Protocol-dependent aging responses of polyurethane-bound pavement composites under thermo-oxidative, hygrothermal, and UV exposure

1 Introduction

Polyurethane (PU)-bound aggregate composites are increasingly studied as alternatives to conventional asphalt mixtures because the cured polymer phase can provide high rutting resistance, favorable low-temperature deformability, and reduced temperature sensitivity (Cong et al., 2019; Li et al., 2021; Xu L. et al., 2024). Unlike polyurethane-modified asphalt, in which polyurethane acts within a bituminous matrix, the present material uses cured polyurethane as the primary binder between mineral particles; deterioration of the polymer phase can therefore directly affect aggregate-to-aggregate load transfer. Laboratory studies have examined moisture-curing behavior and mixture performance (Xiang et al., 2021; Moon et al., 2023; Li et al., 2023; Yang et al., 2021), dense- and open-graded dynamic response (Zhao et al., 2023a,b; Zhao et al., 2026a; Zhuang et al., 2023), pavement structural response (Pang et al., 2026; Chen et al., 2025), temperature-regulating mixtures (Jiao et al., 2023), and polyurethane-based steel-bridge-deck systems (Xu Y. et al., 2024; Chen et al., 2024; Niu et al., 2022; Gao et al., 2022). Construction-stage energy and emissions have also been assessed for polyurethane pavement systems (Cong et al., 2020). These studies establish the engineering potential of PU-bound pavement materials, but long-term environmental durability remains a key barrier to broader application.

Environmental aging of polyurethane may involve thermal oxidation, photo-oxidation, hydrolysis, hydrogen-bond rearrangement, and changes in network organization. Prior studies have reported molecular-weight evolution, tensile deterioration, surface or interfacial changes, and aging-dependent performance in neat polyurethane, polyurethane coatings, and polyurethane-modified asphalt (Zhang et al., 2024; Li K. et al., 2025; Wang et al., 2025; Ji et al., 2023; Godlewska et al., 2024; Potolinca et al., 2026; Yang et al., 2025a,b; Albergamo et al., 2023; Mayer-Trzaskowska et al., 2024; Liu et al., 2024). However, a dense PU-bound aggregate composite responds not only through changes in the polymer phase but also through binder continuity, aggregate interlock, and aggregate–polyurethane stress transfer. A severity ranking obtained from a neat-binder indicator may therefore not be reproduced by every mixture-scale response.

Two protocol-comparability issues are especially important for accelerated durability assessment. First, thermo-oxidative and hygrothermal exposures should be compared at a common elapsed duration before a difference in measured severity is attributed to environmental condition rather than exposure time. Second, continuous and intermittent ultraviolet schedules may accumulate different irradiation times even when their reported laboratory durations are identical. Water can further alter aging responses when wetting is coupled with irradiation (Qian et al., 2021). Thus, a controlled comparison must distinguish a whole-protocol contrast from a comparison in which UV-on time, UV-stage temperature, and cycle length are matched.

Cross-level interpretation also requires complementary indicators. Gel permeation chromatography (GPC) characterizes the filtered tetrahydrofuran (THF)-soluble polymer population, whereas the THF-insoluble fraction provides an operational measure of material that remains outside that soluble population. Binder tensile strength and elongation at break quantify load-bearing capacity and deformability, while dynamic stability, low-temperature flexural strain, splitting strength, and Cantabro loss represent distinct mixture-scale failure modes. The present study therefore evaluates a moisture-curing MDI-based polyether polyurethane binder and a dense PC-13 aggregate composite under thermo-oxidative, hygrothermal, continuous UV, intermittent UV, and intermittent UV–immersion aging. Unlike studies in which polyurethane functions as an asphalt modifier, the cured polyurethane phase in the present composite acts as the primary binder and directly contributes to aggregate-to-aggregate load transfer. Accordingly, the study focuses on three protocol-comparability questions: (i) whether thermo-oxidative and hygrothermal severity rankings remain consistent when exposure duration is matched; (ii) whether molecular-, binder-, and mixture-level indicators provide consistent rankings of aging severity; and (iii) which mixture indicators discriminate the tested UV protocols when differences in accumulated UV-on time are explicitly considered. The resulting experimental design links protocol control with complementary material responses rather than relying on a single indicator to represent overall aging severity. The cross-level assessment concept is summarized in Figure 1.

2 Materials and methods

2.1 Raw materials

The binder was a commercial single-component, moisture-curing MDI-based polyether polyurethane (6170F). Its nominal properties were an isocyanate (NCO) content of 13.8%, a viscosity of approximately 750 mPa·s at 25 °C, a tack-free time of 2.5 h, and a hard-dry time of 4.3 h at 25 °C/80% relative humidity (RH). Basalt was used as the coarse aggregate, while limestone was used for the fine aggregate and mineral filler.

2.2 Mixture design and specimen preparation

A dense PC-13 gradation with a nominal maximum aggregate size of 13.2 mm was used; the adopted sieve passing percentages are reported in Supplementary Table SA1. The binder-to-aggregate ratio for all PC-13 aging specimens was 4.7% by mass. Aggregates were oven-dried at 110 °C for at least 3 h and cooled to approximately 30 °C. The conditioned aggregates were dry-mixed for 60 s; the polyurethane binder was then added, followed by wet mixing for 85 s. Marshall specimens were compacted symmetrically on both faces in two stages while retained in the mold. After leveling, the first face received 45 blows and the specimen was turned for 45 blows on the second face. The molded specimen was then conditioned at 50 °C/50% RH for 2 h, after which 30 additional blows were applied to each face. Each face therefore received 75 blows (150 blows per specimen). The staged procedure maintained the conventional per-face compaction effort while limiting the recovery of air voids associated with curing expansion. The molded Marshall specimens were subsequently cured at 25 °C/80% RH to a total age of 7 d before aging or performance testing.

2.3 Accelerated aging protocols

The experimental design first separated exposure-duration effects from environmental-protocol effects under static aging and then compared three UV protocols (Figure 2; Table 1). Hygrothermal aging was conducted at 85 °C and 80% RH for 20 h. Thermo-oxidative aging was conducted in a dry forced-draft oven at 85 °C for 16 h, and an additional 20 h group provided the equal-duration comparison with the hygrothermal endpoint. The static aging temperature and exposure window were selected from the established experimental program to produce measurable deterioration while retaining sufficient residual integrity for subsequent mechanical characterization. The TO-20 h group was introduced specifically to provide a duration-matched comparison with HT-20 h. UV exposure was conducted in a fluorescent UV aging chamber equipped with UV-340 lamps. The irradiance at the specimen surface was 11 × 10³ μW·cm^-2, and the black-panel temperature during irradiation was 60 °C. The dry intermittent protocol comprised 2 h of UV irradiation at a 60 °C black-panel temperature followed by 4 h of dark conditioning at 20 °C under programmed chamber control. Each 6 h immersion-assisted cycle followed the sequence of 2 h UV irradiation in the chamber at 60 °C, removal of the specimens for 1 h static water immersion at 25 °C, and return to the chamber for 3 h dark conditioning at 20 °C. The reported durations of 72, 120, 240, 360, and 480 h are elapsed protocol times. The 72–480 h UV sequence was selected to capture both early-stage and progressive deterioration, with 480 h used as the terminal comparison point for replicate testing of the three UV protocols. Accordingly, the 480 h endpoint corresponds to 480 h of UV irradiation under continuous exposure and 160 h of UV irradiation under each intermittent protocol. Comparisons between continuous and intermittent UV therefore represent whole-protocol contrasts. The comparison between the two intermittent protocols evaluates the combined effect of replacing 1 h of 20 °C dark conditioning with 1 h of static water immersion at 25 °C while retaining the same UV-on duration, UV-stage temperature, and total cycle length. Representative aging chambers and specimen arrangements are shown in Supplementary Figure SA1.

TABLE 1. Accelerated aging protocols and principal exposure stressors.

RH, relative humidity; UV, ultraviolet. At an elapsed protocol time of 480 h, the accumulated UV-on time was 480 h for continuous UV and 160 h for each intermittent protocol.

2.4 Binder-scale characterization

2.4.1 Tensile properties

Tensile tests were performed according to GB/T 1040.2–2022 (Standardization Administration of China, 2022) using type 1BA dumbbell specimens with a thickness of 2 mm and a gauge length of 25 mm. Tests were conducted at 23 °C ± 2 °C with a crosshead speed of 50 mm·min^-1. The unaged, TO-16 h, TO-20 h, and HT-20 h groups each contained five independently prepared specimens. The continuous-UV tensile-strength series also contained five specimens at each duration. Tensile strength and elongation at break represented binder-scale load-bearing capacity and deformability. The continuous-UV trajectory was presented using the measured specimen values and normalized mean retention.

2.4.2 Molecular-weight characterization

GPC was performed on a Waters 1515 system equipped with a Waters 2414 refractive-index detector and three serial PLgel 5 µm Mixed-C columns (300 mm × 7.5 mm; nominal separation range 500–1,000,000 Da). Chromatographic-grade THF containing 0.025% BHT was used at 1.0 mL·min^-1 and 35.0 °C ± 0.1 °C with a 100 µL injection. Cured-film fragments were dissolved at approximately 2 mg·mL^-1 for 24 h under sealed, dark conditions and filtered through a 0.45 µm PTFE membrane. Ten monodisperse polystyrene standards and third-order calibration were used in Empower 3; M n, M w, M p, and M z are therefore reported as polystyrene-equivalent relative molecular masses. Dispersity was calculated as Đ = M w / M n. GPC tracked molecular-weight evolution in the filtered THF-soluble fraction, while the THF-insoluble measurement provided a complementary operational indicator of aging-induced insolubilization. Full method parameters and complete molecular-weight series are reported in Supplementary Tables SB1–SB6. Each aging endpoint was represented by one independently aged binder sample analyzed by a single injection; no repeat injections were performed. The GPC results were interpreted descriptively at each protocol endpoint, whereas inferential comparisons were based on the mechanical replicate datasets.

2.4.3 THF-insoluble fraction

Selected aging states were evaluated by 48 h Soxhlet extraction in THF using 250 mL extractors and quantitative filter-paper thimbles. Residues were vacuum-dried at 80 °C for 12 h to constant mass. Two blank thimbles accompanied each batch, and their before/after masses were used to correct thimble mass change. The resulting THF-insoluble fraction is an operational solvent-insoluble measure and is not equated with absolute crosslink density. Corrected triplicate results are reported in Supplementary Table SB6.

2.5 Mixture-scale performance

Wheel-tracking characterized high-temperature stability through dynamic stability (DS) at 60 °C and 0.7 MPa, with a tracking rate of 42 passes·min^-1. The 300 mm × 300 mm × 50 mm slabs were rolled for 16 passes, conditioned in the mold at 50 °C for 2 h, and rolled for a further 8 passes. The slabs remained in the mold at room temperature for 4 d, were then demolded, and were conditioned at 25 °C/80% RH to a total curing age of 7 d before aging or performance testing. Low-temperature beams and splitting specimens were cut from the 7 d slabs and reconditioned at 25 °C/80% RH for 24 h after cutting. Low-temperature cracking resistance was evaluated by three-point bending at −10 °C using 250 mm × 35 mm × 30 mm beams (200 mm span, 50 mm·min^-1 loading, and 4 h conditioning at −10 °C). Moisture-conditioned splitting strength was evaluated using a freeze–thaw conditioning sequence adapted from JTG E20-2011 (T 0729) (Ministry of Transport of the People’s Republic of China, 2011), which was in force when the tests were conducted. Specimens were vacuum-saturated at a vacuum level of 97.3 kPa for 15 min, returned to atmospheric pressure, and immersed in water at room temperature for a further 30 min. The saturated specimens were frozen at −18 °C for 16 ± 1 h and then transferred immediately to a 60 °C water bath for 24 h. After surface drying, the specimens were loaded at 50 mm·min^-1. The conditioned splitting strength itself was used as the comparison indicator across aging protocols. Moisture-conditioned particle retention was assessed by Cantabro loss after 48 h immersion at 60 °C and 24 h at room temperature; the Los Angeles drum was operated without steel balls at 30–33 r·min^-1 for 300 revolutions. Together, the four tests represented rutting resistance, low-temperature deformation capacity, residual cohesion, and particle-retention resistance.

For UV-aged mixtures, the pavement top face remained upward toward the lamps and specimens were not turned during exposure. The aged top face was used for wheel tracking and placed on the tensile side in bending; for splitting and Cantabro testing, specimen orientation was held constant across all three protocols. This fixed exposure geometry provided a common basis for comparing protocol-specific surface-to-bulk performance responses. Representative binder- and mixture-scale test configurations are shown in Supplementary Figure SA2.

2.6 Data processing and statistical analysis

Static results are reported as mean ± sample standard deviation. Sample sizes were n = 5 for binder tensile strength and elongation at break, n = 3 for dynamic stability, low-temperature bending, and Cantabro loss, and n = 6 for splitting strength in the unaged, TO-16 h, and HT-20 h groups; the TO-20 h splitting group contained n = 5. One-way ANOVA evaluated overall differences among the static-aging groups. The TO-20 h versus HT-20 h contrast was treated separately as a planned comparison because it addressed the specific equal-duration protocol comparison, whereas the omnibus ANOVA assessed overall group differences. The planned equal-duration comparison used two-sided Welch independent-samples t-tests, followed by Holm correction across six indicators at α = 0.05. At the 480 h UV endpoint, each protocol contained three independently tested mixture specimens; pairwise protocol comparisons used two-sided Welch tests with Holm correction across the three comparisons within each indicator. Retention ratios were calculated relative to the corresponding 0 h reference of each experimental series, while Cantabro loss was reported as an absolute loss-type indicator. For the UV-series splitting-strength retention, a separately prepared 0 h control value of 2.72 MPa was used as the normalization reference; this single control value was not included in inferential statistical comparisons. Supplementary Appendices A–C provide the static replicate data, the complete GPC series, selected UV key-point data, and the replicate-tested 480 h UV endpoint results.

3 Results

3.1 Equal-duration static-aging response at the binder scale

The binder-scale static-aging responses are summarized in Table 2 and Figure 3. All static aging conditions reduced binder tensile strength and elongation at break. Tensile strength decreased from 27.90 ± 0.84 MPa in the unaged state to 15.35 ± 0.57 MPa after TO-16 h, 13.52 ± 0.60 MPa after TO-20 h, and 18.13 ± 0.55 MPa after HT-20 h. The corresponding strength retentions were 55.02%, 48.48%, and 64.98%. Elongation at break decreased from 99.98% ± 5.62% to 34.96% ± 2.73%, 29.740% ± 1.927%, and 45.02% ± 3.00%, respectively. At the equal 20 h endpoint, the thermo-oxidative group had significantly lower tensile strength and elongation at break than the hygrothermal group (Holm-adjusted p < 0.001 for both indicators).

TABLE 2. Binder- and mixture-scale performance under static aging conditions.

Values are mean ± sample SD. Tensile strength and elongation at break: n = 5; dynamic stability, flexural strain, and Cantabro loss: n = 3. Splitting strength: n = 6 for unaged, TO-16 h, and HT-20 h and n = 5 for TO-20 h. TO, thermo-oxidative aging; HT, hygrothermal aging; DS, dynamic stability.

Extending thermo-oxidative exposure from 16 to 20 h reduced mean tensile strength from 15.35 to 13.52 MPa and mean elongation at break from 34.96% to 29.74%, confirming continued mechanical deterioration over the tested interval. Static binder and mixture results are summarized in Table 2, and the planned duration-matched statistical comparisons are reported in Table 3.

TABLE 3. Planned equal-duration comparisons between thermo-oxidative and hygrothermal aging at 20 h.

3.2 Molecular-weight and THF-insoluble-fraction evolution

The molecular-scale responses are summarized in Figure 4 and Supplementary Appendix B. The unaged THF-soluble fraction had M n = 7122.9, M w = 8569.1, M p = 6718.3, and M z = 9636.4, with dispersity Đ = 1.203. At the available static molecular endpoints, HT-20 h reduced M n and M w to 5697.8 and 7540.7 and increased Đ to 1.323, whereas TO-16 h produced M n = 6054.8, M w = 7711.9, and Đ = 1.274. Because GPC was available at TO-16 h and HT-20 h rather than at equal duration, these measurements are interpreted as protocol-specific molecular endpoints. The HT-20 h endpoint showed the lower soluble-fraction molecular masses, whereas the independent equal-duration mechanical comparison showed greater tensile deterioration under TO-20 h.

UV exposure produced progressive molecular-weight reduction and distribution broadening. At 480 h, M n values were 2135.9 under continuous UV, 2491.3 under intermittent UV, and 1779.4 under intermittent UV–immersion; the corresponding Đ values were 2.566, 2.338, and 2.888. The immersion-assisted protocol produced the strongest terminal enrichment of lower-molecular-weight material in the THF-soluble population.

The THF-insoluble fraction increased from 9.69% ± 0.67% in the unaged binder to 19.30% ± 1.36% after TO-16 h and 15.17% ± 1.16% after HT-20 h. It reached 37.80% ± 2.34% after 480 h continuous UV and 32.39% ± 2.11% after 480 h intermittent UV–immersion. These measurements provide complementary information on material outside the filtered THF-soluble fraction.

3.3 UV-aging trajectory of binder tensile response

The continuous-UV tensile-strength series declined nonlinearly from 27.90 ± 0.84 MPa to 12.09 ± 0.93 MPa after 480 h, corresponding to 43.35% retention. The specimen-level measurements and normalized mean trajectory show rapid early deterioration followed by a progressively smaller rate of strength loss under continued laboratory exposure (Figure 5).

At the 480 h endpoint, the three UV protocols also produced different THF-soluble molecular-weight distributions: intermittent UV–immersion had the lowest soluble-fraction molecular mass and the highest dispersity, while the selected THF-insoluble measurements followed a different protocol order. These molecular results are reported descriptively because each GPC endpoint was represented by one independently aged sample and one injection.

3.4 Mixture-scale static-aging response

Mixture performance declined under all static aging conditions (Table 2; Figure 6). Extending thermo-oxidative exposure from 16 to 20 h reduced dynamic stability from 51,308.0 to 48,683.9 passes mm^-1, flexural strain from 6826.1 to 6502.8 με, and splitting strength from 1.708 to 1.580 MPa, while Cantabro loss increased from 12.481% to 13.213%. The consistent progression across the four mixture indicators demonstrates continued deterioration as thermo-oxidative exposure increased from 16 to 20 h.

At the equal 20 h endpoint, TO-20 h caused significantly greater losses in binder tensile strength, elongation at break, and mixture splitting strength. The HT-20 h group showed lower mean dynamic stability and flexural strain and higher mean Cantabro loss, although those three pairwise differences were not significant after Holm correction (Table 3). Thus, the equal-duration comparison produced a scale- and indicator-dependent response pattern rather than a single protocol ranking. Overall static-group effects were significant for all six indicators (Supplementary Table SC6).

Splitting strength was particularly sensitive to thermo-oxidative exposure, reaching 1.580 ± 0.094 MPa after TO-20 h compared with 1.825 ± 0.114 MPa after HT-20 h (Holm-adjusted p = 0.014). Cantabro loss increased most under HT-20 h, reaching 13.890% ± 1.084%, while TO-20 h produced 13.213% ± 0.967%; the difference was not significant after correction.

3.5 Mixture response at the 480 h UV endpoint

At the 480 h elapsed-protocol endpoint, the three UV schedules produced distinct response patterns (Table 4; Figure 7). Flexural strain differed significantly among all three protocols after Holm adjustment (adjusted p = 0.0192, 0.00327, and 0.0192 for the three pairwise comparisons), making it the most discriminating mixture indicator in the present dataset. Dynamic stability under intermittent UV–immersion was significantly lower than under continuous UV and dry intermittent UV (adjusted p = 0.0113 and 0.0159, respectively). Most importantly for isolating the protocol modification, the dry-intermittent and immersion-assisted schedules had the same UV-on duration, UV-stage temperature, and 6 h cycle length; replacing 1 h of dark conditioning with 1 h of water immersion reduced terminal dynamic stability from 28,764 ± 951.9 to 24,652 ± 1,071.2 passes mm^-1. Splitting strength under continuous UV was significantly higher than under both intermittent protocols (adjusted p = 0.0142 for both comparisons), whereas the two intermittent protocols did not differ significantly (adjusted p = 0.489). Cantabro loss increased from 12.92% ± 0.999% under continuous UV to 14.56% ± 1.231% under intermittent UV and 15.95% ± 1.515% under intermittent UV–immersion, but none of these pairwise differences was significant after correction (adjusted p ≥ 0.159). The 2.72 MPa value used to calculate UV-series splitting-strength retention was a separately prepared 0 h normalization reference and was not part of the 480 h pairwise statistical comparisons.

TABLE 4. Mixture performance after 480 h under the three UV aging protocols.

Values are mean ± sample SD (n = 3). Retention ratios were calculated using the corresponding 0 h references for the UV-aging series: 64,350.7 passes mm^-1 for dynamic stability, 7,935.9 με for flexural strain, and 2.72 MPa for splitting strength. The 2.72 MPa splitting-strength reference was obtained from a separately prepared 0 h control and was used only for normalization of the UV-series retention values; it is therefore distinct from the replicated static-aging unaged group reported in Table 2. Cantabro loss increases were calculated relative to 9.23%. Pairwise statistics are reported in Supplementary Table SC5.

4 Discussion

4.1 Cross-level interpretation of molecular and mechanical responses

The static endpoints show why THF-soluble molecular mass should not be used as a stand-alone surrogate for mechanical aging severity. HT-20 h had the lower soluble-fraction molecular masses, whereas TO-16 h had the higher THF-insoluble fraction; independently, the duration-matched 20 h comparison showed greater thermo-oxidative deterioration in binder tensile strength and elongation at break. Because GPC characterized only the filtered THF-soluble population, these results are consistent with simultaneous changes in the soluble molecular population and in the less-soluble network fraction. Accordingly, the apparent mismatch between soluble-fraction molecular-mass evolution and tensile deterioration should be interpreted as a cross-level decoupling rather than as contradictory evidence. The macroscopic response of the cured binder reflects contributions from material populations and load-bearing structures that are not fully represented by the filtered THF-soluble fraction analyzed by GPC. Prior polyurethane-aging research has shown that thermal, photo-oxidative, and hydrolytic exposures can involve competing chain scission, oxidation, and network reorganization (Zhang et al., 2024; Godlewska et al., 2024; Mayer-Trzaskowska et al., 2024; Liu et al., 2024). The present dataset supports the narrower conclusion that soluble-fraction molecular mass alone is insufficient to rank mechanical severity; it does not establish a unique chemical mechanism.

This distinction is particularly important for PU-bound aggregate composites because the cured polymer is the primary binder rather than a modifier dispersed within an asphalt phase. Mixture-scale performance additionally depends on air-void structure, aggregate gradation, curing state, and aggregate–polyurethane stress transfer (Zhou et al., 2025; Zhao et al., 2026a,b; Xu Y. et al., 2024; Chen et al., 2024; Gao et al., 2022). Recent composite-asphalt studies likewise show that aging can alter molecular or microstructural indicators while improving some high-temperature responses and degrading low-temperature or fatigue responses (Fan et al., 2025). More mechanism-intensive work has combined rheology, spectroscopy, microscopy, and molecular simulation to establish a stronger micro-to-macro causal chain (Hu, 2026). Relative to those approaches, the role of the present GPC and THF-insoluble measurements is deliberately operational: they provide molecular-level context for the measured mechanical responses without being used to claim a complete degradation mechanism.

4.2 Protocol comparability and the role of water immersion

The interpretation of continuous versus intermittent UV requires a separate boundary. At 480 h elapsed protocol time, continuous exposure accumulated 480 h of UV irradiation, whereas each intermittent schedule accumulated 160 h. Consequently, those contrasts cannot be assigned solely to intermittency or UV dose and are treated here as whole-protocol comparisons. Among the UV comparisons, the dry-intermittent versus intermittent UV–immersion contrast is the most controlled because UV-on time, UV-stage temperature, and cycle length are matched, while 1 h of dark conditioning is replaced by 1 h of water immersion. This comparison therefore provides the strongest direct evidence in the present dataset for the performance consequence of introducing the immersion step within the intermittent schedule. Across the four terminal mixture indicators, intermittent UV–immersion produced the lowest mean dynamic stability, flexural strain, and splitting strength and the highest mean Cantabro loss, but this ranking is specific to the tested protocol set and should not be generalized as a dose-equivalent UV severity ranking.

The duration-matched static comparison serves a similar purpose. Adding the TO-20 h group prevented the original TO-16 h versus HT-20 h contrast from being interpreted as an environmental effect when exposure time also differed. This control materially changed the strength of the inference: thermo-oxidative aging was more damaging for binder tensile strength, elongation at break, and splitting strength at equal elapsed duration, whereas dynamic stability, flexural strain, and Cantabro loss did not support the same statistically significant ranking. For the UV protocols, the matched intermittent comparison provides the most defensible evidence for the consequence of introducing water. Previous asphalt-binder work has also shown that coupling water with UV exposure can intensify aging-related changes (Qian et al., 2021), but the present results should be interpreted at the composite-performance level rather than as proof of a specific hydrolytic or interfacial mechanism.

4.3 Indicator-specific implications for accelerated durability assessment

The results show that indicator sensitivity depends on the exposure comparison being made. Under the equal-duration static comparison, binder tensile strength and elongation at break quantified losses in load-bearing capacity and deformability, while splitting strength tested whether the same direction of response was retained at the mixture scale. At the 480 h UV endpoint, flexural strain provided the strongest statistical discrimination among the three tested protocols. Dynamic stability was particularly sensitive to introducing immersion within the matched intermittent schedule, whereas splitting strength differentiated continuous from intermittent dry UV only as a whole-protocol contrast because accumulated UV-on times differed.

Table 5 summarizes these observations for the tested moisture-curing MDI-based polyether polyurethane and PC-13 composite. The table is intended as a concise summary of indicator behavior within the present dataset rather than a universal test-selection hierarchy. Transfer to other polyurethane chemistries, aggregate gradations, specimen geometries, or field exposure conditions requires independent validation.

TABLE 5. Summary of indicator responses to the tested aging protocols.

TO, thermo-oxidative aging; HT, hygrothermal aging. Statistical decisions are based on the planned equal-duration comparisons and the 480 h UV endpoint comparisons reported in Table 3 and Supplementary Table SC5.

The practical implication is that no single response variable captured every deterioration mode. Combining a binder-level mechanical response with deformation- and cohesion-sensitive mixture indicators provides a more defensible interpretation of accelerated aging, while the most informative mixture indicator depends on the environmental contrast being evaluated.

4.4 Limitations and transferability

The study has three principal boundaries. First, each GPC endpoint was represented by one independently aged binder sample analyzed by a single injection, and the available static molecular endpoints (TO-16 h and HT-20 h) were not duration-matched. The molecular results are therefore descriptive and are not used for inferential ranking of the static protocols. Second, the continuous and intermittent UV schedules accumulated different UV-on times at the same elapsed protocol duration; comparisons between them are whole-protocol contrasts rather than isolated intermittency or dose effects. Third, the response patterns were obtained for one moisture-curing MDI-based polyether polyurethane, one dense PC-13 gradation, and fixed laboratory specimen geometries. Polyurethane chemistry, aggregate gradation, air-void structure, and curing state can influence not only the absolute engineering response but also the relative sensitivity of individual aging indicators. Recent PU-mixture studies demonstrate that air voids, gradation, and curing conditions can alter engineering response (Zhou et al., 2025; Zhao et al., 2026a,b), so external transfer should be verified rather than assumed. No laboratory-to-field acceleration factor or service-life conversion is claimed. Future work should prioritize duration-matched replicated molecular characterization, complementary spectroscopic or viscoelastic measurements, additional polyurethane formulations and mixture structures, and field-exposure validation.

5 Conclusion

This study quantified protocol-dependent aging responses of a moisture-curing polyether polyurethane binder and a dense polyurethane-bound pavement composite. The principal findings are:

• At the equal 20 h endpoint, thermo-oxidative aging produced significantly lower binder tensile strength and elongation at break than hygrothermal aging (Holm-adjusted p < 0.001 for both indicators). Splitting strength was also lower after thermo-oxidative aging (1.580 ± 0.094 MPa) than after hygrothermal aging (1.825 ± 0.114 MPa; Holm-adjusted p = 0.014). For these load-bearing and deformation responses, the dry thermo-oxidative protocol was more damaging at equal elapsed duration.
• Static protocol severity remained indicator-dependent. The hygrothermal group had lower mean dynamic stability and flexural strain and higher mean Cantabro loss than the equal-duration thermo-oxidative group, but these differences were not significant after Holm correction. A single mixture indicator therefore cannot define a universal static-aging ranking for the tested composite.
• The available TO-16 h and HT-20 h molecular endpoints showed complementary soluble- and insoluble-fraction responses: HT-20 h had the lower THF-soluble molecular masses, whereas TO-16 h had the higher THF-insoluble fraction. Considered together with the independent equal-duration mechanical comparison, these results show that filtered THF-soluble molecular weight alone is insufficient to rank mechanical aging severity. Because GPC used one independently aged sample and one injection per endpoint, this molecular interpretation remains descriptive.
• At 480 h elapsed protocol time, intermittent UV–immersion produced the lowest mean dynamic stability, flexural strain, and splitting strength and the highest mean Cantabro loss. Flexural strain distinguished all three UV protocols, while the matched intermittent comparison showed that introducing the immersion step significantly reduced dynamic stability. The monotonic increase in mean Cantabro loss was not statistically significant among the three protocols after Holm correction. Differences between continuous and intermittent UV are interpreted only as whole-protocol contrasts because accumulated UV-on duration differed (480 h versus 160 h).

Overall, the measured aging severity was not invariant across observation levels or performance indicators. Duration matching strengthened the static protocol comparison, and matching the intermittent UV cycle isolated the performance consequence of the immersion step more defensibly than comparisons with continuous UV. The results therefore support protocol-specific, multi-indicator interpretation of accelerated aging for polyurethane-bound pavement composites, while field transfer and mechanistic attribution require additional validation.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. OpenAI ChatGPT (web application; GPT-5.6 Sol; OpenAI; accessed August 2026) was used after completion of the experimental and statistical analyses to assist with English-language refinement and manuscript organization. No generative AI tool was used to generate experimental data, perform statistical analyses, determine statistical significance, or independently interpret the results. All AI-assisted outputs were reviewed and verified by the authors, who take full responsibility for the accuracy and integrity of the final manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmats.2026.1964302/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Xiaolong Yang, Guangxi University, China

Reviewed by: Zipeng Ma, Jilin University, China

Xi Chen, The University of Alabama, United States