Procter & Gamble's Jeff Kent Discusses AI and Machine Learning

Summary

Jeff Kent, vice president, Smart Platforms Technology & Innovation at Procter & Gamble, discussed experience with the application of AI and machine learning at the ARC Forum 2023. Kent described goals of cutting manufacturing operating expenses and improving operations, including over 5-10% staffing efficiency improvement, 50% maintenance & and repair cost reduction, and 50% reduction of QA costs. In addition to 25 years at P&G as a controls and automation subject matter expert, he has experience in the U.S Air Force working with enterprise systems & networking. 

Kent believes AI and machine learning are fundamental components of production digitalization. Kent has been on the Procter & Gamble (P&G) digitalization journey for about five years in a corporate group that began seven years ago. The group, which has grown to include 25 people, is focused on practical application of Industry 4.0 concepts and has engineering innovation centers in Cincinnati, Ohio, as well as Kronberg, Germany, near Frankfurt. "Things couldn’t be any more exciting," he said.

Kent emphasized, “The control system is a very powerful asset,” and P&G has a very important operational intelligence program with the goal of introducing thousands of machine learning algorithms at the equipment level throughout 120 sites in over 40 countries. “The control should not be forgotten,” he noted, describing adding intelligence at this edge level as an important part of IT/OT convergence.

WISE Initiative

The company has created the internal WISE branding for the initiative including the SmartBox that is an edge control and computing device that P&G is partnering with industry leaders to deliver “in a very practical and affordable way.” Procter & Gamble is committed to introducing AI and machine learning models across all operations with a particular focus at the equipment level with collaboration across key work systems whether it be quality assurance maintenance, and material utilization. Wise is the overarching internal P&G service that supports the DevOps of the Smart Box device as well as the overarching Machine Learning (ML) lifecycle.

SmartBox Edge Device

P & G is using an edge device they term the "SmartBox" that interfaces and harvests data from existing controls, new controls and OEM equipment including interfaces to Mitsubishi, Rockwell and Siemens. Kent emphasized that computing at the control edge is important because many of the algorithms that run machine learning in P&G’s core work processes must be tightly coupled to the control system in real time to accomplish functions including adaptive control. 

Information is communicated to the OT stack above and below the firewall using OPC UA, and this also supports cloud applications collaborating with Microsoft. The information becomes available throughout the organization accelerating digitalization improving efficiency, quality and profits. “We map the whole data cycle from getting the data to contextualizing the data to working through model development through model deployment and then all the way to delivering information to operators," Kent explained. 

Manufacturing Machine Learning (ML) Lifecycle

The P&G architecture is designed to support the Manufacturing Machine Learning (ML) Lifecycle Kent describes as:

• Capture data
• Historize and contextualize data
• Explore data
• Develop machine learning models
• Test and validate machine learning models
• Deployed machine learning models
• Monitor machine learning models
• Maintain machine learning models

OPC UA Foundation

Kent clearly stated, “OPC UA is vital; I see no better way to have a to have a common language between OT and IT than with OPC UA.” P&G developed a strong relationship applying OPC UA throughout the architecture achieving seamless data access that scales reliably and repeatably throughout the organization.

Reaching beyond Purdue Model

Reflecting a view shared by many, Kent pointed out the structural architecture of systems needs to change:
“The Purdue model has served us well for many decades. In the end, we’ve been respecting that for too long. What we really need to do is start to blur some of the levels of the Purdue model so this can be a much more broadly communicative architecture, much more open for everyone to participate.”

“If we don’t introduce more agnostic network of things, I don’t think we can deliver industry 4.0 or the power of intelligence. Initiatives from field to cloud are very important and the OPC Foundation is at the forefront of defining what it should be.” 

Kent further explained, “We are not going to respect the Purdue model. Why should we? Why send information through different layers rather than communicate directly where it needs to go? We want to roll out to all 120 factories within a year. I see no better way to have a to have a common language between OT and IT than with OPC UA."