Generative AI for Design and Sustainability in European Manufacturing

Over the last decade, Europe’s manufacturing sector has made enormous progress in digitalisation connecting machines, data, and people through advanced automation and Industrial IoT. Yet, as industries move toward sustainable and circular models, digitalisation is no longer enough, it must evolve into digital intelligence that can design, adapt, and regenerate. The Horizon Europe DaCapo project embodies this evolution by embedding intelligence into every stage of the product lifecycle. Its vision is to support and accelerate the adoption of Circular Economy practices by helping industrial users make informed, data-driven decisions about sustainability and resource efficiency. Through AI-based design, predictive models, and interoperable digital assets, DaCapo turns raw data into actionable insight, enabling not only optimization, but a conscious transformation toward more sustainable value chains.

In this context, Generative Artificial Intelligence (GenAI) is reshaping the landscape. Beyond analysing data, GenAI learns from patterns to create: it proposes new layouts, predicts system behaviour, and reimagines products and processes with sustainability built in. In manufacturing, this paradigm merges design thinking with AI-driven reasoning and sustainability engineering, turning digital tools into active co-creators that embed eco-performance directly into design. Within DaCapo, this philosophy materialises in a family of smart digital tools that together form an AI-driven ecosystem:

• the Eco-Storage Architect, applying conditional generative models to design sustainable warehouse layouts;
• the Eco-Remanufacturing Architect, reconstructing geometries for predictive, resource-aware repair; and
• the Eco-Case Architect, combining diffusion models and 3D workflows for eco-evaluated product customization.

Designing Sustainable Layouts with the Eco-Storage Architect

(Video 1 – EcoStorage Architect Demonstration)

Industrial layouts define how efficiently materials, energy, and people move. Traditional planning tools often optimize only for throughput or cost. In collaboration with PESMEL Oy, the Eco-Storage Architect[1] brings genAI into warehouse design, introducing sustainability as an active design constraint for automated warehousing systems.

Using a Conditional Tabular GAN (ctGAN),  the tool learns from historical and simulated warehouse data to propose new 2D configurations of aisles and stacker cranes. Unlike manual CAD iterations, the ctGAN can explore thousands of feasible alternatives under user-defined technical and eco indicators—space usage, steel consumption, CO₂ footprint, and roll accessibility. During training, the ctGAN model learns the statistical dependencies among parameters; at inference, users simply set targets (e.g., “maximize capacity, minimize CO₂”), and the generator conditions on them to synthesize realistic layout candidates.

A post-processing layer evaluates distributional fidelity (≈ 92 %), ensuring that generated layouts remain consistent with real operational constraints. In parallel, a genetic algorithm (GA) evaluates and refines the layouts proposed by the ctGAN using a multi-objective fitness function that balances technical efficiency and eco-indicators. In this way, the tool provides decision-makers with a concise, yet diverse set of options optimised across both operational and environmental dimensions.

Why it matters

The Eco-Storage Architect moves sustainability upstream. Instead of assessing impact after construction, it lets engineers generate and evaluate sustainable options from the very first sketch.
Its generative approach to sustainable layout design can be replicated across logistics centres, manufacturing plants, and other facility-planning contexts, supporting cross-sector energy and material optimisation. It demonstrates that generative AI can act not only as a creativity engine but also as a measurable sustainability optimizer, directly supporting the EU’s goals for digital and green transitions.

Learning to Repair — The Eco-Remanufacturing Architect

(Video 3 – Eco-Remanufacturing Demonstration)

Extending product life is essential to the circular economy. Yet deciding when and how to repair a degraded part remains complex. The Eco-Remanufacturing Architect[2], developed in collaboration with GKN Aerospace Engine Systems, uses AI to reconstruct worn geometries and estimate the sustainability impact of each repair option.

The workflow begins with Holistically Nested Edge Detection (HED) applied to metallographic images of repaired components, isolating contour shapes of additive layers. These curves are encoded into Fourier Descriptors, transforming spatial geometry into frequency-domain features that capture both global and local shape characteristics. Two Random Forest Regressors then learn to map process parameters—laser power, wire feed, speed—to the real and imaginary components of these descriptors, enabling direct prediction of repaired shapes from process settings alone. A Variational Autoencoder (VAE) augments the dataset with synthetic but realistic parameter combinations, ensuring robustness even with limited experimental data.

Evaluation across 200 samples confirms high predictive accuracy (cosine similarity ≈ 0.99; RMSE ≈ 0.33).
Beyond geometry, the tool estimates energy consumption, material use, and CO₂ emissions for each predicted repair.

Why it matters

Eco-Remanufacturing Architect turns repair into a data-driven design problem.
Instead of relying on experience or trial and error, engineers can predict the geometric and environmental outcome before touching the machine. This reduces waste, shortens repair cycles, and supports the EU’s shift from linear production to regenerative manufacturing. Although demonstrated with GKN Aerospace, the same predictive framework can be retrained for repair and remanufacturing in tooling, energy, or automotive sectors, making additive sustainability intelligence transferable across industries.

From Prompt to Product — The Eco-Case Architect

(Video 4 – Eco-Case Architect Pipeline)

At the product level, generative AI opens new frontiers for customization. In collaboration with Fairphone, the Eco-Case Architect demonstrates how text-to-image diffusion models can be combined with manufacturing simulation to create custom 3D-printed phone cases that balance creativity, functionality, and eco-efficiency.

The workflow links diffusion models such as Stable Diffusion, Flux 1-Schnell, and SD3-Medium models to Blender-based geometry processing. A user describes a design concept (“leaf-pattern texture,” “geometric relief,” “minimalist wave”), and the system generates a 512×512 texture, maps it onto a base mesh, and applies controlled surface displacement. Each resulting design is automatically evaluated for semantic alignment (CLIP Score ≈ 0.27, R-Precision ≈ 0.9), aesthetic preference (HPS-v2 ≈ 0.25), and eco-indicators—build time, material use, energy, cost, and carbon footprint.

Average results show ≈ 205 min print time, 0.69 kWh energy, and 0.14 kg CO₂ per case—comparable to or better than baseline designs—while maintaining a printability score ≥ 12.8/15. Among the evaluated models, Flux 1-Schnell achieved the best balance between semantic fidelity and computational cost, allowing for efficient iteration cycles. The tool integrates directly with slicing and simulation environments, ensuring that each digital prototype is manufacturable before printing. This coupling of generative design and eco-assessment accelerates design loops, reduces trial-and-error waste, and embeds sustainability feedback within the creative process.

Why it matters

Eco-Case Architect shows that generative AI can extend beyond imagery into manufacturable, sustainability-aware product design. It allows users to translate creativity directly into 3D objects whose ecological impact is quantified instantly, connecting human imagination with measurable circular performance. By embedding environmental assessment within the design loop, it helps shift product creation from intuition to evidence-based decision-making. While developed with Fairphone for consumer electronics, its generative design workflow can be replicated to other product lines such as wearables, packaging, or healthcare accessories, where personalised, low-impact production offers both sustainability and market differentiation.

Toward an AI-Driven Circular Pipeline

Each of these tools plays a distinct role, yet they form a coherent digital ecosystem:

Across all tools, sustainability metrics are not an afterthought—they are embedded within the AI generation itself. Each model, whether generative or predictive, integrates eco-indicators directly into its decision space as part of an in-loop sustainability assessment. This means that every new layout, repair strategy, or design proposal is generated under explicit sustainability constraints, producing outputs that are both technically valid and environmentally optimised.

Beyond individual functionality, all tools are developed following a common interoperability layer based on the Asset Administration Shell (AAS) standard. Each AI service is represented as a digital asset with machine-readable metadata, ensuring full traceability and alignment with Industry 4.0 and Digital Twin architectures. This approach allows every model to be deployed, updated, or reused across systems without re-engineering data structures or communication interfaces.

On top of this interoperability layer, DaCapo integrates the Circular Economy Decision Support System (CE-DSS)—a shared intelligence layer that aggregates data and insights from all Eco-Architect tools and other digital tools developed within the project. The CE-DSS analyses sustainability indicators, compares alternative options, and guides users toward the most circular and resource-efficient decisions. Acting as the cognitive core of the ecosystem, it connects AAS-based digital assets with practical decision-making, ensuring that AI and data-driven outputs translate directly into measurable environmental and economic value.

This interoperability makes the DaCapo ecosystem more than a collection of isolated prototypes: it becomes a scalable digital framework where AI services, environmental intelligence, and industrial twins co-evolve. By combining AAS-compliant data exchange with embedded sustainability reasoning, the project sets the foundation for a new generation of modular, transparent, and circular AI tools—capable of operating within larger European manufacturing platforms and supporting the long-term vision of interoperable, sustainable, and human-centric digital factories.

Why It Matters for European Industry

Europe’s manufacturing base is evolving under growing pressure to remain competitive, sustainable, and digitally sovereign. To meet these challenges, industries must not only digitalise processes but embed intelligence that drives circularity—optimising resources, extending product life, and reducing environmental impact. The DaCapo ecosystem demonstrates how this can be achieved in practice: through replicable, AI-driven tools that link data, design, and sustainability across the full product lifecycle.

• Strengthening industrial competitiveness through circular intelligence. DaCapo turns AI into a productivity and sustainability driver, enabling manufacturers to optimise energy, materials, and design processes simultaneously —reducing costs while meeting environmental targets.
• Ensuring interoperability and replicability across sectors. Built on open architectures and standards such as the AAS, DaCapo’s AI models, data structures, and KPIs can be transferred across domains—from logistics and aerospace to consumer goods—accelerating cross-sector adoption.
• Advancing Europe’s twin digital and green transition. DaCapo shows how Digital Twins evolve into circular twins, embedding environmental performance directly into design and operation. This supports the EU Green Deal, the Circular Economy Action Plan, and the Digital Europe Programme by making sustainability measurable, traceable, and actionable.

Together, these elements position DaCapo as a replicable European model for AI-driven circular manufacturing—bridging innovation and policy by turning intelligent digitalisation into tangible industrial competitiveness and environmental value.

Acknowledgements

This work has been carried out under the Horizon Europe project DaCapo (Grant Agreement No. 101091780). Views and opinions expressed are those of the author only and do not necessarily reflect those of the European Union or HADEA. Neither the European Union nor the granting authority can be held responsible for them.

[1] Troncoso, J., Landin, S., Artigues, R., Anttila, E., Maunula, J., Martinez, A.  (2025). Generative AI for Sustainable and Efficient Layout Designs. In: Tareq Z. Ahram and Renate Motschnig (eds) Human Interaction and Emerging Technologies (IHIET 2025). AHFE (2025) International Conference. AHFE Open Access, vol -5. AHFE International, USA.http://doi.org/10.54941/ahfe1006700

[2] Jaziri, S., Martinez, A., Vallhagen, J., Landin, S.  (2025). Learning to Repair Through AI-Driven Geometry Reconstruction for Sustainable Manufacturing. In: Tareq Z. Ahram and Renate Motschnig (eds) Human Interaction and Emerging Technologies (IHIET 2025). AHFE (2025) International Conference. AHFE Open Access, vol -5. AHFE International, USA. http://doi.org/10.54941/ahfe1006744

 [AT1]we use ctgans together with genetic algorithms, and then the 3  best candidates proposed by them are presented. Do you want to mention GAs?