get the car to go faster, we go back and starting producing. We tell the engineers, “Look, that’s the physical property of the part you need to design.” That way, we are just physically producing one part, the part we are actually putting [into] the car. We save time, we save money, we iterate faster—and all of that brings performance.”
Ted Doheny, CEO of mining equipment company Joy Global: “In mining, the challenges today are immense. In the last four years, the prices of commodities across the board have been cut in half, so our customers are under a lot of pressure. We use sensors and the IoT to solve mining challenges.
Our longwall shearing system, one of the most automated pieces of mining equipment, is a good example. Where a mine once had more than 35 people underground, we now have 7,000 sensors and 140 cameras that constantly monitor and control processes, transmitting data between machines and up to the surface. If that longwall system goes down, it’s $1,500 a minute. So we help our customers use this data to predict a problem and get the inventory in place to make a repair before it goes down. As we drive automation, we look to simplify the process, eliminate waste and remove people from harm’s way. For one of our customers, automation was able to eliminate $100 million of waste and reduce 35 people working underground down to less than five.”
ON WHAT THEY LEARNED ALONG THE WAY…
Meyer: “At the end of the day, it’s about making the right decisions—in our case, do we pit the car or not? But sensors can have the wrong readings. Whenever we have an alert, we crosscheck it with two or three other sensors and put the physics behind it to see if what the sensor is telling us is even possible. Otherwise, we could retire a car just because we have wrong information coming from the sensor.”
Teske: “You do need to decide what to do with all this data. We went back to our framework and how we talk about innovation, because innovation is an overused word that means something different to everybody. We defined it as user-driven problem-solving and that’s where we started. We looked at how we accumulate data and communicate it so that it becomes information that ultimately solves a problem, whatever that problem is, for the user.”