Context
AGC wanted to evaluate an alternative to Abaqus for simulating the de-icing of automotive glass using the Joule effect.
The reference case combined thin glass, a conductive layer, busbars and cables, with a transient thermal analysis coupled with an electrical analysis.
Simvia’s approach
- Development of a 3D and 1D monolithic thermoelectric model directly within code_aster.
- Step-by-step verification of the physical contributions, followed by a comparison with Abaqus using a simple test case and the complete industrial windscreen.
- Delivery of a ready-to-use Docker environment, test cases and data input for the AGC model.
Results
- 73 verification and non-regression test cases
- 0.61% relative error on a simple case
- 0.73°C mean absolute error on the industrial case
- An initial operational version, verified and ready for use in the industrial use cases targeted by AGC.
Technical view
Technology stack
- Core: a fork of code_aster 17.4, with Fortran routines and Python libraries.
- Solver: THER_NON_LINE with MUMPS, with temperature and electric potential solved jointly.
- Models: 3D_THER_ELEC for volumes and 1D_THER_ELEC for cables with variable cross-sections.
- Interfaces: LIAISON_MAIL and ECHANGE_PAROI extended to non-conforming meshes.
Constraints
- No thermoelectric shell elements: volumetric extrusion of the glass in 5 to 6 layers.
- 1D cables connected to busbars and to 3D glazing via non-conforming meshes.
- Field-dependent materials and transient calculation over 1,800 s.
How it went
- Definition and verification of the mathematical model based on Abaqus and Ansys references and the literature.
- Prototyping, systematic code review and manual rewriting of critical sections.
- Analytical validation of the element contributions, followed by 73 automated test cases.
- Final comparison with Abaqus on a simple case and on the AGC industrial model.
Security, performance, quality
- 1D–3D interfaces: distribution of the injection over a representative area of the cable cross-section.
- Diffusion through the thickness: 5–6 layers required to reproduce the Abaqus overheating level.
- Symmetry: verification of currents and potentials, followed by testing on a mirrored mesh to isolate the numerical artefact.
- Time scheme: alignment of the θ parameter and meshing refinement to ensure a consistent comparison.
