“If we take a bunch of data from a bunch of sensors and dump it all in a data lake, nobody knows what it means,” Sobel says. “The labels on the data, the way data's been named on shop floors, data's been added year after year by different OT integrators ... nobody knows what the data even is, never mind how it all fits together.” 

Beyond inconsistent labeling, manufacturers must also reconcile data collected at different speeds and times. Machine sensor data may be available in real time, while quality results, inspection data or defect reports may not arrive for hours, days or even months. 

“You actually want to be able to understand, am I making good stuff or bad stuff?” Sobel says. “But you don't get your quality data until maybe an hour, or a day, or a week later ... integrating all of that across time is an extremely challenging problem.”  

Sobel says manufacturers must first organize, map and contextualize production data before AI can produce reliable insights. Once those challenges are addressed, AI can identify relationships that would otherwise remain hidden. 

“We think of data and AI as being another next-generation tool in the hands of experts that lets you see deeper and farther than you could with your own senses,” Sobel says.