Simulations of dry fabric forming are used to decide whether a carbon-fiber preform will drape over a mould without wrinkles or gaps. They pair a constitutive law, the formula giving stress from deformation, with a mesh of simple elements. These laws are deliberately extreme. Fibers are almost inextensible, so stiffness along the fiber is orders of magnitude above shear stiffness. Recent laws for unidirectional non-crimp fabrics (UD-NCFs, parallel carbon rovings held together by stitching) add a "perpendicular compressive locking" branch: a stiffness wall that switches on when rovings are squeezed together, so that heavily sheared elements cannot collapse. Each of these stiff terms acts as a near-constraint on the deformation. Finite element analysts know what happens when cheap low-order elements meet a near-constraint. If rubber is nearly incompressible, the elements cannot satisfy the volume constraint and still deform, so they stiffen artificially and produce noisy stresses. This is called volumetric locking.
This note proposes treating fiber inextensibility and compressive locking as the fabric analogue of incompressibility. The hypothesis is that some of the error currently blamed on missing physics, notably the force overestimation in 30° off-axis tension of UD-NCF, is actually locking caused by these directional constraints. The proposal has two parts. The first is a diagnosis: tests that separate discretization error from constitutive error. The second is a cure that is cheap enough for explicit forming codes. Locking-free methods replace the volume ratio $J$ with a smoothed version. We do the same for only the stiff directional strain measures, while the soft shear measure stays pointwise.
Core Claim
Investigation of the Membrane Behavior of UD-NCF in Macroscopic Forming Simulations built an 18-parameter membrane model for UD-NCF. It reproduces 45° off-axis tension well, but it "tended to overestimate the required force in the 30° orientation". The authors attributed this to decoupling transverse tension from shear. The same work, and the later invariant-based model in A hyperelastic approach for modeling the membrane behavior in finite element forming simulation of unidirectional non-crimp fabrics (UD-NCF), found that compressive locking is required or elements collapse in the 60° test. These two observations motivate the work. Neither paper varies mesh orientation or density relative to the fibers, so neither can separate constitutive error from discretization error.
Our claim has three testable parts. First, with standard four-node membrane elements and fibers misaligned with the mesh, the combination of near-inextensible fibers and compressive locking over-constrains the element in the same way incompressibility over-constrains a linear tetrahedron. Second, the signature of this would be reaction forces that depend on mesh orientation and fall slowly with refinement, concentrated at bias angles where fiber stretch and roving compaction are both active. Third, an energy-consistent projection of only the stiff directional invariants removes this dependence at roughly the cost of an element average. If the 30° overestimate is mesh-independent, the claim is wrong, and coupling physics is the right explanation. The experiments are built so that this outcome is clearly visible.
Research Program
The method borrows from two lines of locking work. Circumventing volumetric locking in explicit material point methods maps particle Jacobians onto the background grid and back to form a smoothed deformation gradient $\bar{F}$. This removes checkerboard stresses in explicit dynamics at small cost. Advanced discretization techniques for hyperelastic physics-augmented neural networks goes further and treats strain invariants as independent fields in a seven-field Hu-Washizu formulation, a mixed method that solves for strains and stresses alongside displacements. That approach is locking-free but leads to a saddle-point system, which is awkward in explicit codes. We take the middle path. We write the fabric strain energy as a function of invariants, as all the cited fabric laws already do. We then evaluate only the stiff ones through a linear projection operator $\Pi$, such as an element average or a nodally smoothed field:
$$ \Psi(u) = \int_\Omega W\big(I_s(u),\ \Pi I_f(u),\ \Pi I_c(u)\big)\, d\Omega . $$
Here $u$ is the displacement field. $I_s$ is the in-plane shear invariant, which is evaluated pointwise. $I_f$ is the fiber-stretch invariant. $I_c$ is the perpendicular-compaction invariant that drives compressive locking. $W$ is the unchanged constitutive energy, for example the UD-NCF potential of the 2024 hyperelastic model. Because $\Pi$ is linear, internal forces are exact derivatives of $\Psi$. The scheme therefore conserves energy in explicit time stepping, and no separate tangent derivation is needed, which removes a stated obstacle for assumed-strain methods. The design choice that matters is which invariants are projected. Projecting $I_s$ as well should create zero-energy shear modes, the "hourglassing" that Potential and challenges of a solid-shell element for the macroscopic forming simulation of engineering textiles found isotropic stabilization cannot control in textiles. We will check this directly.
The second stage moves the same operator into a three-dimensional setting. Solid-shell elements are thin hexahedra that also resolve thickness, using enhanced assumed strain (extra internal strain modes condensed out per element) and assumed natural strain against thickness and shear locking. 3D solid-shell element for macroscopic composite forming simulation enabling thickness prediction supplies such an element for bidirectional NCF. Rate-dependent 3D forming simulation of thermoplastic composite materials supplies a viscous variant. Both use constitutive laws without tension-shear or compression coupling. Neither has studied how a compressive-locking wall interacts with the enhanced strain modes. In the solid-shell, thickness compaction adds a third stiff invariant. We will measure the condition number of the element-level enhanced-strain condensation matrix as compressive locking engages, with and without projection. The question is whether the constitutive wall brings back the volumetric locking that the enhanced modes were meant to remove. An accurate, robust, and efficient finite element framework with applications to anisotropic, nearly and fully incompressible elasticity found that the standard split into volume-changing and volume-preserving parts misbehaves with strong fiber anisotropy. This constrains our design: we avoid that split and act on the directional invariants directly.
Baselines
The naive reference is the UD-NCF membrane law exactly as published in Investigation of the Membrane Behavior of UD-NCF in Macroscopic Forming Simulations and A hyperelastic approach for modeling the membrane behavior in finite element forming simulation of unidirectional non-crimp fabrics (UD-NCF). It is run as an Abaqus/Explicit user material on standard fully and reduced-integrated membranes, with the published parameters. The second model is the strongest existing method for this question, because it already includes the transverse tension–compression coupling and compressive locking whose numerical side effects we are testing. Neither paper reports force values for off-axis tension, so we found no reported result on Lacuna for UD-NCF off-axis tension force error. We will rerun both models ourselves on the same meshes and time steps, using the published parameters without re-tuning.
The only reported forming-scale figure in the file comes from Modeling multiaxial stress states in forming simulation of woven fabrics. In the fiber-metal-laminate tetrahedron test, its coupled model reaches a maximum shear angle of about 25°, compared with about 18° for the uncoupled Abaqus FABRIC model. Where we use woven-fabric material, that 25° is the number our projected version must match without loss. Beating the baseline has three conditions. Off-axis force error must fall at 30° and 60° on misaligned meshes. The spread between fiber-aligned and misaligned meshes must shrink by at least half. Accuracy at 45° and on aligned meshes must be unchanged.
The key ablation projects $I_f$ alone, $I_c$ alone, both, and all three invariants. This isolates which constraint causes locking and confirms that the soft shear invariant must stay pointwise. The accuracy reference is a mixed formulation in the style of Advanced discretization techniques for hyperelastic physics-augmented neural networks, run implicitly on the coupon tests only. Wall-clock time is reported alongside. For the three-dimensional stage, the baseline is the solid-shell of 3D solid-shell element for macroscopic composite forming simulation enabling thickness prediction with the coupled law placed into it without modification.
Why not just…?
Why not just add the missing tension-shear coupling, as the original authors suggest? This is the main competing explanation, and it may be correct. The test is mesh dependence. Missing physics produces the same error on every mesh. Locking produces error that changes with fiber-to-mesh angle and falls with refinement. We will run the 30° off-axis test on aligned and on 15°- and 30°-misaligned meshes at three densities, with and without a coupling term. If the error is orientation-independent and removed by coupling alone, we report that, and the projection reduces to a robustness tool.
Why not just align the mesh with the fibers? In coupon tests this is possible and might remove the effect entirely. During forming, however, fibers rotate by tens of degrees under shear, so no single mesh stays aligned. Multi-ply laminates also carry plies at different angles on a shared mesh. The hemisphere and tetrahedron runs test whether the advantage survives once fibers reorient.
Why not just soften the compressive locking function or retune its threshold? That moves the error around rather than removing it. A wall soft enough to avoid locking at 30° should allow the element collapse at 60° that the wall exists to prevent. Our test is to calibrate the wall freely against the 30° test and then score it on 60° and the tetrahedron. A single softened wall that passes both would weaken the case for projection, and we would report that.
Why not just use a full mixed formulation? It is the principled cure and serves as our accuracy reference. The anisotropic framework paper reports projection-stabilized elements about 3 times slower and MINI elements, which add extra internal displacement modes, about 7 times slower than penalty formulations on coarse grids. They also need saddle-point solvers that do not suit explicit forming codes. Our projection must come within the mixed method's accuracy at close to the baseline's cost.
Proposed Experiments
On coupons, we simulate off-axis tension at 30°, 45° and 60° and compare against the DIC strain fields and force curves in Membrane behavior of uni- and bidirectional non-crimp fabrics in off-axis-tension tests. We report force error, local shear angle, and the perpendicular-compression field in the gauge region, which the 2024 model was the first to predict at 60°. The sweep covers mesh orientation and density. A pure-membrane patch test with misaligned fibers measures the smallest-eigenvalue spectrum of the stiffness matrix, to show directly whether low-energy modes are being suppressed. As a sanity check that the projection itself works, we replicate the isotropic case and confirm agreement with the reported Prandtl strip-footing load of 5.14 from the $\bar{F}$-MPM paper using our operator on $J$.
At forming scale, we run UD-NCF hemisphere and tetrahedron forming and compare outer contours and shear fields against experiments. We also run the woven fiber-metal-laminate tetrahedron, where the reference is the roughly 25° maximum shear angle. In the solid-shell stage we repeat the hemisphere and double-dome benchmarks and report thickness distributions, condensation-matrix conditioning, and hourglass energy relative to internal energy.
Two outcomes settle the question. The first is whether misaligned-mesh force error at 30° falls with projection while aligned-mesh results stay unchanged. The second is whether projecting $I_s$ triggers hourglassing as predicted. A clean yes on both would show that fabric forming errors have been partly misattributed to constitutive physics. A no on the first would point modeling effort back to constitutive coupling.
References
- Investigation of the Membrane Behavior of UD-NCF in Macroscopic Forming Simulations
- A hyperelastic approach for modeling the membrane behavior in finite element forming simulation of unidirectional non-crimp fabrics (UD-NCF)
- Circumventing volumetric locking in explicit material point methods: A simple, efficient, and general approach
- Advanced discretization techniques for hyperelastic physics-augmented neural networks
- Potential and challenges of a solid-shell element for the macroscopic forming simulation of engineering textiles
- 3D solid-shell element for macroscopic composite forming simulation enabling thickness prediction
- Rate-dependent 3D forming simulation of thermoplastic composite materials using visco-hyperelastic material modeling and 3D hexahedral solid-shell elements
- An accurate, robust, and efficient finite element framework with applications to anisotropic, nearly and fully incompressible elasticity
- Modeling multiaxial stress states in forming simulation of woven fabrics
- Membrane behavior of uni- and bidirectional non-crimp fabrics in off-axis-tension tests