I attended my first ARC Forum event, virtually, in February, and here I’ll talk about two sessions that I attended focused on using technology for operations and maintenance. A couple of themes ran throughout both presentations, despite the use case or the industry, from software solution providers to equipment manufacturers. Two things are most important: data and people. Does that seem counterintuitive? Do we use data to do the work of people? Or do they work together? Can the answer be both?
This article is part of our monthly Your Space column. Read more from our Your Space series.
To help find some answers to those questions, let’s consider some specific case studies from the two sessions: “Digital Twins for Improving Operations and Maintenance” and “Machine Learning and Analytics in Operations and Maintenance.” Emily Grams, process analytics engineer at Owens Corning, outlined the company’s digital twin solution used at its residential roofing plant. For this production process, the plant mixes asphalt with a pre-heated filler material, and the plant was having issues with its filler system. The technical lead noticed that it was not able to maintain optimal temperature for production.
The team used a digital twin, Grams said, to understand what had changed in the process. Analyses showed an issue with the filler heater, which was significantly limiting the filler temperature and the overall process. “This was an issue that the tech leader mentioned would normally have taken weeks to identify and traditionally would not have been noticed by the plant until it was causing a significant run issue,” Grams said. The plant made the necessary repairs and tracked those variables to prevent future issues and if possible, further optimize the production process.
This goes a step further than traditional statistical process control (SPC), Grams said. By identifying ranges for the filler system variables and monitoring how well the process can maintain those set points, the model can see the real-time impact on the process. It’s used not only to keep track of individual variable performance within a certain range, but, “We’re actually looking to control particular inputs and see how they affect an output. That’s the main benefit of using a digital twin to model your process,” Grams says.
Jane Arnold, senior vice president and head of global process control technology at Covestro, detailed the work that the chemical manufacturer has done in going digital, after making the announcement to do so in late 2017. The company has worked for years to measure and collect operational data, putting together its process centered solution for operations—ProDAVis (Process Data Analysis and Visualization).
One common problem Arnold found with plant operators was that they didn’t necessarily trust what data was coming back to them. They would assume first that the sensor was bad. If the data was good and not the result of a bad sensor, over time that could become a significant problem in production, Arnold said. Operators could compare measured field data with dynamic simulators to find anomalies, but it still required human intervention to diagnose problems. “Is there something that can automatically tell us where there’s a problem with a sensor,” Arnold asked.
Covestro found help from Aperio Systems, which makes industrial data integrity solutions. By integrating the into Covestro’s current system, Aperio’s solution automatically detects problems with sensors and much more. Covestro tested the software through some pilot projects with detecting forged data and other use cases, but the big break-through happened with redundant sensor failures. A Covestro plant in China had to shut down production, when the system couldn’t tell which sensor was bad. As a test case, Arnold’s team sent all the autonomized data to Aperio, and the system easily detected which sensor was bad. “That perked our interest,” Arnold said. Aperio tackled a broader set of data, and not only found the redundant sensor failures, but other problems the team was unaware of. Covestro is currently connecting the Aperio solution to its software, where the team can begin testing its function in real-time.
Right data, right time
Both Grams and Arnold echoed a message about quality data. During the panel discussion, Grams was asked where do you start? What are the first steps in setting up a digital twin? Her answer was not surprisingly at the data. She said start with defining the inputs and outputs of the process; learn the different sensors that you have that can track flow levels; and understand the capabilities that you have. “Make sure you have all the relevant pieces of information to model the system, and once you’ve established the baseline with all the relevant data, you can go in and use that data and establish relationships,” Grams said.
Arnold said data excellence has been a priority at every step at Covestro, yet acheivement in that is still challenged by many factors: the need to upscale the workforce to meet technology needs; data security risks; the lack of effective applications; and redefining roles and responsibilities to accommodate new technologies.
Many facilities are not lacking in a quantity of data; in fact, some have too much and don’t know what to do with it. “The right data needs to be available at the right time for operations to make timely and accurate decisions,” Arnold said.
Domain knowledge
Even with a process to collect data, is it collecting the right data? The data can help find the answers to help production, but operations has to be asking the right question. That’s where people come in. Nearly every speaker in both sessions talked about the need for domain expertise. The knowledgeable operators and maintenance team that know the equipment and the process are needed to refine the data and focus on the specific problem that company is trying to solve with technology. Domain knowledge is necessary to model, understand, and utilize a digital twin, as well as automate any production process with analytics and machine learning.
Shawn Collins, chief revenue officer at Flanders, a motor manufacturer and repair service company, is interested in using the power of AI, condition-based monitoring, and machine learning to optimize systems and help maintenance and operations teams. What’s most necessary to make that work, as he sees it? Domain expertise in combination with these emerging technologies. “There are many different places you can get that deep domain expertise, but you need to apply it to the problem you’re trying to solve, so you know what data to collect, where to collect it from, how often you need to collect it, the resolution of the data you need to collect,” Collins said. “When you repair things for a living, you learn a lot about failure modes. What we’ve found is a lot of people are looking in the wrong place for the answer.”
Mike Brooks, senior director of APM consulting for AspenTech, joined the panel discussion, where he also hailed the need for domain knowledge to make sense of the data. “That’s the answer. What’s the question? What’s the problem you’re trying to solve?” Brooks asked. “The most important guide rail for machine learning is the domain knowledge. That will make sure you’re tracking toward a causation, rather than a correlation,” he added.
Shifting mindsets
Even with software to automatically detect process anomalies, Arnold noted that the technology requires operations to feed new skill sets and habits. “It’s important to keep all of the data up to date,” she said. “It’s human nature. They put it off to do it later, and it never gets updated properly.”
For Owens Corning, Grams said the shift has required a new mindset too. “The digital twin is a technology that’s useful in understanding your process, but without data gaps being closed and proper maintenance procedures in place, process modeling is limited or not possible at all.” Experts are needed to fill data gaps and uphold maintenance procedures. Resources also need to shift to a predictive mindset, Grams said, from preventive. “It’s difficult to optimize for the future, if you’re constantly thinking about the next fire to put out,” Grams said.
- Read "Case study: How digital manufacturing helped one facility survive after losing 90% of its revenue"
The moderator of the machine learning and analytics session, Ed O’Brien, director of research at ARC Advisory Group, said that before tackling data analytics or machine learning projects, organizations should look inward to ask: are they ready? This involves considering the skill sets of personnel, training opportunities, vendor professional services, and the journey that’s right for the organization. Sustaining the momentum needed for a culture shift within a company requires a lot of supportive staff to be on-board.
Both the digital twin modeling at Owens Corning and the digitalization process at Covestro required immense amounts of quality data and strong knowledge of the process at hand. Will data analytics and digital twins with augmented reality (and an army of autonomous robots) replace people on the shop floor? No, people are still an integral part of the process. Digitalization can help do some work, and it’s most useful focused where analysis and data can find the answers or problems quicker than people. Operators are needed to identify those questions, or collect data accurately, so the technology can get it right.
