AI and ML generating real value in manufacturing

By Jim Montague

May 31, 2018

TechED 2018 LeadImage 2

People have borrowed ideas and picked one another's brains as they've collaborated throughout human history. So it should come as no surprise when they do the same with today's emerging forms of artificial intelligence (AI) and machine learning (ML) know-how.

"For those of us on the operations technology (OT) side, these technologies can be a mystery,” admitted John Dyke, global director of software business development at Rockwell Automation. “But they’re no longer futuristic. We have vital customer pilots and projects going on right now. Productization is still a little ways out, but the ability of AI and ML to generate value is real."

Dyke and David Crook, senior engineer at Microsoft, presented "The Role of AI in Manufacturing" on the opening day of Rockwell Automation TechED in San Diego. It covered how AI, ML, computing, data and algorithms are changing the face of manufacturing.

Extending AI reach, functions

"In all the discussion about AI, there's not much about how to do it,” Crook said. “But potential users need to understand its deeper levels, so they can decide where and how to invest in it. As our CEO states, 'Microsoft's goal is to democratize AI to empower every person and organization to achieve more.' Previously, we had to hand-code many tasks, but now many can be done with automation."

Crook added that Microsoft is also seeking to:

  • Extend AI to the edge and the device level of facilities and applications
  • Perform closed-loop AI at the edge to fix devices and prevent problems 
  • Make the "leap" to embedded and remote components

The venue for much of this development and activity is the Microsoft Azure cloud-computing platform and service. Crook reported that it serves as an across-the-board programming environment that lets users develop, update and adjust software products. Azure presently operates worldwide in 42 primary regions, but its Azure Stack version enables hundreds of service providers and thousands of individual enterprises to run it as a standalone instance.

Basic concepts

Crook explained that one of the most important aspects of AI is to learn how it's different than ML, including understand that ML is really a subset of AI.

"AI is based on perception of intelligence; encompasses a variety of techniques including machine learning; turns its autonomous perception into action; and generally describes more complex systems," said Crook. "ML is more statistically based; uses algorithms with weighted values that are 'learned;' and typically applies to simpler systems."
The main categories of ML are:

  • Supervised, requiring users to provide labeled data, and delivering relationships and context
  • Unsupervised, which means it can figure out relationships on its own, but can't add context
  • Reinforcement, including an actor-driven model with a “reinforcement” loop

All three types of ML can be characterized by “deep learning,” which is where most technology advances have occurred in recent years. “Reinforcement is interesting because it's closer to AI, and allows users to take existing algorithms, and update them with more recent feedback," explained Crook.

AI in manufacturing

Crook reported that the foundations of AI and ML's role in manufacturing will be laid down over the next two or three years, and they'll accelerate what they can do afterwards, whether it's self-driving semi-trucks, self-repairing robots or other applications.

"AI and ML will develop many building-block capabilities, and combining them will make up the factories of the future," explained Crook. "All of these independent pillars have the same base—to collect device-level data, and optimize by predicting conditions and events."

Thanks to AI and ML, Crook added that "surfacing" or emerging intelligence in manufacturing will enable:

  • Optimizing supply chains and production operations, such as using real-time weather data to reroute shipping fleets
  • Transforming services and products, which consists of using production and ambient-conditions data along with AI and ML to assist new "as a service" offerings
  • Engaging customers in new and powerful ways, such as providing self-service options, product and services suggestions, and even collaborating with users on product development

Crook added that the basic framework for implementing an AI transformation occurs at six main levels, including devices, controls, supervisory, supervisory control and data acquisition (SCADA), information and manufacturing execution systems (MES), and the enterprise.

Further, the primary steps for adopting AI include: 

  • Instrumentation and unification of systems to gather data; 
  • Collecting and normalizing data "estates" or environments, so information can be viewed in common ways;
  • Implementing intelligence technologies, so users can decide where to diagnose devices and when; and
  • Using closed-loop AI technologies where applicable. 

"Scalable analytics are the key differentiator because they can cross borders to provide interesting insights, such as which production line is running the best," said Crook. "These offerings include FactoryTalk Analytics running on data centers and FactoryTalk Analytics for Devices to automatically gather information from ControlLogix modules.”