15y

How Simulation & AI are Redefining EV Development

When we see an Electric Vehicle (EV) drive by, one of the first things we notice is its silent operation. It seems effortless but, for us in the engineering world, that silence is the loudest challenge we face.

Without the noise of an internal combustion engine to mask vibrations, every gear mesh and mechanical whine becomes audible. Furthermore, the instant torque delivered by an EV puts immense stress on mechanical components.

One of our consultants, Pedro Gomez, recently had the opportunity to dive into these challenges, working within an electrification department on gearboxes designed specifically for the next generation of EVs.

The Digital Proving Ground

Generally, within an industrial setting, one cannot simply “build and break” prototypes; it is too costly and too slow. Instead, they are tested and “broken” digitally.

  • Structural Integrity (FEA): Using Finite Element Analysis (FEA), the durability of the gearbox housing and its components is assessed. Think of this as a digital stress test. Virtual mechanical loads and constraints that mimic the real working conditions are applied to the model. This ensures that the housing will not crack or warp, all while trying to keep the weight as low as possible for battery range.

  • Lubrication (CFD & SPH): This is where the job gets even more fascinating. In an EV gearbox, oil is not meant just for lubrication; it is also a coolant. Smoothed Particle Hydrodynamics (SPH) is used to model the oil and its interaction with solid bodies. Unlike traditional fluid dynamics techniques that treat liquid as a block, SPH allows the simulation of millions of individual "particles" of oil. This allows the visualization of how and where the oil splashes, coats the gears, and flows into the bearings. Armed with these types of analyses and results, it can be ensured that no component runs dry, regardless of vehicle speed.

The Future: AI and Geometric Deep Learning (GDL)

While FEA and CFD are incredibly powerful, they have one major drawback: solution time. A high-fidelity lubrication model or a system-level structural simulation can take several hours or even days to run on powerful computing clusters.

This is where the industry is slowly shifting towards AI and GDL. Unlike simple AI models that deal with spreadsheets or text-based inputs, GDL learns directly from the 3D shape (the mesh) of the mechanical component.

  • Mining Historical Data: Every engineering company sits on a goldmine of old simulations with gigabytes of past projects. GDL models can digest this historical data to learn the relationship between a specific 3D shape and its performance (strength, deflection, etc.).

  • Prediction vs. Calculation: Once trained, the GDL model does not need to solve complex physics equations from scratch. It can look at a new gearbox housing design and predict stress hotspots or fluid flow patterns in seconds rather than days.

This allows the exploration of thousands of design variations instantly. One may then ask, "What if we change this fillet/chamfer radius?" and get an immediate answer, saving the heavy-duty physics solvers for a final validation run.

The Human/Engineering Element

The transition to AI-assisted engineering does not mean we stop understanding physics. It means we spend less time waiting for progress bars to complete and more time innovating.

At VeroTech, this is the mindset we bring to our clients: blending the reliability of high-fidelity simulation with the incredible speed of data-driven prediction.

The future of engineering isn't just about building better meshes; it's about making our past data work for our future designs.