Big data is just like salt water. There’s lots of it everywhere, but you can’t consume it without putting in plenty of time, money and labor to clean it. This is why many end users lament that their analytics projects are drowning in data, but starving for useful insights and actionable intelligence, so they can improve decisions and add value to their products and companies.
“Some company managers have the impression that their operations people have the time to look at and analyze lots of data, and they really don’t. One client told us he had 10 pages of data with 50 columns each from one casting machine, and he was expected to analyze it, but he was too busy keeping his processes running, dealing with downtime, and going to meetings,” says Sam Hoff, president and CEO at Patti Engineering in Auburn Hills, Mich., a certified member of the Control System Integrators Association. “This is where system integrators with process knowledge can help, so we do analytics projects that combine existing data with tool like Sorba.ai, Seeq and Power BI.
“For example, 13-15 years ago, we built a feeding machine for Ancor that stuffs eight to 10 National Highway Traffic Safety Administration (NHTSA) documents and labels into the envelopes along with the Monroney window stickers on new vehicles. We upgraded the machine’s controls a couple of years ago, and now collect data from it. We have regularly scheduled production reviews with Ancor, where we analyze their data, discuss ways to improve their OEE and provide suggestions. I see these types of engagements becoming more the norm in the future as manufacturers try to improve OEE using the expertise of an integrator that can contextualize their data.”
Location and testing
Hoff reports that one of the main questions any data analytics project must answer is where to do it—onsite, on the edge, remotely, in the cloud or elsewhere?
“Our rule of thumb is operational activities including analytics that need to be done that same day or shift should probably be done as locally as possible. This includes tasks like alarms, machine health and overall equipment effectiveness (OEE). However, more historical tasks like data acquisition (DAQ) and archiving for longer-term analysis can likely be done remotely or in the cloud. In the cloud, you want your data stored in an open format, so you can use analytic tools,” says Hoff. “So, for Ancor’s feeding machine, we used Power BI to analyze data captured by Ignition web-based SCADA software from Inductive Automation.”
Hoff adds its latest envelope-stuffing machine can insert 2,200-2,400 labels per hour, which are added to the automaker’s final assembly process. “Faster-responding machines like this can contribute to increased throughput,” he says. “This can help an operating shift get down to eight hours from 10 hours.”
Race to higher resolution
While it’s already important to verify data is correct to avoid a garbage-in/garbage-out situation before it’s analyzed, Hoff reports that testing results becomes even more crucial for subsequent, high-resolution simulations and digital twins.
“I ask people if they know the difference between a simulation and a digital twin, which can be hard because they’re often used interchangeably, but it’s an important distinction,” says Hoff. “A simulation is just a model of a system such as a picture. A digital twin is a model that integrates real-world data, so it can act the same as its physical counterpart. This lets users see how devices and systems will operate in the real-world, evaluate expected results, and adjust designs before actual components are constructed. In the past, we knew much less about the effects on prototypes, but now we can do trial-and-error tests without real-world hindrances. The process of making digital twins is also iterative, so new data can be added and tested in multiple cycles before real-world items are built.”
In another case of industry following consumer technology, Hoff adds that today’s digital twins for manufacturing should likely seek to become as finely rendered as the iRacing digital, online game, which simulates real cars running in real NASCAR and IMSA races. In fact, its simulations are so good that professional drivers reportedly spend hours on iRacing to prepare for upcoming races.
“We keep thinking wouldn’t it be nice to have such high-resolution digital twins for industrial production,” he says. “All the tools are available. We just need to feed more and more real-world data into our models, and make sure the simulations produce outputs that are the same as those in the physical world. Then, we test their results in successive iterative cycles, and add each update to the digital twin.”
