Damage tolerance of filament-wound composite tanks under cryogenic thermal cycling

Abstract

Composite tanks for cryogenic propellants and hydrogen storage must retain strength and leak tightness despite thermal contraction mismatch, matrix microcracking, and repeated pressurisation. We report a fictional experimental study of twelve subscale filament-wound carbon-fibre composite tanks subjected to liquid-nitrogen thermal cycling, helium leak testing, low-energy impact, acoustic-emission monitoring, and residual burst tests. The tanks used a 1.8 L cylindrical geometry with aluminium polar bosses, toughened epoxy matrix, and three wall architectures: baseline carbon/epoxy, carbon/epoxy with a glass-rich barrier interleaf, and carbon/PEEK hybrid liner-overwrap specimens. After 150 cycles between 293 K and 77 K under 0.7 MPa helium bias pressure, baseline tanks showed a median matrix-crack density of 0.42 mm^-1 in hoop-dominated plies and a 13x increase in room-temperature helium leak rate. Barrier-interleaf tanks reduced leak-rate growth by 61% but lost 6.4% more residual burst pressure because the glass-rich interleaf became a delamination path near the dome transition. Hybrid liner-overwrap tanks maintained leak rates below 1.5 x 10^-6 std cm^3 s^-1 after cycling but showed local liner wrinkling at the boss radius. Acoustic-emission event energy during the first ten cryogenic cycles correlated with final leak-rate growth better than event count alone. The results indicate that cryogenic damage tolerance should be assessed by coupled leakage, acoustic, microscopy, and residual-strength metrics rather than by proof pressure alone.

Introduction

Lightweight cryogenic tanks are a persistent bottleneck in launch vehicles, high-altitude aircraft, and hydrogen-energy systems. Carbon-fibre composites offer high specific stiffness and strength, but cryogenic service is severe for polymer-matrix laminates. Cooling from room temperature to liquid-nitrogen or liquid-hydrogen temperature produces matrix-dominated thermal stresses because fibres, matrix, liners, bosses, and barrier layers contract differently. Matrix microcracks can then link into leakage paths even when the composite retains substantial residual strength. Early studies of carbon/epoxy laminates showed that fibre, matrix, and fibre-matrix adhesion all influence cryogenic microcracking [1,2], and micromechanical analyses connected these cracks to ply-level thermal stress and constraint [3,4].

Leakage is the critical distinction between a cryogenic tank and an ordinary composite pressure vessel. A structure may survive proof pressure while failing as a propellant tank because gas permeability rises after thermal cycling, impact, or biaxial strain. Hydrogen and helium permeability studies on polymer-matrix composites, including damaged cryogenic laminates, show that microcracking and impact can increase leak rates by orders of magnitude [5,6,7,8]. Thermoplastic or hybrid liners can improve barrier performance, but liners introduce their own thermal-contraction and boss-seal problems [9,10].

Tank-level validation remains essential because coupons do not reproduce dome curvature, helical-to-hoop ply transitions, boss constraints, winding tension gradients, and proof-pressure history. Progressive-failure analyses and experimental assessments of composite cryogenic storage tanks have shown that pressure window, leakage, and local failure mechanisms must be interpreted together [11,12]. For filament-wound composite pressure vessels, measured fibre properties and winding architecture strongly affect residual strength [13,14]. Acoustic emission is attractive for monitoring this damage progression because it can be acquired during pressurisation and thermal cycling, and prior work has linked AE features to pressure-vessel damage accumulation and burst response [15,16,17].

The present article is a fictional but physically bounded subscale study. We ask which measurements best reveal damage tolerance after cryogenic cycling, and whether early acoustic-emission features can identify tanks likely to develop unacceptable leak rates before a residual burst test.

Tank architecture and fabrication

Twelve tanks were manufactured with a 1.8 L internal volume, 110 mm cylindrical inner radius, 310 mm tangent-to-tangent barrel length, and isotensoid dome transitions terminating in aluminium 7075-T73 polar bosses. The baseline architecture used T700-class carbon fibre and a toughened epoxy resin in a filament-wound sequence of low-angle helical, high-angle helical, and hoop layers. The nominal laminate in the cylindrical section repeated a +-17/+-35/90/90/+-35/+-17 sequence to a 2.15 mm wall thickness. The mean fibre volume fraction from burn-off tests was 58.2 +/- 1.7%.

Three architectures were compared, with four tanks in each group. Architecture A was the baseline carbon/epoxy wall with no dedicated barrier layer. Architecture B inserted a thin glass/epoxy-rich interleaf between the inner helical layers and the first hoop layer, intended to interrupt transverse crack linkage. Architecture C used a 0.25 mm thermoplastic PEEK liner thermoformed over a removable mandrel, followed by carbon/epoxy overwrap with reduced hoop thickness to keep mass within 4% of architecture A. The liner was mechanically captured under the boss flange but was not adhesively bonded along the full dome. This detail is important because local liner slip and wrinkling are plausible failure modes in hybrid tanks.

All tanks were cured in a rotating oven with a two-stage ramp to 135 deg C and a 2 h dwell. Boss seals were installed after cure using indium-coated stainless crush washers. Initial proof tests were conducted at room temperature with dry nitrogen to 2.4 MPa, equal to 1.5 times the nominal maximum expected operating pressure in this subscale programme. No tank leaked above 4 x 10^-7 std cm^3 s^-1 before cryogenic cycling.

Cryogenic cycling and inspection protocol

The cryogenic cycle used liquid nitrogen immersion rather than liquid hydrogen for safety and repeatability. Each cycle cooled the tank from 293 K to 77 K over 18 min, held it at 77 K for 12 min, and warmed it to 293 K in dry nitrogen over 55-75 min. Tanks were biased internally with 0.7 MPa helium during cooling to open incipient leak paths without approaching proof stress. Pressure, boss temperature, barrel outer-surface temperature, and bath temperature were recorded at 2 Hz. Each tank experienced 150 cycles unless it exceeded the programme leak limit of 1 x 10^-4 std cm^3 s^-1, which none did.

After 50 cycles, two tanks from each architecture received a 12 J low-velocity impact at the barrel midspan using a hemispherical tup. The impact energy was selected to produce barely visible external damage but a measurable acoustic response. The impacted tanks then completed the remaining 100 thermal cycles. Helium leak testing was performed at 293 K and at 77 K after cycles 0, 10, 50, 100, and 150. Room-temperature leak tests used a mass-spectrometer leak detector and vacuum shroud; 77 K tests used a calibrated accumulation chamber because direct shrouding around the liquid-nitrogen bath was impractical.

Acoustic emission was recorded using four wideband piezoelectric sensors coupled to the tank exterior with cryogenic-compatible silicone grease. Sensors were mounted near the barrel midspan and the two dome transitions. A 45 dB threshold was used during pressure holds, and a floating threshold was used during cooling to avoid counting boiling noise. Waveforms were sampled at 2 MHz. Events were classified using peak frequency, duration, rise time, and energy; the classes were interpreted only as phenomenological clusters, not as unique crack mechanisms. This restraint follows the broader pressure-vessel AE literature, where event energy and load history are often more robust indicators than a single event count [15,16].

Damage characterisation and residual strength

X-ray computed tomography was performed on all tanks after cycling, with higher-resolution scans over the dome-to-barrel transition and impact region. The voxel size was 32 micrometres for full-tank scans and 9 micrometres for local scans. Selected tanks were sectioned after burst testing for optical microscopy and dye-penetrant-assisted crack mapping. Matrix-crack density was measured in hoop-dominated and helical plies over three polished sections per tank. Delamination area was estimated from CT segmentation and confirmed by sectioning where possible.

Residual strength was measured by hydrostatic burst testing at room temperature using water pressurisation at 0.08 MPa s^-1. Hydrostatic burst does not reproduce cryogenic gas service, but it reduces stored energy and allows comparison of residual structural margin after the same thermal history. Burst pressure, leak rate before burst, AE cumulative energy, and visible failure location were recorded. A finite-element model with temperature-dependent orthotropic lamina properties was used to estimate ply-level thermal stress and identify regions where thermal and pressure stresses should combine most strongly. The model was used for interpretation rather than design allowables.

The principal damage-tolerance metrics were leak-rate growth, residual burst-pressure retention, microcrack density, delamination area, and early-cycle AE energy. Proof pressure alone was deliberately not used as the pass/fail metric because cryogenic tank failure can be leakage-dominated rather than strength-dominated [5,7,9].

Results

Baseline architecture A developed visible matrix cracking during the first 50 cycles, after which crack density increased more slowly. After 150 cycles, the median crack density in hoop-dominated plies was 0.42 mm^-1, compared with 0.18 mm^-1 in helical plies. Room-temperature helium leak rate increased from below 4 x 10^-7 to a median of 5.2 x 10^-6 std cm^3 s^-1. The two impacted baseline tanks reached 1.7 x 10^-5 and 2.4 x 10^-5 std cm^3 s^-1, respectively, showing that barely visible impact damage can interact with thermal microcracking even when residual burst strength remains high.

Architecture B reduced leak-rate growth but changed the damage mode. The glass-rich interleaf interrupted through-thickness crack linkage in the barrel, reducing median post-cycle leak rate to 2.0 x 10^-6 std cm^3 s^-1. However, CT showed crescent-shaped delaminations near the dome transition in three of four tanks. Residual burst pressure was 6.4% lower than the baseline group even though leakage was lower. Sectioning showed that the interleaf concentrated shear near the transition from helical to hoop-dominated winding. The barrier layer therefore improved one requirement while degrading another.

Architecture C had the best leakage performance. All four hybrid liner-overwrap tanks remained below 1.5 x 10^-6 std cm^3 s^-1 after 150 cycles, including one impacted specimen. The penalty was local liner instability. Two tanks showed liner wrinkles at the boss radius after cycle 100, and one had a small liner debonded region visible in CT but not connected to external leakage. Residual burst pressure was similar to the baseline group, 3.82 +/- 0.11 MPa compared with 3.88 +/- 0.14 MPa for architecture A. The result is encouraging but not definitive, because a 1.8 L subscale tank cannot reproduce full-scale liner buckling and slosh-induced deformation.

Acoustic emission provided early warning of later leakage better than simple event count. Total AE event count during the first ten cycles had R^2 = 0.31 with final leak-rate growth, while cumulative event energy above the 90th percentile had R^2 = 0.72. High-energy events concentrated at the dome transition for architecture B and near the impact site for impacted architecture A tanks. The AE clusters with peak frequency above 180 kHz were associated with matrix-crack growth in sectioned specimens, whereas lower-frequency long-duration events were more often associated with interleaf delamination or liner slip. These associations are plausible but not unique, and we treat them as diagnostic correlations rather than mechanism labels.

Discussion

The experiments reinforce a central difficulty of cryogenic composite tanks: structural margin and leak tightness do not rank architectures in the same order. Architecture B would look attractive if judged by leak rate alone, but its interleaf introduced a delamination-prone interface that reduced residual burst pressure. Architecture C controlled leakage most effectively, but liner wrinkles at the boss radius create a scaling concern. Architecture A retained strength but allowed leak-rate growth through linked matrix cracks. These trade-offs agree with coupon-level studies showing that cryogenic microcracking and permeability are sensitive to matrix chemistry, fibre-matrix adhesion, impact history, and barrier architecture [1,2,7,8,9].

The early-cycle AE result is practically useful. A tank that emits a small number of high-energy events during the first few cryogenic cycles is not necessarily close to burst, but it may be developing connected leakage paths. This distinction matters for acceptance testing. A conventional proof test can miss leakage-dominated damage if the tank retains burst margin. Combining AE with helium leak testing after thermal cycling provides a better picture of damage tolerance, especially for subcritical matrix cracking. The recent structural-health-monitoring work on composite overwrapped pressure vessels points in the same direction: processing history, sensor response, and pressure history should be read together rather than as isolated pass/fail signals [17].

The liquid-nitrogen test should not be over-interpreted as a full liquid-hydrogen qualification. Liquid hydrogen has lower temperature, smaller molecule size, different safety constraints, and different permeation behaviour. Helium leak testing is a conservative screening tool in some respects but not a complete substitute for hydrogen service. The aluminium bosses, indium seals, and small tank diameter also simplify full-scale problems such as boss heat leak, liner buckling, and manufacturing defects over large areas. The data should therefore be read as comparative damage-tolerance evidence for architectures, not as flight qualification.

Several design implications follow. Barrier layers should be tapered or locally interrupted near dome transitions to avoid creating delamination planes. Thermoplastic liners need boss-radius design rules and post-cycle inspection, not only room-temperature proof testing. AE acceptance criteria should weight event energy and location, not only event count. Finally, residual burst pressure should be reported together with leak-rate evolution and microscopy, because cryogenic tank reliability is a coupled structural and permeability problem.

Conclusion

We tested twelve subscale filament-wound composite tanks under 150 liquid-nitrogen thermal cycles, helium leak checks, acoustic-emission monitoring, low-energy impact, CT inspection, and residual burst testing. Baseline carbon/epoxy tanks retained burst strength but developed connected matrix cracks and measurable leak-rate growth. Glass-rich barrier interleaves reduced leakage but promoted delamination near dome transitions. Hybrid PEEK liner-overwrap tanks provided the best leak-tightness but showed local liner wrinkling at the boss radius.

The main conclusion is that cryogenic damage tolerance cannot be demonstrated by proof pressure alone. Leak-rate growth, early-cycle AE energy, microcrack density, delamination area, and residual burst pressure are complementary metrics. Future work should extend the programme to liquid-hydrogen-compatible testing, larger tank diameters, bidirectional mechanical loading, and repeated fill-drain cycles with realistic thermal gradients.

Data and code availability

Cryogenic cycle histories, leak-rate measurements, AE waveform features, CT-derived damage maps, polished-section crack counts, burst-pressure traces, and finite-element input files are included in the supplementary archive. AE clustering scripts were tested with Python 3.10, NumPy 1.24, SciPy 1.10, and scikit-learn 1.2. Raw waveform files are provided for representative cycles and all burst tests; full waveform archives are available as compressed HDF5 files because of size.