Digital Twins and AI Helping Industries Stay Flexible During COVID-19 Uncertainties
When COVID-19 clobbered all of humanity in early 2020, the resulting government lockdowns in countries and communities across the world negatively impacted entire industries and turned people's lives and whole economies upside down. For many, a full recovery may still be months or years away.
But for some large industrial operations, such as oil refining or glass manufacturing, the innovative use of digital twins based on AI technology has meant the difference between succumbing to the interruptions or rapidly adjusting to the new normal established by the pandemic.
We all remember the shock that the oil industry experienced in the spring of 2020, when the value of crude oil futures temporarily dipped into negative territory, meaning companies were literally paying others to take a glut of oil off their hands. But in addition to impacting the price of oil, the lockdown also threw oil refineries’ carefully orchestrated chemical processes into disarray.
Figuring out how to efficiently operate a hugely complex piece of machinery, like an oil refinery, is not easy during the best of times. When you throw COVID-19 into the mix, you end up with a lot of unknowns. However, the refineries that had invested in digital twin technologies–where machine learning techniques are used to create a statistical model of how an actual physical machine works–were able to navigate their way through it.
Paige Morse, the industry marketing director for chemicals at digital twin software company AspenTech, says her customers are still adjusting to the supply and demand impacts wrought by COVID-19.
“Refineries right now are not operating the way they normally would,” Morse told Datanami in a December interview. “Refineries love to run at least 92% throughput. You look at the number from the U.S. Energy Information Administration. They’re in the 70s right now. That is not how a refinery wants to run.”
Running a refinery at 20% reduced capacity can introduce problems that are hard to anticipate. The stability of a chemical reaction may be compromised by running outside of normal operating parameters, Morse said, which can compromise the safety of the plant. Instead of running all the distillation columns at a reduced capacity, a company may elect to completely shut down an entire column, enabling the other three or four to run at the prescribed rate.
“That’s been a very big issue for refineries over the last several months, because they’re no longer in the preferred design range for equipment. They’re being asked to run much more slowly,” she said. “That’s why units are shut all over the world because they can’t operate them safely, so it’s better to shut them.”
Delays in crude oil deliveries also impacted the carefully crafted schedules of refineries. When an oil tanker is not able to make an on-time delivery due to a harbor’s decision to reduce the number of workers or docks available to offload goods–a situation that is currently happening at the twin Los Angeles/Long Beach harbor–then downstream customers are inevitably impacted.
“Crude types vary significantly, so if you get a different crude coming into your refinery, it’s going to completely change how it operates and what the cuts are and how you actually direct different product streams,” Morse said. “Refineries are extremely complex and this is where digital twins can be so valuable. It’s true in chemical plants as well. There’s a lot of challenges with feedstock variance. A lot of plastics processes are not easy to manage, so digital twins can be very useful there as well.”
2020 was a tough year for everybody. For customers of AspenTech, which was founded nearly 40 years ago as a joint project out of MIT between Dow Chemical and Exxon, the ability to leverage AI technology to see how different data inputs impact complex machinery and supply chains was invaluable.“
“Some of our customers have been running hundreds of scenarios as they look at changes in their ability to get feedstocks, to make product, and move their products around the world,” Morse said. “You start with ‘This is the normal world. What did 2019 look like?’ Then they started making huge adjustments for what 2020 has become.”
Getting the Last 5% to 10%
Over in Europe, the German glassmaker SCHOTT AG has been using digital twin technology from a company called NNAISENSE to help it optimize massive new melt tanks used to manufacture glass vials, which have been in high demand due to COVID-19.
SCHOTT uses melt tanks that are about 30 meters long and infused with about 300 sensors. It can take up to 12 hours for the raw material to make its way through all the different chambers that refine the glass. SCHOTT, which has been in business for about 130 years, has amassed a huge database of time-series data gathered from these types of tanks, which run continuously for years. This gives it a head start in building digital twins.
According to NNAISENSE CEO and co-founder Faustino Gomez, deep learning techniques go far beyond traditional chemistry or physical simulators that have been used for digital twins in the past.
“If you try to use traditional statistical methods to do it, or an off-the-shelf chemistry simulation, when you put in the boundary conditions–the shape of the tank and that stuff–you can get some results. But it’s very hard to model all the non-linearities that are there, the disturbances and the specifics of the particular tank and hardware and the physical system you’re talking about,” Gomez told Datanami.
“Basically, we train the neural network to behave like the tank,” he continued. “Once you have that, then you can do things like evaluate alternative control actions that the operator can do, and determine the consequences that will have for the process.”
Gomez founded NNAISENSE with several other members of the Swiss AI Lab (IDSIA), where they were at the cutting edge in developing long short-term networks (LSTM) algorithms and HighwayNets, among others. He couldn’t go into the specific algorithmic techniques that the company employed with Schott AG. But needless to say, there was nothing off-the-shelf about the implementation.
“Even though the basic thermodynamics and chemical process have been understood for a long time…it’s this uncodified knowledge of how to do that,” Gomez said. “They have intuition and certain heuristics that they apply to get it done. But no human could really ever take into account all 300 sensors and say, Okay, this is what should be done.”
When SCHOTT decided to invest about $40 million into building the new tanks for the specialty glassware (in part due to increased demand from COVID-19), the company decided to bring in NNAISANCE to get the last 5% to 10% of efficiency and quality out of the melt tanks.
“They kind of reached the plateau of the efficiency they can get using the traditional methods. That’s why they came to us,” Gomez said. “They’re the best in the business. But there’s still a certain amount of guesswork when dealing with an entirely new material. There are certain things that are hard to account for in anything they do, like things outside the tank, ambient conditions and stuff like that, that can affect how the tank works.”
Digital twins aren’t new, but when they’re infused with the latest AI technology, they can take simulation and modeling of the physical world to a whole new level.
This article first appeared on sister website, Datanami.