Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a Steel Mill Cut Furnace Energy by 25 Percent with Digital Twins
Transcript
- Lucas: If these conversations have moved your work forward in some small way, we want to keep doing them. And the way we keep doing them ad-free is listener support. Luna: Yeah, it's literally just people chipping in what a coffee costs. Lucas: Exactly. Buy me a coffee dot com slash fexingo. If today was useful, that's the way to make sure the next one is too. Luna: Appreciate it. Back to the furnace. Lucas: Alright, so I want to talk about a specific operations win that doesn't get enough attention. It's from a steel mill. But not just any steel mill—Nucor's mini-mill in Crawfordsville, Indiana. Luna: The one that pioneered thin-slab casting back in the 80s? Lucas: Same plant. But the innovation here isn't about casting. It's about the electric arc furnace. That's the big power-hungry machine that melts scrap steel. It eats about 400 kilowatt-hours per ton of steel. For a mill that produces two million tons a year, that's a massive electricity bill. Luna: So they built a digital twin of the furnace to optimize energy use? Lucas: Exactly. And they cut energy consumption by 25 percent. That's not a pilot project number. That's real, sustained savings across multiple furnaces since late 2024. Luna: Walk me through how the digital twin actually works. I think people hear the term and picture a 3D model you can spin around. Lucas: It's not a visual model. It's a physics-based simulation running in parallel with the real furnace. Every few seconds, it ingests data from sensors—temperature, current, voltage, electrode position, slag thickness. Luna: And it predicts what's about to happen? Lucas: More than predicts. It calculates the optimal power input for the next 30 seconds. The furnace has three electrodes that plunge into the scrap. The arc is unstable by nature—it flickers, it creates harmonics. The digital twin adjusts the electrode position and the transformer tap to keep the arc stable at the lowest possible energy draw. Luna: So it's not just turning down the power. It's smoothing the arc so you waste less energy in flicker and reactive power. Lucas: Right. The system was developed by a company called FreeFall, working with Nucor's engineers. They spent about 18 months calibrating the model on one furnace before rolling it out. Luna: And they needed accurate sensor data for that. Which is actually hard in a steel mill because of heat and dust. Lucas: Huge challenge. They had to install water-cooled cameras and acoustic sensors to measure arc length. They also used strain gauges on the electrode arms to measure physical forces. The model fuses all that data. Luna: So the digital twin is only as good as the sensors. What happens if a sensor drifts? Lucas: The model is designed to flag sensor drift automatically. If the predicted power consumption diverges from actual by more than two percent, the system alerts operators and falls back to a conservative baseline. That's a key design choice—they didn't want a black box that could go rogue. Luna: That's smart. I've seen too many AI projects fail because they had no fallback. Lucas: Exactly. The operators also get a dashboard showing the twin's recommendation and the current arc stability index. They can override it. In practice, they override maybe five percent of the time—usually during unusual scrap mixes or electrode changes. Luna: So you've got a 25 percent energy reduction. What does that mean in dollars? Lucas: For a typical mini-mill, electricity is about 15 to 20 percent of operating cost. A 25 percent reduction on that line item translates to roughly 3 to 4 percent total cost savings. For a plant doing two million tons at say $500 per ton, that's on the order of 30 to 40 million dollars annually. Luna: Per plant. Lucas: Per plant. Nucor has about a dozen electric arc furnace plants in the US. If they roll this out to all of them, we're talking hundreds of millions. Luna: And it's not just cost. It's carbon. If the grid mix is natural gas or coal, saving a megawatt-hour reduces emissions. Lucas: Absolutely. Nucor already positions itself as the green steelmaker because they use scrap and electric arc furnaces instead of blast furnaces. This cuts their Scope 2 emissions by roughly 15 percent per ton. Luna: Okay, but I want to push back a little. A digital twin of a continuous process like a furnace—that's one thing. But a lot of operations are batch processes. Does this approach transfer? Lucas: It transfers better to batch than you'd think. In batch, the twin can simulate the entire cycle and optimize parameters for the next batch based on the previous one's outcome. A pharmaceutical company, for example, used a similar approach to reduce batch cycle time for a biologic drug by 30 percent. Luna: But batch processes have more variability—different raw material lots, different operators. Lucas: That's where the physics-based model helps. Unlike a pure machine learning model that needs thousands of examples, a physics model can extrapolate to unseen conditions because it encodes the underlying laws. The pharma case I mentioned used a model of the bioreactor's mass and heat transfer. Luna: So the prerequisite isn't continuous vs. batch. It's whether you can model the physics accurately enough. Lucas: And whether you can instrument the process. The steel mill succeeded because they invested in sensors and validated the model against real data for over a year. That's the unglamorous work. Luna: What about older mills? Nucor's Crawfordsville plant is from 1989 but it's been upgraded. Can you retrofit a 40-year-old furnace with this? Lucas: FreeFall says yes, as long as the furnace has basic instrumentation—current transformers, voltage sensors, electrode position encoders. Most mills already have that. The additional sensors they added—acoustic, strain gauge—cost maybe $50,000 per furnace. Luna: And the software licensing? Lucas: FreeFall charges an annual license based on the furnace's rated capacity. For a 150-ton furnace, it's roughly $200,000 a year. Against $30 million in savings, that's trivial. Luna: So the ROI is under a month. Lucas: Easily. And the intangible is operator adoption. The operators at Crawfordsville initially resisted because they felt the system was second-guessing them. But once they saw the arc stability index smooth out—the furnace stopped making that harsh roaring sound during unstable periods—they bought in. Luna: The sound is a proxy for wasted energy. Quieter furnace means more efficient arc. Lucas: That's the kind of sensory feedback that makes digital twin adoption stick. It's not just a number on a screen. It changes the feel of the plant. Luna: So what's the next frontier for this? Can they close the loop entirely and run the furnace without human intervention? Lucas: Nucor has publicly said they're aiming for lights-out operation of the EAF by 2028. That means no operators on the melt shop floor. The digital twin is the core of that vision. Luna: That would be a first in the steel industry globally. Lucas: It would. And it shows that digital twins aren't just a buzzword. They're a practical tool for reducing energy, cost, and carbon in heavy industry. Luna: Aligned with the show's theme: process, people, and profit. This one hits all three. Lucas: It does. And it starts with good sensors and a physics model. No magic. Luna: If only every operations problem were that tractable.