Simulation App Speeds Up Electric Truck Battery Design

We live in an era where waiting forty-eight hours for a package feels like a personal affront. Consumers demand speed, and that demand has forced a massive logistical shift. We have fulfillment centers everywhere now, but the middle mile—the long stretches of highway connecting these hubs—remains the backbone of the system. For founders and engineers in the EV space, this is the hardest problem to solve. It is not about making a cool car for a weekend drive; it is about making an electric truck that can survive 500 miles of highway vibration and thermal stress every single day for a decade.

The weak link in this chain isn't always the chemistry; often, it is the manufacturing. Specifically, how we join thousands of individual cells into a single, cohesive pack. A new project, nicknamed LaserBATMAN, recently wrapped up a three-year stint involving Volvo Group, Aurobay, and the Technical University of Denmark. They weren't fighting crime; they were fighting battery degradation through multiscale simulation. The result is a design tool that uses deep neural networks (DNNs) to predict how a single laser weld on a busbar affects the lifetime of a multi-ton vehicle.

The Manufacturing Bottleneck

If you are building in the hardware or climate-tech space, you know that scaling from a prototype to a production line is where dreams go to die. In battery packs, laser welding is the industry standard for connecting cells to busbars. It is fast and precise, but it is also fickle. If the laser doesn't penetrate deep enough, you get high resistance and heat. If it goes too deep, you damage the cell chemistry. Both lead to a truck that dies on the side of the I-95 two years before its warranty expires.

Historically, testing these designs meant building physical packs, cycling them until they failed, and then performing autopsies. That is too slow for the current pace of AI-driven development. The LaserBATMAN consortium decided to flip the script by creating a "digital twin" of the welding process. They used COMSOL Multiphysics to model the electrochemistry at the cell level, then fed that data into a neural network to create a reduced-order model. For those of us in the builder community, this is the real takeaway: they took a complex physics problem and turned it into a software tool that an engineer can use to run "what-if" scenarios in seconds.

Small Welds, Big Data

The app developed by Resolvent, a Danish engineering consultancy, acts as a bridge between high-level physics and practical R&D. When you are sitting in a design meeting at a company like Volvo, you don't want to wait three weeks for a supercomputer to tell you if a thinner busbar will overheat. You want to toggle a slider and see the result immediately. The app allows users to input geometry, cell counts, and ambient temperature, but the "secret sauce" is a feature called the Busbar Weld Design Tool.

This is what I call a "founder's tool." It allows an engineer to simulate production errors. You can intentionally input a "bad" weld—one with shallow penetration—and see exactly how that affects the Solid-Electrolyte Interface (SEI) layer thickness over thousands of charge cycles. The SEI layer is essentially the battery's skin; if it grows too thick due to heat and stress, the battery loses capacity. By seeing this degradation in a simulation, builders can find the "sweet spot" where the design is robust enough to handle manufacturing variances without being over-engineered and too heavy.

The Shift to Builder-First Simulation

For a long time, high-end simulation was locked away in the ivory towers of academia or specialized simulation departments. The rest of the team just got the reports. What Resolvent and the LaserBATMAN team have done here is democratize that data. By wrapping a complex neural network in a clean UI, they allow the people actually making the trucks to experiment. If the simulation shows that better airflow could eliminate two battery cells while maintaining the same range, that is a massive win for the bottom line. It reduces weight, cost, and environmental impact.

From a skeptical founder's perspective, I always look for where the hype ends and the utility begins. We hear a lot about "AI in manufacturing," and usually, it's just marketing fluff. But using a DNN to speed up electrochemical modeling—that is a legitimate use case. It solves the speed-vs-accuracy trade-off. You get the accuracy of a physics-based model with the speed of a lookup table.

What This Means for the Future of Logistics

The long-term goal here isn't just better welds; it's making electric freight viable. Currently, fleet owners are hesitant to go all-in on electric because the ROI depends entirely on battery longevity. If a pack fails early, the economics of the entire fleet collapse. Tools like this reduce that risk. They allow manufacturers to guarantee a certain lifetime because they have already "run" that truck for ten years in a virtual environment before the first weld was even made.

For those of us watching the intersection of AI and heavy industry, this project is a roadmap. It shows that the future of building isn't just about better hardware, but about the software layers that allow us to understand that hardware. We are moving away from trial-and-error and toward a model where the manufacturing process itself is optimized by the same intelligence that powers the vehicle's navigation.

The Takeaway for Builders

Stop looking at manufacturing as a separate step from design. The most successful founders in the next decade will be those who integrate simulation and machine learning directly into their production workflow. If you can predict a failure at the weld level, you can dominate at the fleet level.

The LaserBATMAN project might have a catchy name, but its real value is in the boring details: the SEI layer thickness, the laser penetration depth, and the thermal turbulence. That is where the money is made, and that is where the electric transition will actually be won. We don't need more hype; we need more tools that tell us the truth about how our products will perform in the real world.

---

Read the original at IEEE Spectrum →