Case study

MaChAwAI

We're reinventing material testing for additive manufacturing

Our Partners: MaCh3D and 3DPR

The Project

MaChAwAI is an innovative project that combines artificial intelligence and Material Testing for additive manufacturing, developed in collaboration with MaCh3D and 3DPR.

The goal is to reinvent material analysis in 3D printing, making it simpler, faster and cheaper to verify the physical characteristics of materials.

The project was supported by the European Horizon 2020 program through the open call KYKLOS 4.0 #2 (KYKLOS 4.0 project, Grant Agreement n. 872570).

Challenges

The Challenges in Material Testing

High Costs

Traditional Material Testing machinery is expensive, bulky, and often inaccessible to SMEs.

Significant Complexity

Only specialized operators can correctly perform the tests, requiring highly qualified personnel.

Long Time

Tests can be lengthy to prepare and conduct, slowing development and production processes.

The MaCh3D device, while reducing certain cost and complexity aspects compared to standards, still needs an additional tool to obtain reliable data — such as an extensometer (or "distance meter") — with consequent increases in costs and difficulty of use.

Goals

Project Goal

"Build a system that delivers precise measurements from the MaCh3D device, avoiding the use of additional instrumentation."

Faster

No need to position the extensometer.

Simpler

By eliminating additional tools, the procedure becomes streamlined.

Cheaper

Costs related to extensometers or other equipment are saved.

Equally Accurate

Even though we remove a device traditionally considered essential, high accuracy is maintained.

AI model

Our Solution

We developed an innovative Deep Learning model that makes MaCh3D measurements as accurate as those obtained with an extensometer.

Input

Device that needs an extensometer to obtain reliable data

Process

Deep Learning model that analyzes and corrects raw data

Output

Data with accuracy equivalent to that obtained with the extensometer

The neural network acts as a "transformer" of the data generated by MaCh3D, returning a curve that's much closer to reality.

Roadmap

Implementation Phases

1. Preliminary Study

Analysis of scientific literature and selection of existing public datasets on Material Testing, to build a solid background for deep learning techniques.

2. Data Collection

Leveraging the many instrumented tests by MaCh3D and 3DPR on various materials and printing technologies to build a custom dataset representative of multiple scenarios.

3. Ideation & Development

Based on the collected data, we conceived and developed dozens of Deep Learning models, experimenting with different architectures to find the ideal tailored solution.

4. Validation

The best model was cross-validated by the engineering teams of MaCh3D and 3DPR, using data never "seen" by the model to evaluate its performance on new scenarios.

5. Iteration & Optimization

With new measurements and printing samples, the AI continued to learn and further reduce measurement error, creating a virtuous loop: the more tests run, the more robust and accurate the models become.

Results