A matrix-free inexact Gauss–Newton solver with automatic differentiation for photometric DIC: Application to a self-occluded torsion test
Abstract
Digital Image Correlation (DIC) has established itself as an essential tool in experimental solid mechanics. However, since it relies on the assumption of grey-level conservation, its performance can be compromised in specific experimental configurations that violate this hypothesis (artifacts, occlusion, scale changes, large deformations, etc.). Photometric DIC replaces this assumption with a physically based, differentiable rendering model, thereby extending full-field measurement to these complex configurations. Its deployment has so far been limited by the cost of the Jacobian of the rendered image. A matrix-free inexact Gauss-Newton (GN) solver is introduced. The linearised GN subproblems are solved by a truncated Krylov solver whose matrix-vector products are computed by automatic differentiation, at a cost of two forward model evaluations per inner iteration, independent of mesh and image sizes. A diagonal preconditioner, assembled at the cost of three renderings, makes as few as two inner iterations sufficient. The solver is characterised on a synthetic test, then applied to a torsion experiment performed on a polymer cylinder twisted up to 418° and monitored by a single camera, a configuration that consequently exhibits severe self-occlusion.
Related
- Derivative Computation in PINNs: Automatic Differentiation, Finite Differences and Beyond
- Locating critical solutions in numerical relativity using automatic differentiation
- Optimal Control of Periodic Nonequilibrium Mechanochemical Systems via Automatic Differentiation
- Variational Multi-Gaussian Phase-Space Bosonic Dynamics via Automatic Differentiation
- Truncated automatic sparse differentiation for machine learning interatomic potentials
BibTeX
@article{person2026matrixfree,
title = {{A matrix-free inexact Gauss–Newton solver with automatic differentiation for photometric DIC: Application to a self-occluded torsion test}},
author = {Lucas Person and Jean-Charles Passieux and Jean-Noël Périé and John-Eric Dufour},
journal = {Computer Methods in Applied Mechanics and Engineering},
year = {2026},
month = oct,
doi = {10.1016/j.cma.2026.119494},
url = {https://hal.science/hal-05844210v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A matrix-free inexact Gauss–Newton solver with automatic differentiation for photometric DIC: Application to a self-occluded torsion test
AU - Lucas Person
AU - Jean-Charles Passieux
AU - Jean-Noël Périé
AU - John-Eric Dufour
PY - 2026
DA - 2026/10/24
JO - Computer Methods in Applied Mechanics and Engineering
DO - 10.1016/j.cma.2026.119494
UR - https://hal.science/hal-05844210v1
AB - Digital Image Correlation (DIC) has established itself as an essential tool in experimental solid mechanics. However, since it relies on the assumption of grey-level conservation, its performance can be compromised in specific experimental configurations that violate this hypothesis (artifacts, occlusion, scale changes, large deformations, etc.). Photometric DIC replaces this assumption with a physically based, differentiable rendering model, thereby extending full-field measurement to these complex configurations. Its deployment has so far been limited by the cost of the Jacobian of the rendered image. A matrix-free inexact Gauss-Newton (GN) solver is introduced. The linearised GN subproblems are solved by a truncated Krylov solver whose matrix-vector products are computed by automatic differentiation, at a cost of two forward model evaluations per inner iteration, independent of mesh and image sizes. A diagonal preconditioner, assembled at the cost of three renderings, makes as few as two inner iterations sufficient. The solver is characterised on a synthetic test, then applied to a torsion experiment performed on a polymer cylinder twisted up to 418° and monitored by a single camera, a configuration that consequently exhibits severe self-occlusion.
SN - 0045-7825
SN - 1879-2138
ER -