Physics-Driven Generative Design for Laser Powder Bed Fusion in Aerospace
Interview with Thomas Rees – ToffeeX.
ToffeeX is developing a new class of generative design tools that give engineers precise control over how functional geometry is shaped—balancing physical performance with real-world constraints like manufacturability and aesthetics.
Built around multiphysics simulation, adjoint optimization, and a streamlined user interface, ToffeeX enables both early-stage design exploration and late-stage validation across industries from aerospace to thermal management for consumer electronics.
At CDFAM Amsterdam 2025, ToffeeX’s Thomas Rees will present recent advances in the platform, including new tools for customizing optimization behavior and navigating multi-objective trade-offs.
With a background in aerospace engineering, Thomas brings a perspective grounded in applied complexity—where performance, cost, and manufacturability must be reconciled in every decision.
In this interview, we speak with Thomas about what’s changed since ToffeeX’s presentation at CDFAM NYC, how their platform is evolving, and what it means to empower engineers without turning design into a black box.
Marco from ToffeeX presented at CDFAM NYC last year. Can you start by telling us what’s new with the company since then, and give us an overview of what you’ll be presenting at CDFAM Amsterdam?
At CDFAM NYC last year, Marco’s presentation shared some of our early learnings at ToffeeX around creating a better user experience for generative design tools, particularly when it comes to user control and workflow.
One key takeaway was that users tend to reject “black box” experiences when it comes to controlling the design itself. While they may not mind not knowing how a geometry is generated, they do want control over the general characteristics of the geometry, especially in the context of manufacturability or geometric aesthetics.
Since then, we’ve been thinking about the best ways to give users more control over the kind of geometry they generate through ToffeeX. We’ve been developing tools that allow for far greater customization of design characteristics while maintaining the strength of our optimization algorithms.
These tools give engineers the ability to guide design and optimization outputs not just toward non-negotiable requirements like manufacturability, but also toward specific geometric qualities relevant to their requirements.
What kinds of physics models and solvers are you integrating today, and how do you balance accuracy versus computational speed for fast optimization?
Our physics library is constantly expanding as we identify new, cross-industry opportunities where our generative design tools can have an impact. Right now, we’re focussed on the two-phase fluid model we’re building.
It is capable of simulating phenomena like evaporation and condensation.
This has drawn interest from a broad range of industries, from data center cooling to aerospace.
We’re also developing a coupled thermal-fluid-structural solver, which enables simultaneous consideration of thermal stresses, pressure containment, and external structural loads.
When it comes to balancing simulation fidelity with computational speed – this is always a challenge (or an art…).
Our philosophy is that our optimization tools don’t necessarily require absolute high-fidelity simulation capability per se. For that reason, we use lower-fidelity simulations for early-stage exploration and reserve high-fidelity models for validating the most promising design candidates.
Our platform supports both modes, allowing users to quickly iterate through the early design stages before validating and ensuring that results are grounded in good physics simulation models.
Can you walk us through how these different physics domains are coupled together in your workflow, and how they directly inform the geometry generation process?
We try to keep things as simple and modular as possible for the user. Depending on their goals, engineers can selectively enable or disable specific physics and optimization models. They might choose to simulate a physical system without optimizing it, or they can co-optimize multiple physical behaviors simultaneously.
To get technical for a moment: the nice thing about the adjoint-based optimization frameworks which we use is that the Lagrangian function which we construct is a linear sum of individual objectives and constraints. This makes it mathematically straightforward to add or remove different physics domains or goals. It gives users a flexible toolkit without added complexity.
ToffeeX allows engineers to optimize for multiple objectives simultaneously — like fluid dynamics and thermal performance. How does the platform help users navigate the trade-offs when those goals are in tension?
This is where things can get complex!
The key to managing trade-offs is being able to visualize, analyze, and interpret the Pareto front of the optimization problem.
This is a map of optimal solutions between competing objectives. Once that’s established, engineers choose the appropriate trade-offs with clarity and confidence.
So – construct the Pareto front for your optimization problem and all your problems are solved!
Or not… The challenge scales quickly. With each additional objective function, the curse of dimensionality strikes – the dimensionality of the Pareto front increases, making it harder to explore exhaustively. With four competing objectives you have a four dimensional Pareto front – how on Earth do you visualize this?!
Ultimately, our approach at ToffeeX is not to automate the entire process, but rather to empower the human. We provide intuitive tools that allow them to target and explore specific regions of the Pareto front – those they deem most promising or relevant based on their expertise.
For companies used to traditional design workflows with heavy iteration and simulation loops, what does adopting a ToffeeX-driven approach change in their development process? How early should teams be thinking about using your platform?
At ToffeeX, we’re not trying to eliminate iteration, we’re trying to accelerate it. Iteration is how engineers learn, and we’re fundamentally against hiding that process behind some automated black box.
Our platform enables rapid iterations that allow teams to explore radically different design directions early on directions that might never be considered in traditional workflows.
For example, early in a project, an engineer could use ToffeeX to compare the performance trade-offs between a highly complex (but expensive) additively manufactured component and a simpler (cheaper), stamped alternative. Maybe the performance gains of the additive component justifies the increased cost. But without a rapid, easy way to explore this trade off it is difficult to justify putting the effort into performing the study if you know you are working in a particularly cost-sensitive industry. However, these kinds of potentially high-impact design decisions can be explored quickly, cheaply, and thoroughly with our tools.
As for when to use ToffeeX? Simple – the earlier, the better.
As soon as component requirements are available from systems and cost engineering, designers should be using tools like ours to rapidly investigate their ideas. These early insights can shape the entire product architecture and as the project matures more and more detailed design and optimization cycles can be run.
Many engineers are domain specialists in mechanical or thermal design. How have you structured ToffeeX so that non-experts in simulation, like generalist engineers, can still confidently generate and evaluate high-performance designs?
This is where UI/UX decisions make all the difference. Our interface is intentionally streamlined, designed for generalist engineers who understand the functional requirements, use cases, and engineering contexts of their components, but not necessarily the inner workings of multiphysics solvers.
Most of the heavy lifting, meaning meshing, multiphysics solutions, and adjoint optimization, can be abstracted away from the user. Meanwhile, post-processing tools make it easy to understand things like trade-off analyses and Pareto fronts, even for non-experts.
Of course, if a domain expert is using our tool then they’ll likely want to dig down into the gory details of what is going on behind the scenes – which they should be fully empowered to do (albeit with a ‘here be dragons’ sign).
That said, we fully acknowledge the need for deep engineering expertise.
An analogy I like to make is to medicine – AI tools can make huge advances in diagnosis and interpretation of things like scans or biopsy samples. However, I think it unlikely anyone would start a treatment course without the review of an expert physician!
As you prepare to present at CDFAM Amsterdam, what are you most looking forward to sharing, and what kinds of conversations or collaborations are you hoping to build with other computational design and engineering experts?
CDFAM is such a unique event because it brings together experts from a (sorry) niche field – people who are normally scattered across disciplines and industries. What makes it so refreshing is that it’s not anchored to a single vertical like automotive or aerospace. We get to hear from architects, software developers, engineers, industrial designers, and more, all tackling computational design problems from different angles.
At CDFAM Amsterdam, I’m particularly excited to share some of the aerospace-focused projects we’ve been working on at ToffeeX. At my core I’m still an aeronautical engineer so any time I get to talk about aerospace I’m happy! It’s such a complex, exciting, and high tech field, and while at ToffeeX we’re just one small piece of the puzzle, the work is incredibly rewarding.
Most of all, of course, I’m looking forward to the rich conversations and idea exchanges that happen in between the talks – the kind of cross-pollination that only happens when the right people are in the same room.
To learn more, join us at CDFAM Amsterdam 2025, where you can connect with Thomas Rees and others working in ‘niche applications’ of computational engineering.
If you’re interested in design optimization, simulation, and data driven tools that let engineers stay in control of their designs, you’ll be in good company.
And for those who prefer their geometry with some semblance of supporting math, be sure to check out ToffeeX’s research papers and technical publications—they cover far more than can fit in a 20 minute conference presentation.
