How deeptech is already revolutionising EV powertrain engineering

Learning from Nature: The Evolution of Engineering Design

To understand the potential of deeptech in automotive engineering, we can draw inspiration from nature itself. Organisms evolve over millions of years as a direct response to their local environment, with the freedom to create complex organic geometries that perfectly balance physical forces. Think of a bird's wing, optimised for both lift and structural integrity, or a tree's branching pattern that maximises sunlight exposure while minimising wind resistance.

Nature's true genius, however, lies in system-level optimisation. A natural ecosystem, for instance, thrives not because individual organisms evolve in isolation, but because entire networks of organisms evolve together as balanced systems. This concept aligns perfectly with Monumo's two-pronged approach: first, we increase the geometric and parametric freedom in how we define individual components. Second, we combine these components into systems and optimise them all together. This dual approach enables us to find higher-performance system designs that would be impossible through traditional methods.

Where Engineering Must Surpass Nature

While we look to mimic nature's efficiency, there are two critical areas where deeptech needs to depart from the natural process. We cannot wait millions of years to evolve designs - we need to converge on apex designs within weeks to remain competitive. And evolution struggles to adapt when environmental change is rapid, which perfectly describes today's automotive sector, with China's EV dominance and shifting global supply chains.

The Computational Challenge: Why Traditional Methods Fall Short

The traditional engineering approach works well for simple problems. You might vary five parameters in a template with ten values each, creating 100,000 possible designs - manageable with modern computing power in about half a day.
But true geometric freedom changes everything. A freeform topology optimisation approach might require evaluating a 50x50 grid with four alternatives (three materials and air), creating 4^2,500 design variations. Even if you could harness every CPU on Earth, you'd need the entire age of the universe to complete the analysis.

This is why intelligent search algorithms are essential. Brute force computation cannot solve the complex, high-dimensional optimisation problems that modern powertrain design demands.

The Anser® Engine: Bringing Intelligence to Scale

At Monumo, we've developed the Anser® Engine specifically to address this challenge—a multi-physics platform built from the outset with automation, large-scale computing, and speed of execution at its core.

But scale alone isn't enough. The sheer scale of simulation enables machine learning to come into its own, providing the intelligence we require to search these vast design spaces efficiently. Think of it like Google's search algorithm—it doesn't search every webpage for your query but uses intelligent ranking to find the most relevant results quickly.

Real-World Results: Case Studies in Cost Reduction

Let me share concrete examples from our collaborative projects with automotive partners.

Phase 1: Rotor optimisation only - our first project focused on a state-of-the-art automotive motor design. The results were impressive:

• Over 250,000 valid designs were generated in just 3 days from a much larger pool of candidates (over 3Million)
• 8.5% reduction in magnet mass
• 4.1% cost reduction, saving €15 per motor

Phase 2: System-level expansion - we expanded the design space to include rotor radius, stator dimensions, motor length, and gearbox ratio - a total of 28 optimisation parameters. The results were even more dramatic:

• 550,000 valid designs generated in 5 days from a much larger pool of candidates
• 23.0% reduction in magnet mass
• 11.4% cost reduction, saving €43 per motor
• Alternatively, 12.5% loss reduction at the same cost

This level of optimisation is impossible through traditional manual design studies.

The Machine Learning Advantage

The ability to generate large training sets enables us to deploy machine learning to support our large-scale simulation in a number of pathways:

Accelerated simulation: We have demonstrated data-driven models using Vision Transformers that can predict electromagnetic performance with 99% accuracy, while delivering a 10x speed improvement. This has the potential to reduce an optimisation run time from 4 days to 4 hours and works well in cases when optimisations need to iterate quickly as requirements change.

Intelligent optimisation: Machine learning can make the entire optimisation process smarter. We have demonstrated the use of classifiers to identify designs that have a higher probability of passing mechanical constraints before running expensive simulations, spending less time investigating invalid designs.

The Future: Generative Design

We are also exploring the next frontier: generative design. This represents a paradigm shift from searching through design spaces to directly creating optimal designs. Instead of running thousands of optimisation iterations, a trained generator could potentially propose high-performing designs directly.

We are currently demonstrating early-stage generative models that can propose promising starting points for optimisation searches. The ultimate vision is trained generators that can directly produce complete, optimised designs, compressing what currently takes days of optimisation into minutes of generation.

The Technology Development Roadmap

Our experience shows a clear progression in the benefits that deeptech can deliver, and we believe we can go much further as our technology evolves

• Motor-only optimisation: ∼5% cost reduction (∼€23 per motor)
• System-level optimisation: 10%+ cost reduction (∼€46 per motor)
• System-level + freeform + ML/AI: 20%+ cost reduction (€92+ per motor) - projected

The first two levels represent results we're delivering, today, with partners. The third level represents our projected capability, as we integrate increasing component and system-level freedom in our optimisations, enabled by the intelligence from the ML techniques we are working on today.

In conclusion, deeptech is not a distant promise for EV powertrain engineering; it's already a reality and delivering significant, quantifiable benefits. Monumo's Anser® Engine, powered by intelligent search and machine learning, is proving that embracing computational freedom and system-level optimisation is the key to unlocking unprecedented cost reductions, and performance gains, in electric vehicle development.

Simon Shepherd is Head of eDrive and Chief Product Officer at Monumo, a deeptech startup based in Cambridge and Coventry, UK, specialising in AI-driven automotive engineering solutions. Ready to explore how deeptech can transform your powertrain development? Contact our team at monumo.com or reach out directly to discuss your specific challenges and opportunities.