Thin-Ply Composite Analysis

Conventional-thickness Double-Double (DD) laminates carry a well-documented unnotched tensile penalty relative to stiffness-equivalent quasi-isotropic (QI) laminates, usually attributed to elevated free-edge stresses, earlier matrix cracking and delamination.

This project asks whether that penalty is intrinsic to the DD architecture, or whether it is only a consequence of matrix- and interface-dominated damage mechanisms that thin plies suppress. It asks the same question of the classical strength penalty associated with ply blocking.

Each laminate is 32 plies at 0.03 mm, giving a 0.96 mm laminate. The blocked stacks contain the same proportion of each orientation as their dispersed counterparts, so their extensional stiffness (A) matrices are identical. That equivalence is the premise the whole comparison rests on, and it is asserted by an automated test rather than assumed.

Panels were laid up by hand from TC33/K51 spread-tow thin-ply prepreg, vacuum-bagged and autoclave-cured, then cut into untabbed coupons.

The non-contact video gauge matters on a 0.96 mm coupon. A bonded extensometer or strain gauge would introduce exactly the local stiffening and stress concentration the experiment is trying to avoid. Coupons were run untabbed after preliminary tabbed tests failed prematurely at the tab ends.

The FE work is the numerical half of the argument. If DD really carries an intrinsic penalty, it should show up as a more severe free-edge stress state than QI at matched in-plane stiffness.

Every one of the 32 plies is meshed discretely through the thickness. The free-edge problem is a ply-scale effect, so smearing the laminate into an equivalent shell would delete the very thing being measured.

Two results come out of this. QI and DD develop genuinely different free-edge stress states despite matched in-plane stiffness, so the architectures are not interchangeable at the ply scale even when their ABD matrices say they are. Blocking also raises the local magnitudes in both families, which is the conventional explanation for the blocking strength penalty.

Feeding these fields into the O'Brien energy release rate gives per-interface values of 45 to 254 J/m², sitting within or below the expected critical range for this material system (Gc ≈ 200 to 500 J/m²) for three of the four laminates. The free-edge stresses are real and architecture-dependent, but mostly not large enough to drive delamination first. That is the crux of the whole result.

All four laminates failed at similar global strains (1.41 to 1.53 %) and stresses (496 to 538 MPa). Blocked DD gave the highest mean, but the question worth answering is which of the differences are real.

Classical laminate theory converts the laminate-level measurements into something comparable against the material's own fibre allowable. Both architectures underperform their references and DD underperforms by more, which taken alone looks like the classic DD penalty. It is not. The shortfall tracks the grip-failure caveat below, and the architectures are statistically indistinguishable when compared against each other rather than against literature values.

That pair is what decides the question. Raw pairwise tests throw up differences, but once Holm correction is applied for the number of comparisons actually being made, almost all of them disappear. The headline QI-versus-DD difference does not survive it.

SEM of the fracture surfaces makes the mechanism visible directly. QI shows matrix-dominated separation localised at the 90° ply, so a single weak orientation is doing the failing. DD shows fibre fracture across all four orientations, with damage distributed rather than concentrated at one ply.

That is what the thin-ply suppression hypothesis predicts. With no weak matrix-dominated ply to fail first, the DD architecture loses the mechanism that conventionally penalises it. The statistics say the penalty is gone, and the fractography says why.

The reported numbers come from the MATLAB pipeline, which covers ABD construction, Hashin first-ply failure across all four modes, O'Brien energy release rates, and the Welch and Holm statistics. Because MATLAB needs a licence, the central calculation is also implemented in Python and runs against a clearly labelled synthetic sample dataset with no licensed software at all.

A 65-test suite asserts the properties the analysis depends on rather than just that the code runs. It checks that a symmetric laminate really produces a zero B matrix, that the A matrix is genuinely unchanged by blocking (the stiffness-matching premise of the whole comparison), that a unidirectional laminate recovers its engineering constants exactly, that Hashin returns unity at each allowable, and that the Holm implementation matches hand-computed adjusted p-values.