Skip to main content

ARCHITECTURE

Technology

A mixed-reality module mounted on the helmet, a computer-vision pipeline and feedback displayed in the welder's field of view.

TECHNICAL CHAIN

From capture to consolidation

The full chain, from the welding scene to the recorded gesture. Each link is described as currently designed: some are validated in a virtual environment, others still have to be proven in real conditions.

  1. 01

    Capture

    The helmet-mounted module films the work zone and the torch position during the pass, without changing the workstation or the process.

  2. 02

    Pre-processing

    Luminance range correction, image stabilisation and filtering of the noise caused by fumes and spatter, before any analysis.

  3. 03

    Computer vision

    Detection of the torch, the workpiece and the weld pool, then reconstruction of the trajectory in the workpiece frame of reference.

  4. 04

    Analysis

    Computation of gesture parameters — travel speed, arc height, angles, regularity — and comparison with the ranges expected for the procedure.

  5. 05

    Feedback

    Visual feedback displayed in the field of view during the pass, kept minimal so it never pulls attention away from the pool.

  6. 06

    Consolidation

    Session recording, per-learner history and a workshop-level view for the instructor.

How these steps are split between on-board and off-board computing is still an open design decision.

Close-up of the optical module built into the helmet: two sensors and light strip

The helmet module

The module mounts on the welder's helmet without changing the station or the torch.

Precision industrial optical module, close-up in cold lighting

The optical assembly

Cameras capture the weld zone and its immediate surroundings, despite the arc's extreme light dynamics.

Representation of a gesture path with a detected variation

Trajectory analysis

Algorithms reconstruct the torch trajectory and compare the performed gesture with the expected one.

DATA AND COMPLIANCE

The principles we apply

These principles describe our design intent. They are not a contractual commitment: exact terms will be formalised with each partner organisation.

Educational purpose
Gesture data serves training and learner progress. It is not designed as a tool for individual monitoring of employees.
Minimisation
We aim to collect only what the gesture analysis requires, for as long as needed, and to limit directly identifying data.
Hosting
The target is hosting within the European Union, with restricted and logged access.
Models and annotated datasets
Models and annotated datasets produced within the project remain the intellectual property of ARC FRANCE, unless a specific agreement is signed with a partner.

COLLABORATION

We are looking for TRAINING partners

Partner centres

Training centres and workshops willing to host supervised trials in real conditions.

Propose a collaboration