The world’s largest glass company, AGC, makes glass for nearly everything, from “skyscrapers to microwave windows, automobiles, and everything in between,” said Bradley Willson, electrical engineer for AGC Glass North America. The company has facilities around the world and has been optimizing operations at its float glass manufacturing operations in Spring Hill, Kan., Richmond, Ky., and Church Hill, Tenn., using OSIsoft’s PI system for production data analyses. (Editor’s note: At press time, Aveva had just announced its pending acquisition of OSIsoft. Read more at http://awgo.to/bSDYc.)
AGC Glass North America has been using OSIsoft’s PI System since 1996 to help make sense of its production data. But while attending a recent PI World event, AGC learned about PI’s move from data historians to a data infrastructure, giving AGC new ideas for how to improve its production operations.
AGC has been using PI products such as ProcessBook, DataLink, and Manual Logger to help visualize data from PI historian servers and create reports. These tools are largely used for process tuning of PID loops to make sure they’re performing optimally. Now the company is “migrating towards OSIsoft’s new product, PI Vision, to build out our asset framework, create notifications, develop complex analyses, and integrate with our ERP system,” Willson said.
PI Vision enables users to analyze data in a number of ways on any device. It can import graphic process monitoring displays created by PI ProcessBook, for example, and allow them to be viewed in a web browser. It also supports mobile browsers and customized views for small-screen devices to enable easy access from anywhere.
Monitoring multiple processes
AGC uses PI tools to monitor its intense production process. “This furnace is capable of melting roughly 600 tons of batch per day. And it’ll hold up to 1,700 tons of molten glass within the tank. It is a natural gas furnace that consumes roughly 120,000 cubic feet of natural gas per hour,” Willson said. “The interior temperature of the furnace reaches roughly 3,000°F, and this furnace has been constantly melting batch 24 hours a day, seven days a week for 15-20 years.”
AGC monitors about 1,500 data points on each of its float lines. It also monitors auxiliary processes throughout its production facilities, such as its sputter coating line, which produces high-efficiency glass through the application of a metallic film added to glass surfaces. Likening the process to how ships pass through the Panama Canal—but lowering air pressure instead of water level—Willson explained how the glass gets transferred from chamber to chamber about every 20 seconds. “Once the glass reaches roughly 10-6 mbar, it gets transferred into the sputter coating compartments,” he said. “Here, we inject argon and oxygen into each compartment and then ignite a large DC plasma field inside the compartment. The plasma grabs atoms of material from the precious materials, bounces around inside the compartment, and then deposits those precious materials onto the top surface of the glass.”
At AGC’s U.S. manufacturing facilities, data from various automation devices are sent to the PI Vision Data Archive server via OPC and a relational database management system, and from there into the PI Asset Framework server. Both of those servers are located on premise at each site. At this point, a centralized, cloud-based PI Vision data server gathers data from each of the Asset Framework servers for further analysis.
The Richmond site stands apart from the other two facilities using these PI technologies because AGC is experimenting with machine learning there. “We’re using the PI OLEDB interface to push data up into the Amazon Web Services (AWS) cloud, where we’re doing some complex analyses and pushing data out through the clients via Tableau (business intelligence and dashboard software),” Willson said.
Implementing new tools
The shift to PI Vision has delivered new tools for AGC’s analytics. Willson’s favorite, he said, are the email notifications. “You set these notifications up once and then PI does the rest of the work,” he said. “It’s completely automated and, most of the time, these emails get sent out even before the operator notifies maintenance that there’s a problem. This helps reduce the amount of downtime and production outage.”
The Greenland facility in Church Hill has begun to add preemptive notifications as well, which can tell them when a process is starting to go out of tolerance and notify the team to take action before it becomes an issue.
PI Vision also enables production reporting, making it easy to compare, for example, what gets loaded in with what gets loaded out of a coater. “[We] can see how many square feet was loaded into the coater per hour, how many square feet of glass was loaded into the coater by shift, and how many square feet of glass was actually packed on the unload end,” Willson said. “This could be expanded to a complete material balance as well.”
PI Vision is also being used for troubleshooting. Willson explained how the software was able to discover a seven-year trend in which the temperature on a melter backwall was slowly increasing. Without the long-term trending capability, the changes would’ve been too small to notice. “Due to this long-term trending [capability], we were able to identify that we had a problem, trace the events back to the start, and then install countermeasures to verify that the problem had been corrected—ultimately reducing downtime,” Willson said. “In this example, we estimated that we would’ve had a two-week downtime if we had not noticed that the backwall was heating up. And that equates to roughly $2 million worth of downtime plus materials and labor.”
AGC plans to continue exploring the additional tools available in PI Vision, with a plan to move progressively toward preventive rather than reactive maintenance. “I’ve seen some really neat use cases for event frames, so I want to learn more about those. And we’re going to continue developing new analyses,” Willson said. “We’ll continue to develop additional PI Vision screens. And then we’re going to begin rolling out these PI tools to our other facilities.”
The facilities also plan to continue their efforts with machine learning, using AWS and Tableau to make even better use of the data.