Towards AI-Augmented Engineering: what you missed at the Transvalor Tech Days!

Written by Lisa Mas | Jul 8, 2025

This ongoing revolution is driven by our R&D teams and illustrated through concrete user cases. Discover the distinct AI development tracks aimed at streamlining your simulations.

From Experimentation to Artificial Intelligence: The Four Pillars of Modeling

There are four key pillars supporting the use of AI in digital simulation:

Experimentation, the traditional foundation, though costly;
Analytical models, fast but simplistic;
FEM modeling, currently the standard for high-fidelity results;
AI/ML, the new pillar enabling unprecedented prediction and optimization capabilities.

Why Integrate Machine Learning?

Two main user cases:

Direct prediction: “Will my component fail?”
Inverse optimization: “What configuration would prevent this failure?”

With ML, we move from classic simulation to an adaptive learning process capable of exploring broader solution spaces—faster and more efficiently.

Complex Materials & Hybrid Approaches

A M.L. Approach to Assist Process Modeling:

Advanced material modeling now enables you to:

Create models from experimental databases;
Combine complex phenomena across multiple scales.

This modeling streamlines the transition from raw data obtained in controlled lab environments to evaluations of new industrial configurations.

Example: Predicting hardness properties in an aluminum alloy family
We demonstrated the accuracy of an Artificial Neural Network (ANN) model trained using a physical model developed from experimental data. This was part of a collaborative project between Transvalor and the SiMaP laboratory in Grenoble.
The goal was to predict the evolution of yield strength in 7XXX series aluminum alloys under various heat treatment and pre-deformation conditions. Once trained, the ANN model enables accurate mechanical behavior assessments on more complex geometries, particularly in FEM simulations at the part scale.

GNN + FEM: Accelerating Simulation with Graph Neural Networks

Graph Neural Networks (GNNs) can handle complex and irregular meshes—where traditional networks fall short. Combined with FEM, they form a hybrid model that:

Integrates physical knowledge (via PINNs – Physics-Informed Neural Networks),
Adapts to diverse geometries,
Dramatically reduces computation time.

Case Study: how works Geometric and Topological Extrapolation?

A major strength of GNN approaches lies in their ability to learn "local" phenomena and propagate them spatially across a geometric domain. This allows training models on various geometries and extrapolating to new ones.

True (reference) solution

Prediction on unseen geometry

This capability also enables GNNs to make predictions across different FEM mesh refinement levels.

Read the full project:

What’s Next? Transvalor’s AI Assistants

We are also developing an in-house Large Language Model (LLM) assistant, designed to:

Guide users through their workflows,
Automate certain setup tasks,
Make simulation more accessible—even without deep technical expertise.

Conclusion: Simulation Enters a New Era

The 2025 Tech Days were an opportunity to share this ambitious vision: intelligent, hybrid, and more efficient simulation tools than ever before. At Transvalor, we believe in the future of augmented engineering—where AI doesn’t replace the human, but empowers them.

View full post