How Dessia helps to accelerates architecture generation for sustainable aviation

Overview

‍

As the aviation industry accelerates its shift toward sustainable propulsion, hybrid energy systems — particularly those combining electric components with hydrogen fuel — are gaining momentum. Yet, this hybridization introduces significant design complexity. That’s why Safran Tech launched a forward-looking initiative to explore how AI could support early architectural choices in this domain.

‍

The challenge? Identifying viable propulsion architectures that meet performance goals, safety constraints, and regulatory requirements, all while minimizing CO₂ emissions, and doing so early enough to influence the rest of the development cycle.

‍

To support this initiative, Dessia and Safran developed an AI-powered software framework capable of automating the generation, evaluation, and optimization of hybrid propulsion system architectures. This case study explores how this solution enables engineers to navigate complexity, accelerate development, and make informed architectural choices from day one.

‍

The challenge: Too many possibilities, not enough time

‍

Designing a hybrid propulsion system means exploring a vast number of possibilities:

‍

• Different topologies (Ways to connect components)
• Multiple component options (Fuel cells, batteries, turbines, motors, tanks)
• Conflicting constraints (Performance, weight, emissions, safety)

‍

Today, engineers rely on trial-and-error, evaluating a few handpicked architectures through simulation. But this approach can’t keep up with the complexity of modern design. The risk? Missing out on more efficient, lower-emission solutions.

‍

What the industry needs is a systematic and automated approach to explore, validate, and optimize system architectures at scale.

‍

Dessia's solution: AI-powered architecture generation

‍

To support Safran Tech’s initiative, Dessia developed an AI-based design application that helps engineers explore complex hybrid propulsion architectures — automatically, and at scale. The goal: remove repetitive tasks, reduce trial-and-error, and support smarter decision-making from the very start.

‍

1. Modeling the system with real engineering logic

‍

The process begins with object-oriented modeling of the propulsion system. Each component  (Battery, fuel cell, turbine, gearbox, propeller) is modeled as an object with defined ports and physical properties. These objects include technological data (hydrogen or electric ports) and built-in design constraints like forbidden or mandatory connections.

‍

2. Exploring architectures with AI

‍

Once components and rules are defined, the tool uses a decision tree algorithm to automatically generate every valid architecture. At each step, the system checks for port compatibility and applies design rules. Invalid topologies are instantly discarded. What’s left? Only technically sound system architectures that are ready for deeper analysis.

‍

3. Filtering with safety criteria

‍

Every generated architecture is evaluated for reliability. If an option doesn’t meet the defined failure rate target, it’s filtered out before going any further. This built-in safety filter ensures that only robust designs move forward.

‍

4. Simulating performance

‍

Feasible architectures are then simulated to assess their performance over a mission. Key outputs include fuel consumption and CO2 emissions. The simulations are based on Safran Tech’s internal energy solver, designed specifically for next-gen propulsion systems like hydrogen or hybrid-electric aircraft.

‍

5. Optimizing with design of experiments (DoE)

‍

To explore trade-offs and fine-tune designs, the tool includes a DoE module. Engineers can vary inputs like fuel cell efficiency or power density and compare outcomes across the architecture space, without redrawing or re-simulating everything manually.

‍

By automating the generation and validation of system architectures, Dessia’s AI solution enables engineers to systematically explore viable designs

‍

‍

Exploring results and selecting the best designs

‍

Once all viable architectures are generated, the application doesn’t just produce a flat list; it presents them through an interactive dashboard. Each configuration is displayed with key performance indicators such as:

‍

• Failure rate
• CO2 emissions
• Number of connections

‍

To simplify comparison, a clustering algorithm automatically groups similar architectures together. This makes it easier to identify trade-offs, filter out non-competitive options, and focus on the most promising solutions.

‍

The result is not just a large design space, but an organized, explainable set of options ready for selection and deeper evaluation.

‍

‍

Results: Turning complexity into clarity

‍

What happens when you feed real engineering constraints, component libraries, and safety rules into an AI-powered architecture generator? You don’t just get a few ideas — you get options. Lots of them.

‍

In this case, the system explored over 2600 unique propulsion architectures, each built from a realistic mix of components: hydrogen and kerosene tanks, batteries, fuel cells, turbines, motors, gearboxes, and propellers. The AI complied with strict design rules and automatically enforced limits on failure rates.

‍

What emerged wasn’t random. It was structured, diverse, and technically sound:

• Hybrid configurations
• Distributed propulsion models
• Fully electric designs
• And setups with built-in redundancy to boost reliability

‍

Instead of sketching out a few safe ideas, engineers now had a broad, validated landscape of solutions to work with, that are ready to be analyzed, compared, and refined. All of it built on real engineering logic.

‍

Conclusion: A new standard for early-stage aerospace design

‍

Safran Tech’s initiative to explore next-gen propulsion architectures illustrates how early-stage design — when powered by the right tools, can become a driver of clarity, speed, and innovation.

‍

With Dessia’s AI-solution, what once required weeks of iteration was accomplished in less than a day, automatically generating thousands of feasible hybrid configurations, while respecting strict engineering rules and safety constraints.

‍

This approach doesn’t just make exploration faster, it makes it smarter. By consolidating requirements, formalizing design rules, and structuring the space of possibilities, teams gain an unprecedented ability to make confident, data-backed architectural decisions from day one.

‍

Looking ahead, the partnership opens the door to even broader perspectives: incorporating more system parameters, expanding simulation capabilities, and embedding architectural decisions into 3D layouts and trade-off environments.

‍

In an industry where every early decision shapes downstream performance and cost, this project marks a decisive shift — from fragmented iteration to structured, intelligence-driven design.

‍