That lack of consistency is becoming a major obstacle as manufacturers attempt to deploy AI and advanced analytics across operations.  

“If I just say the number 1086, what does that mean to you?” Semle asks. “Absolutely nothing.” ”

He points out that a lot of data is a “whole lot of nothing” without context, and raw industrial data has limited value. Manufacturers may collect enormous amounts of information from equipment, sensors and production systems, but AI systems still struggle to interpret it correctly.  

According to Semle, the problem is rooted in how manufacturing systems evolved over time. Many facilities were designed to solve immediate operational needs, not to share standardized information across the enterprise.

“When a controls engineer was in a factory trying to set up a CNC machine, they weren’t thinking, ‘Hey, this machine needs to be globally recognizable across our entire enterprise,’” he says. “What they were thinking was, ‘I just need to get this job done.’” 

As a result, manufacturers often rely on disconnected point-to-point integrations built around individual use cases rather than broader data strategies. 

“We have a lot of point-to-point connections. I would argue [that] is not a data strategy,” Semle says. “It’s just trying to solve the next fire.” 

The issue is compounded by proprietary systems and machine-level data restrictions that make information difficult to access once equipment is deployed. 

“You might order a CNC machine, and then you’d like to poke into that machine and look at the XYZ axis,” Semle says. “Depending on where that machine came from, that might not be that easy to do.”

That fragmentation creates operational challenges throughout the factory. 

One example involves autonomous guided vehicles operating inside warehouses and manufacturing facilities.

“AGVs are just like Roombas,” Semle says. “They get stuck.” 

The problem is not simply the vehicle itself, but the fact that different systems expose operational data in completely different ways.

“One exposes it over Modbus. One kind of does it over OPC UA, another one has a SQL database,” he says. “This is the problem.” 

To address those issues, manufacturers are increasingly focusing on contextualizing and standardizing industrial data before attempting broader AI deployment.

“The industrial data ops layer is the layer to say, what is that number? What does it mean?” Semle says. 

That standardization effort is part of the goal behind the Industrial Information Interoperability eXchange, or i3X, an emerging CESMII standard designed to create a more open and consistent way to exchange industrial data across systems.

“Think of it as simply a standard open way to define that interface of data in and out of the factory,” Semle says. 

Semle believes those efforts will become increasingly important as manufacturers push AI closer to real operational decision-making on the plant floor. 

“If we can contextualize and make the data in the factory available to AI, I, as an operator, can just ask questions,” he says.

That capability could become especially important as experienced workers retire and manufacturers attempt to preserve tribal knowledge that often exists only inside individual facilities. 

“The very minimum thing you need to do is just get people to start to write that down,” Semle says.

Ultimately, Semle says the future of industrial AI depends less on algorithms and more on whether manufacturers can finally organize the information they already have. 

“AI is the driving force to say, industry-wide, we need to move towards standardization,” he says. “I think it’s going to happen whether folks like it or not.”