AI Surrogate Model Brings Real-Time Simulation to Composite Manufacturing

Published date:16/04/2026 | Last updated date:16/04/2026

Researchers from IMDEA Materials Institute and the Technical University of Madrid (UPM) have developed a deep learning-based surrogate model that could significantly accelerate simulation in composite manufacturing, potentially enabling real-time process control in industrial environments.

The study focuses on liquid composite molding (LCM), a widely used process in aerospace and advanced composites where precise control of resin flow is critical. The research introduces a new computational approach capable of delivering simulation results in milliseconds, compared to the much longer runtimes required by traditional methods.

Breaking the computational bottleneck in LCM

LCM simulations are essential for predicting how resin flows through fiber preforms, helping manufacturers avoid defects such as:

• void formation
• dry spots
• incomplete impregnation

However, conventional simulation tools are computationally expensive, often limiting their use to offline analysis rather than real-time decision-making.

The new approach addresses this limitation by using a deep surrogate model, which replaces traditional physics-based solvers with a trained neural network capable of rapidly predicting flow behavior.

According to the research team, this enables:

• 4–5 orders of magnitude speed improvement
• strong agreement with high-fidelity simulations
• consistency with experimental validation

Designed for real-world industrial complexity

One of the key technical challenges in composite manufacturing simulation is handling complex geometries and unstructured meshes, which are common in real production environments.

To solve this, the researchers developed:

• a multi-branched encoder–decoder architecture
• a grid mapping technique for unstructured 3D domains

This allows the model to break down complex structures—such as T-shaped stringers—into manageable regions while maintaining continuity across interfaces.

The result is a system that combines:

• computational efficiency
• geometric adaptability
• predictive accuracy

These three factors are rarely achieved together in existing AI-based simulation tools.

From simulation to real-time manufacturing control

The most important implication of this research is not just faster simulation—it is the shift toward real-time, data-driven manufacturing.

With millisecond-level prediction capability, the model opens the door to:

• digital twins of composite processes
• adaptive process control during production
• immediate detection and correction of defects

In practical terms, this could allow manufacturers to adjust parameters such as:

• resin injection speed
• pressure distribution
• mold temperature

—while the process is still running, rather than after defects have already occurred.

Strategic impact for composite production

For industries like aerospace, automotive, and energy, where composite parts must meet strict quality and performance standards, this development could directly influence:

• production yield
• material waste reduction
• cycle time optimization

More importantly, it aligns with a broader industry shift toward Industry 4.0, where manufacturing systems are expected to be:

• connected
• adaptive
• predictive

The integration of AI-driven surrogate models into LCM processes represents a critical step toward that goal.

🔒 Content Transparency & Editorial Integrity

This article is developed based on real engineering experience, machine testing data, and practical production knowledge from Jota Machinery’s work in advanced composite manufacturing.

All technical explanations—including material structure, processing methods, and performance characteristics—are reviewed and verified by our engineering team to ensure accuracy and real-world relevance.

To improve clarity and structure, AI-assisted tools may have been used during content organization and language refinement. However:

• All key technical insights originate from first-hand industrial experience
• All data and claims are manually reviewed and validated
• The content is created with the primary goal of educating engineers, manufacturers, and buyers

We do not publish content solely for search ranking purposes. Every article is designed to provide practical, experience-based value to professionals in the composite materials industry.

Editor’s Note

This research highlights a turning point in composite manufacturing. For years, simulation has been powerful but slow—useful for design, but disconnected from real production.

What changes here is speed.

When simulation moves from hours to milliseconds, it stops being a planning tool and becomes a control tool.

That shift has deep implications:

• Defects can be prevented, not just analyzed
• Processes can adapt in real time, not after failure
• Manufacturing becomes predictive, not reactive

However, the challenge ahead is integration. Industrial environments demand robustness, reliability, and seamless connection with existing equipment.

If these models can move from research environments into factory floors, they could redefine how composite parts are manufactured—transforming simulation from a support function into the core of production itself.