Introduction

Natural load-bearing materials, including muscles, tendons, and cartilages, demonstrate autonomous repetitive self-repair capabilities while maintaining mechanical function, providing inspiration for developing durable synthetic self-healing materials. However, conventional self-healing polymers employing supramolecular crosslinking with homogeneous architectures typically exhibit intrinsic mechanical weaknesses, fundamentally limiting their applicability compared to natural counterparts that simultaneously achieve high stiffness, toughness, and fatigue resistance1. Recent advances employ molecular engineering strategies to enhance the mechanical performance of self-healing polymers through energy dissipation mechanisms, such as incorporating dynamic covalent bonds2, dual crosslink3, or interpenetrating polymer networks4. While these approaches successfully improve strength and toughness, they demonstrate limited capacity to improve elastic modulus (E) while showing negligible impact on fatigue threshold (Γth) elevation. In key engineering fields such as aerospace, automotive, and healthcare, high stiffness is essential to resist excessive deformation, while a high fatigue threshold is crucial to withstand crack growth under cyclic loading5. From the theory of the Lake-Thomas model, the Γth value of a single-network polymer is determined by the equation Γth = Jabn1/2, where a is the length of the monomer, b is the number of single bonds per unit volume, J is the energy per covalent bond, and n is the number of monomers in the chain6,7. To significantly enhance the fatigue threshold of a homogeneous self-healing polymer, it is necessary to increase the number of monomers (n) between cross-linkers. However, this typically leads to reduced E value, as indicated by the relation of E ~ n−1, and vice versa. Therefore, traditional repairable polymer networks have been modified to be either stiff or fatigue-resistant, but seldom both8,9,10. This significant challenge in achieving both extreme stiffness and fatigue resistance is also evident on the plot of Young’s modulus (the maximum stress to resist elastic deformation) and fatigue threshold (the minimum energy required to facilitate crack propagation during cyclic loads) (Fig. 1a and Table 1)8,11,12,13,14,15,16,17,18,19,20,21,22, which displays a large area in the top right that is nearly vacant. Overall, achieving ultra-high stiffness, robust fatigue resistance, and highly efficient self-healing in the same synthetic polymer proves to be extremely challenging.

Fig. 1: Interlinking multiscale, dynamic high-order structure to synchronously enhance the fatigue threshold and modulus of self-healing materials.
figure 1

a Comparison of Young’s modulus and fatigue threshold of our work and other reported viscoelastic materials; the statistical data can be found in Table 1. b Schematic of multiscale structures comprising a nanoscale continuous hard phase and microscale 3D-interconnected framework; this hierarchical design enhances energy storage, leading to an amplified fatigue threshold as modeled by the generalized Lake-Thomas theory29. c Schematic illustration of multiscale stress deconcentration, where robust interfacial hydrogen bonds between the continuous hard phase and the 3D-interconnected MXene framework facilitate stress transfer between dynamic high-order structures across different scales; the red region indicates the area subjected to stress. Source Data are provided as a Source Data file.

Full size image
Table 1 Comparison of Young’s modulus and fatigue threshold between our CPU/MP composite and other reported viscoelastic materials
Full size table

The integration of dispersed rigid constituents such as particles, nanocrystals, and heterogeneous domains into the primary polymer networks is a well-established approach for simultaneously enhancing stiffness and toughness23. While this toughening mechanism effectively mitigates crack propagation rates above the fatigue threshold24,25, it demonstrates limited efficacy in elevating the fundamental fatigue threshold itself. In such hybrid systems, the fatigue threshold remains intrinsically governed by the topological architecture of the primary polymer network. This critical parameter exhibits direct correlation with the number of chain monomers in the polymer backbone, maintaining functional independence from the incorporation of secondary reinforcement components. A classic example is natural rubber filled with carbon particles, which has shown an increase in modulus by one to two orders of magnitude, but its fatigue threshold, whether reinforced or not, has remained consistently limited to 50–100 J m−2 for decades26. Given this limitation, a clever bioinspired strategy is often utilized to structuralize the isolated rigid constituents at the nanoscale, mesoscale, or macroscale. This approach enables the formation of high-order structures—whose dimensions extend beyond the molecular scale, typically spanning from nanometers to micrometers—that possess high energy states, including aligned crystalline domains, continuous separated microphases, and percolated particle clusters. These configurations effectively suppress and decelerate crack growth under cyclic loads, mirroring the function of unique microstructures in natural materials17,18,27,28. The enhanced fatigue thresholds through this strategy can be explained by the generalized Lake-Thomas model29, where it is the bionic high-energy structures, rather than the polymer chains, that are disrupted to advance the fatigue crack. Since the elastic energies stored in these stiff high-order structures are much greater than in the primary polymer networks, the energy required for the propagation of fatigue crack is also significantly higher19,30. However, the resultant modulus and fatigue thresholds improved by these bionic structures still remain low, typically less than 1 MPa and 1000 J m−2, respectively (Table 1). Even worse, most high-energy structures developed in these synthetic mimics are manufactured through a top-down process, entirely different from the self-assembly seen in biological tissues19,29,31,32. Whereas various high-order structures can now be customized on demand by top-down technology, few exhibit dynamic attributes, consequently altering the bulk dynamic response of synthetic mimics, particularly by bereaving the self-healing capabilities compared to the initial polymer matrices.

Here, we substantially and synchronously amplify the fatigue threshold and modulus while achieving ultra-fast and highly efficient self-healing by integrating dynamic high-order structures at various scales through robust interfacial self-assembly coupling (Fig. 1b). Our strategy maximizes the comprehensive energy of the interconnected high-order reinforcement while preserving its intrinsic dynamics. Initially, we synthesized a self-healing polyurethane network containing a nanoscale continuous dynamic hard phase—referred to as the first dynamic high-order structure, with a characteristic structural unit size of approximately 10.0 nm (Fig. 1b). This was achieved by utilizing 2-uredio-4-pyrimidone (UPy) supramolecular motifs as structural regulators for their fibrillar assembly tendency20,33. Subsequently, a hydrogen bonding-driven self-assembly process was used to integrate 2D MXene nanosheets within this preformed network, leading to the formation of a microscale three-dimensional interconnected dynamic framework. This constitutes the second dynamic high-order structure, featuring basic structural units with a size of approximately 0.3 μm (Fig. 1b). These structures integrate through robust interfacial hydrogen bonds, forming an interlinked high-energy network that enables multiscale stress deconcentration (Fig. 1c). During cyclic loading, fatigue crack propagation requires fracture of this multiscale interlinked network rather than molecular-scale polymer chains. In addition, due to the ability of MXene nanosheets to convert near-infrared light into heat34,35, the CPU/MP material is capable of autonomous self-healing. Consequently, our CPU/MP composite achieves a fatigue threshold of 8226.3 J m⁻², surpassing many stretchable and self-healable materials (Table 1). Synergy between the dynamic hard phase, 3D framework, and their interfacial adhesion further amplifies Young’s modulus to ~ 51.1 MPa, exceeding several commercial particle-filled rubbers (Table 1). Critically, the interconnected MXene framework facilitates rapid photothermal conversion and heat conduction, allowing for NIR-triggered self-healing in timescales of minutes, with ~100% efficiency. Moreover, this structure enhances thermomechanical stability, enabling reliable CPU/MP operation at 70 °C while exhibiting a fatigue threshold of 965.5 J m⁻². This strategy of interlinking multiscale structures yields a viscoelastic composite combining high fatigue threshold, high modulus, and improved thermomechanical stability, resolving the long-standing conflict between self-healing dynamics, fatigue resistance, and stiffness. Given the prevalence of high-order structures in polymers and composites, our methodology shows broad applicability. Implementation could enable conventional self-healing materials to serve emerging fields like robotics, additive manufacturing, and automotive engineering.

Results

Design of the multiscale interlinked high-order structure

We employed a bottom-up self-assembled approach to successfully fabricate a dynamic self-healing composite with a tunable multiscale interlinked structure. The preparation process of this structural composite primarily involves three key stages: the synthesis of waterborne self-healing polyurethane with a continuous dynamic hard phase (CPU), the exfoliation and modification of MXene by phytic acid (MXene-PA), and the hydrogen bonding-driven self-assembly of MXene-PA-wrapped CPU (MXene-PA@CPU) into a high-performance CPU/MP composite (Fig. 2a). Inside CPU/MP, the microscale 3D-interconnected MXene framework is tightly connected with the nanoscale continuous dynamic hard phase through high-density interfacial hydrogen bonds, as depicted in Fig. 2a. A waterborne CPU was first synthesized via condensation polymerization, where a polycaprolactone (PCL, Mn = 2000 g mol−1)-based prepolymer was initially synthesized in the presence of 2.5 equivalents of methylene-bis(4-cyclohexylisocyanate) (HMDI) and catalytic amount of dibutyltin dilaurate (DBTDL), followed by a chain extension reaction using 2,2-dimethylolbutanoic acid (DMBA) and 5-(2-hydroxyethyl)−6-methyl-2-aminouracil (HMA) in a predetermined ratio. Detailed polymerization procedures of CPU and its chemical composition analysis are provided in the Supplementary Information (Supplementary Figs. S1–S4). The selected HMDI possesses a high level of steric hindrance due to its bulky alicyclic structure, preventing crystallization of the assembled hard domain and facilitating the chain mobility of the resulting CPU (Supplementary Figs. S5, S6), which is crucial for achieving intrinsic self-healing36,37,38. HMA was used to introduce the 2-ureido-4-pyrimidinone (UPy) motif into the hard segments of CPU. The supramolecular UPy dimer tends to undergo fibrillar assembly due to its planar aromatic structure20,33, which can further induce the formation of a continuous dynamic hard phase (Supplementary Fig. S7b–d). In sharp contrast, assembled condensed phases without UPy motifs are islanded (Supplementary Fig. S7a). Meanwhile, the resultant self-healing polyurethane with an islanded dynamic hard nanophase (IPU) serves as an ideal example to demonstrate the particularity of phase structures on the comprehensive mechanical and thermomechanical properties. DMBA was selected to introduce hydrophilic ionic groups, which react with triethylamine to ensure the emulsification of CPU39, producing uniformly dispersed colloidal particles in water (Fig. 2b and Supplementary Fig. S8), which is the prerequisite to manufacture a follow-up 3D-interconnected MXene framework.

Fig. 2: Design of the multiscale interlinked high-order structure.
figure 2

a Schematic representation of the fabrication process for the multiscale interlinked high-order structure in the CPU/MP composite. b Transmission electron microscopy (TEM) image of the CPU emulsions. c–e Cross-sectional TEM images of the CPU/MP composite at different magnifications. f Visual representation of the optimized CPU/M and CPU/MP composite structures; the CPU/MP model (right) comprises five CPU polymer chains (including two PCL soft segments and one UPy-containing hard segment), two MXene nanosheets, and two PA molecules; the CPU/M model (left) has an identical polymer composition but lacks PA molecules. g Calculated cohesive energy density of the CPU/M and CPU/MP composites based on molecular dynamics simulations. h Temperature-dependent FT-IR spectrum of the − C = O groups (the primary hydrogen bond acceptors) in CPU/M and CPU/MP upon heating from 25 to 100 oC. Source Data are provided as a Source Data file.

Full size image

Ti3AlC2 MAX phase precursor was exfoliated into 2D Ti3C2Tx MXene nanosheets using a LiF/HCl solution40. High-magnification transmission electron microscopy (HRTEM) revealed the single-layered structure of the well-exfoliated MXene nanosheets (Supplementary Fig. S9), which was further confirmed by X-ray diffraction patterns (Supplementary Fig. S10). Subsequently, a specific amount of polyhydroxy PA molecules was employed to interact with the abundant functional groups (-OH, -F, = O) on the surface of the MXene nanosheets, forming PA-modified MXene additives (MXene-PA, Supplementary Fig. S11). The dendritic polyhydroxy nature of PA facilitates the scaffolding of individual MXene nanosheets, preventing their agglomeration in water41. The MXene-PA nanosheets demonstrated highly stable dispersion, showing no signs of sedimentation even after two weeks of incubation in water (Supplementary Fig. S12). Based on the well-stabilized MXene-PA dispersion and the uniformly dispersed CPU colloidal particles, we successfully assembled a CPU/MP composite. Specifically, due to the abundant hydrogen-bonding sites in the branches of PA molecules, including multiple phosphoric acid groups and hydroxyls, MXene-PA nanosheets were directionally adsorbed onto the surface of CPU colloidal particles (average size of 350 nm), theoretically forming core-shell-structured MXene-PA@CPU microspheres (Supplementary Fig. S13). As the water medium in this hybrid system was gradually removed, the MXene-PA@CPU microspheres fused together through the high-density hydrogen bonds of their MXene-PA shells, inducing the formation of a 3D-interconnected MXene framework embedded within the interwoven CPU matrix. The strong hydrogen-bonding sites in the CPU chains, including UPy motifs and urethane moieties, are mainly located in its hard segments. Consequently, these two high-energy yet dynamic networks, i.e., the nanoscale continuous dynamic hard phase and the microscale 3D-interconnected MXene framework, are tightly interconnected through robust hydrogen bonds, forming the anticipated multiscale interlinked high-order structure.

The cross-section morphologies of the CPU/MP composite were meticulously examined using TEM, as clearly illustrated in Fig. 2c and Supplementary Fig. S14a. The MXene-PA nanosheets were selectively assembled at the interstitial space of CPU-CPU microspheres, concatenated with each other to form an organized 3D-interconnected MXene framework within the CPU matrix. This structure resembles the interlaced collagen fibrillar network found in biological tissues8,20. Notably, the average thickness of this assembled framework is significantly greater than that of a single layer of MXene nanosheet (Fig. 2c). The increased thickness is attributed to the stacking and adhesion of individual MXene-PA nanosheets through their high-density hydrogen bonding sites during the CPU/MP fabrication process. In essence, the MXene-PA nanosheets function similarly to double-sided tape. As a result, the interfaces between adjacent MXene nanosheets are seamlessly welded, creating mechanically robust cross-points (Fig. 2d). These cross-points are crucial for distributing local stress throughout the entire 3D-interconnected framework under load19. Given the higher TEM imaging contrast of MXene nanosheets compared to the hard and soft domains in the CPU elastomer, the constructed microscale 3D-interconnected MXene framework and the nanoscale continuous dynamic hard phase are challenging to visualize simultaneously at the same magnification. Therefore, we increased the TEM magnification to identify the nanostructures within the CPU portion of the CPU/MP cross-section. A typical continuous nanophase separation structure was observed in Fig. 2e, where the dark areas with relatively high contrast represented the continuous dynamic hard phase, consistent with the AFM results in Supplementary Fig. S7b–d. For comparison, we fabricated a control composite (CPU/M) by directly mixing uniformly dispersed MXene with CPU colloidal particles. As shown in the cross-sectional TEM images, the MXene nanosheets in the CPU/M composite were not structurally organized and exhibited an irregular distribution (Supplementary Fig. S14a). This finding confirms that the key to form a 3D-interconnected MXene framework is the presence of PA molecules, rather than the inherent excluded volume effect of CPU microspheres. Without PA molecules, the MXene nanosheets cannot pre-assemble on the surface of CPU microspheres due to their weak interfacial interactions, resulting in the failure to form the framework.

Molecular dynamics (MD) simulations were conducted to further validate the role of PA molecules in inducing robust interactions among the components of CPU and MXene. In these simulations, systems comprising CPU chains combined with monolayer MXene nanosheets (both PA-modified or unmodified) were placed in the simulation cell to model their equilibrium conformation. As shown in Fig. 2f and in the experimental procedures, the calculated cohesive energy density (CED) of the system comprising CPU chains and MXene-PA nanosheets reached up to 2178.0 kJ mol−1; however, without PA molecules, the corresponding cohesive energy density was only 1235.7 kJ mol−1 (Fig. 2g). Consequently, PA-modified MXene was directionally driven onto the surface of CPU microspheres, forming MXene-PA@CPU assemblies through strong supramolecular interactions. We further simulated the binding energies of the systems comprising MXene nanosheets or MXene-PA nanosheets (Supplementary Fig. S15). As anticipated, the PA molecules provided strong binding sites for the adhesion of MXene nanosheets, facilitating the connection of MXene-PA@CPU particles via their MXene-PA shell to construct a 3D-interconnected MXene framework. The formation of robust supramolecular interactions between MXene-PA and the hard segments of CPU was confirmed by X-ray photoelectron spectroscopy (XPS) and temperature-dependent Fourier-transform infrared (FT-IR) spectrum analysis. First, CPU/M composite exhibited two peaks in its F 1 s region, where the peak with lower binding energy was assigned to the “free” -Ti-F moiety, and the other to the “bonded” -Ti-F…H-N- moiety (Supplementary Fig. S16). After the introduction of PA molecules, the peak intensity of the “free” -Ti-F moiety significantly decreased, and a new peak, referred to as the “bonded” -Ti-F…H-O- moiety, emerged at 691.2 eV in the CPU/MP composite. Temperature-dependent FTIR spectroscopy was further conducted on the CPU/MP system to confirm the enhanced interfacial hydrogen bonding mediated by the -C=O groups, which are the major hydrogen bond acceptors. At 25 °C, the -C=O stretching vibration manifested at lower wavenumbers for CPU/MP than for CPU/M (1723 cm⁻¹ vs 1725 cm⁻¹), consistent with greater hydrogen bond density. This thermal response contrast becomes more pronounced with increasing temperature: while heating to 100 °C induces a 6 cm⁻¹ redshift (1725 → 1731 cm⁻¹) in CPU/M, CPU/MP displays merely a 2 cm⁻¹ shift (1723 → 1725 cm⁻¹). These observations confirm that the hydrogen-bonding networks involving -C=O groups in CPU/MP exhibit both stronger bond energies and superior thermal stability compared to those in CPU/M, as quantitatively demonstrated in Fig. 2h. These significant changes in supramolecular chemistry provide compelling evidence of the high-density non-covalent interactions between PA molecules and MXene nanosheets, as well as between PA molecules and the hard segments of CPU chains, which are responsible for the successful construction of the multiscale interlinked high-order structures.

Synchronously amplifying stiffness and fatigue resistance

Tensile tests were conducted on IPU, CPU, CPU/MP, and CPU/M materials at a speed of 100 mm min-1 to assess their stiffness (Fig. 3a), quantifiable through the mechanical parameter of Young’s modulus. As depicted in Fig. 3b and Table 2, the initial Young’s modulus of the IPU elastomer, calculated as the slope of the stress-strain curve at minimal elongation (Fig. 3a), was as low as 0.9 MPa. With the introduction of structural regulators, namely UPy macromolecular motifs, to create a continuous dynamic hard phase, the Young’s modulus of the resultant CPU elastomer increased to 5.3 MPa (Fig. 3b), which is 5.9 times higher than that of the IPU. Remarkably, the CPU/MP composite, incorporating an additional 3D-interconnected MXene framework, exhibited an even higher modulus of 51.1 MPa (Fig. 3b), representing an ~56.8-fold increase regarding the initial IPU elastomer. However, the control CPU/M composite with the same MXene contents showed only a modest increase in Young’s modulus (12.2 MPa) compared to CPU/MP, affirming the effectiveness of our design in interlinking multiscale high-order structures to enhance the overall load-bearing capacity. To visually demonstrate the high stiffness, sheets of these four materials with the same thickness, including IPU, CPU, CPU/MP, and CPU/M composite, were attached to a holder (Fig. 3c). As shown, the IPU sheet quickly bent to 90 degrees in the single cantilever model due to its low stiffness. With the progressive integration of high-order structures, the materials resistance to deformation improved, as evidenced by decreasing bending angles, indicating increasingly stiffer behavior. In particular, the optimal CPU/MP composite could easily support its own weight without any bending (Fig. 3c). The mechanical parameters such as strength, maximum strain, and work of fracture (also known as toughness) were also measured from the stress-strain curves in Fig. 3a. Both strength and work of fracture were significantly enlarged, although the maximum strain decreased (Supplementary Fig. S17 and Table 2). In addition, the storage modulus (E’) as a function of stretching rate were measured by dynamic thermomechanical analysis (DMA), showing a marked increase after introducing the multiscale interlinked high-order structure (Supplementary Fig. S18), consistent with the observed trends in Young’s modulus (Fig. 3b). Cyclic loading-unloading tests was then conducted to demonstrate the resilience of CPU/MP composite, which exhibited fast and elastic deformation recovery at strain region ≤100% (Supplementary Fig. S19).

Fig. 3: Stiff yet fatigue resistant properties and mechanisms.
figure 3

a Stress-strain curves of the viscoelastic IPU, CPU, CPU/M and CPU/MP materials. b Comparison of the Young’s modulus for the IPU, CPU, CPU/M and CPU/MP materials. Error bars represent the standard deviation of five independent samples; centers indicate mean values. c Photographs of the IPU, CPU, CPU/M, and CPU/MP materials mounted on a holder in the single cantilever model. d Plots of crack extension per cycle versus the applied energy release rate for the IPU, CPU, CPU/M, and CPU/MP materials. e Stress-strain curves of the CPU/MP composite under cyclic stretching. f Validation of a fatigue threshold as high as 7933 kJ m−2 (λA = 2) for the CPU/MP composite using a single-notch test at the 1000th and 10000th cycles. g Log-log plot of dc/dN versus G for the IPU, CPU, CPU/M, and CPU/MP materials. h Changes in the calculated cohesive energy density (CED) and the H-bond counts for the CPU/M and CPU/MP materials during five cycles at 10% strain, with inserted images illustrating the simulation process. i Trends in electronic conductivity for the CPU/M and CPU/MP materials at the 100th, 1000th, and 10000th cycles, where R1 and RN represent the resistance at the first and Nth cycles, respectively. Inserted TEM images showing the structural integrity of CPU/MP composite after 10000th cycles; the red double arrows represent the orientation of the interlinked structure after 10000th cyclic loadings. Source Data are provided as a Source Data file.

Full size image
Table 2 Statistics of strength, strain at break, and work of fracture of the IPU, CPU, CPU/M, and CPU/MP
Full size table

In addition to its stiffening effect, our CPU/MP composite also demonstrated a significantly increased resistance to fatigue fracture compared to IPU, CPU, and CPU/M. A single-notch method developed by Zhao and colleagues was utilized to evaluate the fatigue thresholds21. Cyclic loadings were initially applied to a pre-cracked specimen under a small applied stretch (λA) (Supplementary Fig. S20a). If the crack remained nearly stationary with the total crack extension length less than 20 μm for 1000 cycles (i.e., a crack extension per cycle (dc/dN) less than 20 nm), the λA value was gradually increased for other pre-cracked specimens until crack propagation exceeding 20 μm was observed. The corresponding energy release rate (G) was derived from the cyclic stress-stretch curve of the uncracked specimen under the same loading condition (Supplementary Fig. S20b), where the maximum stress in each cycle gradually diminished and eventually stabilized after thousands of cycles (Supplementary Fig. S21). By systematically varying the λA, we plotted dc/dN against the G, using linear regression to estimate the fatigue threshold (Fig. 3d). With islanded hard phases, the fatigue threshold of IPU was low as 24.2 J m−2, comparable to the energy required to facture a single-layer polymer chain (~10 J m−2)22,29. In contrast, the CPU with continuous hard phases exhibited a significantly enhanced fatigue threshold of Γth = 680.1 J m−2 (Fig. 3d). Not limited to the polymer network, a network of percolated hard domains in the CPU deconcentrates stress to resist the propagation of fatigue crack. Incorporation of MXene nanosheets (CPU/M) resulted in an increase in Young’s modulus (Fig. 3b); however, the fatigue threshold remained at 728.5 J m−2, essentially equivalent to that of the CPU (Fig. 3d). In stark contrast, by further structuring the MXene nanosheets into a 3D-interconnected framework using PA molecules, the restructured CPU/MP composite with multiscale interlinked high-order structure achieved an unparalleled fatigue threshold of 8226.3 J m−2, which is 339.9 times higher than that of IPU. To validate such a record-high Γth value, we applied a G of 7933.0 J m-2 (i.e., λA = 2) to a pre-cracked specimen (Fig. 3e and Supplementary Fig. S22), observing no crack propagation, redirection, or failure over 10,000 cycles (approximating infinite cycles of loads; Fig. 3f and Supplementary Movie 1). Taken together, the multiscale interlinked high-order structure, composed of a nanoscale continuous dynamic hard phase and a microscale 3D-interconnected MXene framework, along with their robust interfacial force, stored more elastic energy and further deconcentrated stress across multiple scales, resulting in a significantly enhanced Γth value.

Interestingly, the multiscale interlinked high-order structure not only increased the Γth value but also slowed the progression of fatigue cracks. The dc/dN above the threshold quantifies fatigue resistance under significant loadings. The Paris law, expressed as dc/dN = AGm, was used to fit the curves of fatigue crack growth rate, where A and m are the constants42. The exponent m is indicative of the resistance to crack growth, determined by the slope of dc/dN versus G on a log-log scale. As shown in Fig. 3g, the calculated m value for IPU was 555.0, reflecting rapid fatigue crack propagation. By contrast, the m value for CPU was 2.9, and reducing further to 0.4 for the CPU/MP composite. This three orders magnitude reduction in the m value for CPU/MP underscores the synergistic effect of continuous dynamic hard phases and 3D-interconnected MXene frameworks in decelerating fatigue fracture under cyclic loadings. Typically, for applications demanding reversible deformation over numerous cycles, load-bearing ability is primarily governed by Young’s modulus and fatigue thresholds. We thus evaluated these properties across various reported viscoelastic materials in the literature. It was found that our structure-tailoring strategy optimizes the multiscale interlinked structure of the CPU/MP composite, markedly enhancing stiffness (Young’s modulus) and fatigue resistance (fatigue threshold) to levels approximately ~56.8 and ~339.9 times greater than those of the initial IPU elastomer, respectively. This approach effectively overcomes the conventional compromise between fatigue resistance and stiffness in elastic networks, surpassing most contemporary techniques that typically enhance only one parameter (Fig. 1a and Table 1).

To elucidate the concurrent enhancement in both stiffness and fatigue resistance, we investigated the crucial role of PA molecules in promoting the formation of a multiscale interlinked high-order structure optimized for load-bearing performance. Initially, DMA was conducted to demonstrate the formation of robust supramolecular hydrogen bonds following the introduction of PA molecules. The CPU/MP exhibited a higher glass transition temperature (Tg) and a lower loss factor (tanδ) compared to CPU/M (Supplementary Fig. S23). The polyhydroxy PA molecules offer numerous hydrogen-bonding sites, facilitating strong interactions among the hard segments of the CPU, the MXene nanosheets, and between different MXene nanosheets themselves. Consequently, the mobility of polymer chains was substantially restricted by these bonded nanosheets, thereby improving deformation resistance and enhancing the modulus43. This was further evidenced by the MD simulations shown in Fig. 2f, g, where CPU/MP displayed significantly higher cohesive energy density than CPU/M system, indicating a greater anti-deformation capability. Additional MD simulations were conducted to model changes in cohesive energy density and hydrogen-bond counts for CPU/MP and CPU/M after cyclic stretching at a preset strain (Supplementary Fig. S24). In contrast to CPU/M, where a gradual decrease in the numbers of hydrogen bonds and cohesive energy density was observed with increasing numbers of cyclic tensile tests, the CPU/MP composite system showed little change in hydrogen-bond counts and cohesive energy density under the same cyclic loading conditions (Fig. 3h). This suggests that the supramolecular cross-linkers within the CPU/MP network are more robust and resilient, remaining intact during cyclic stretching and thus preserving the high energy release rate of the high-order structures, significantly enhancing the fatigue threshold. The robust interactions facilitated by the abundant PA molecules also maintain the integrity of the multiscale interlinked high-order structures throughout thousands of cycles, a critical factor in resisting fatigue crack propagation. This premise is directly supported by the results of electrical conductivity measurements over thousands of cycles, where CPU/MP maintained nearly constant conductivity, in stark contrast to the significant decrease in conductivity observed in CPU/M (Fig. 3i and Supplementary Fig. S25).

Capitalizing on CPU/MP’s unique combination of high stiffness and exceptional fatigue resistance, we propose its potential application in robotic artificial ligament systems. Such components must simultaneously satisfy two critical criteria: sufficient modulus to withstand physiological loading and superior fatigue tolerance for long-term cyclic tensile operation44,45. To validate this, we conducted a biomechanical simulation using an elbow joint model (Supplementary Fig. S26), replacing both lateral and medial collateral ligaments with CPU/MP sheets containing a predefined 2-cm crack initiation site. Subsequent kinematic simulation driven by a universal testing machine (Supplementary Movie 2) demonstrated outstanding durability; after 3000 human-motion-mimicking flexion-extension cycles, the CPU/MP ligament maintained mechanical integrity with no observable crack progression (Supplementary Fig. S26a). This performance starkly contrasts with IPU counterparts, which exhibited catastrophic failure via complete crack penetration after merely 150 cycles (Supplementary Fig. S26b). Collectively, our findings demonstrate the significant potential of CPU/MP materials for next-generation robotic ligaments requiring both mechanical robustness and fatigue durability.

Thermomechanical stability

A key limitation of most self-healing materials is their rapid mechanical degradation and structural failure upon heating46,47,48,49. Exemplifying this limitation, self-healing IPU exhibited a sharp decline in storage modulus (E’) with a flow transition at merely 67.8 °C, as revealed by temperature-dependent DMA experiments (Fig. 4a). In contrast, CPU’s continuous hard phases, formed through multivalent UPy hydrogen-bond aggregation, significantly elevated its flow transition temperature (Tf) to 121.8 °C (Fig. 4a), accompanied by a relatively gradual E’ decrease during heating (Fig. 4b). CPU/M displayed an even higher Tf of 146.3 °C (Fig. 4a), though its E’ values followed a similar declining trend as CPU from 30–130 °C (Fig. 4b). Thus, while MXene incorporation restricted chain mobility and enhanced Tf, thermomechanical stability improvements remained limited. Remarkably, structural CPU/MP maintained relatively stable E’ at high temperatures, decreasing less than twofold from 30–130 °C and retaining > 10 MPa at 200 °C (Fig. 4a, 4b). This contrasts starkly with the orders-of-magnitude E’ reductions observed in IPU, CPU, and CPU/M under identical conditions (Fig. 4b). Furthermore, CPU/MP achieved a Tf of 210 °C, approaching the polymer chains’ thermal decomposition temperature (225 °C, Supplementary Fig. S27). More visual demonstrations of the enhanced thermomechanical stability are provided in Fig. 4c and Supplementary Movie 3, where a CPU/MP spline sustained a weight of 20 g for 35 s even when exposed to a flame (with a distance of 5 cm, where the temperature is around 400 °C). However, under the same conditions, IPU, CPU, and unstructured CPU/M showed rapid decay, with load-bearing times of only 5 s, 9 s, and 11 s, respectively.

Fig. 4: Thermomechanical stability.
figure 4

a Storage modulus and loss modulus versus temperature for the IPU, CPU, CPU/M, and CPU/MP materials. b Trends in storage modulus of the IPU, CPU, CPU/M, and CPU/MP materials at different temperatures, where E30 and ET represent the storage modulus at 30 °C and higher temperatures, respectively. c Photographs of IPU, CPU, CPU/M, and CPU/MP materials supporting a 20 g weight after flame exposure for 5, 9, 11, and 35 s, respectively. d Creep-recovery plots for the IPU, CPU, and CPU/MP materials at 65 °C under a constant stress of 20 kPa. e Stress-strain curves of the CPU/MP composite under cyclic stretching at 70 °C. f Plots of crack extension per cycle versus the applied energy release rate for CPU/MP at 70 °C. g Creep-recovery plots of the CPU/MP composite under different stress levels at a high temperature of 120 °C. h Shift factor as a function of temperature for IPU, CPU, and CPU/MP; the sold line represents Arrhenius equation fitting. Source Data are provided as a Source Data file.

Full size image

To quantitatively assess the enhanced thermomechanical stability imparted by the multiscale interlinked high-order structures, we conducted high-temperature creep experiments at 65 °C (close to IPU’s flow temperature) under 20 kPa constant stress. As shown in Fig. 4d, the IPU elastomer underwent rapid plastic deformation within 100 s due to inadequate percolating structures for creep suppression. Conversely, CPU elastomer, featuring a continuous hard phase, demonstrated significantly reduced creep elongation (εcreep = 4.7%) at 65 °C (Fig. 4d). Remarkably, incorporating an additional 3D-interconnected MXene framework via robust interfacial hydrogen bonding yielded CPU/MP with nearly negligible εcreep (0.8% over 100 s), representing a 15.5-fold reduction versus IPU. After stress removal, CPU/MP fully recovered its original length within 200 s without residual strain, outperforming both reference elastomers (Fig. 4d). These comparable results collectively validate the efficacy of multiscale interlinked architecture in enhancing thermomechanical stability. We investigated CPU/MP’s fatigue resistance around this creep temperature (Fig. 4e), where it demonstrated exceptional performance with a fatigue threshold of 965.5 J m⁻² at 70 °C (Fig. 4f), surpassing room-temperature values of most reported viscoelastic materials9,19,24,30. By contrast, pristine CPU underwent abrupt failure under identical fatigue conditions (Supplementary Fig. S28). Further probing of creep behavior at 120 °C (Fig. 4g), where IPU and CPU completely melted (Supplementary Fig. S29), revealed substantial CPU/MP creep strain (~ 9%) under 20 kPa stress despite maintained structural integrity (E’ > 20 MPa at 120 °C, Fig. 4a). Critically, partial residual strain persisted after 600 s recovery, indicating functional impairment. This thermomechanical limitation was corroborated by fatigue testing at 120 °C (Supplementary Fig. 30a), where progressive degradation occurred during cycling, with tensile stress decaying to ~0 MPa by the 2736th cycle. At 170 °C, CPU/MP suffered complete fatigue failure after merely 3 cycles (Supplementary Fig. 30b). Collectively, the multiscale interlinked structure enables CPU/MP to maintain functionality at 70 °C, which is a critical advantage for viscoelastic fatigue-resistant materials.

The enhanced thermomechanical stability originates from a high-density hydrogen bonding strategy with constrained spatial organization. Specifically, polyhydroxyl PA molecules serve as multifunctional bridges, interconnecting polymer chains (via -C=O motifs) and MXene nanosheets through optimized hydrogen-bond geometries. This forms three-dimensional hydrogen-bond clusters within topologically confined nanodomains (Supplementary Fig. S31). This architecturally confined network mirrors thermal stabilization phenomena in semicrystalline nylon and carboxymethyl cellulose, where geometrically ordered hydrogen-bond arrays elevate thermal transition temperatures. Crucially, the MXene-polymer interfaces exert nanoconfinement effects that suppress polymer chain dynamics at elevated temperatures while preserving structural continuity at the mesoscale. These dual mechanisms operate synergistically: hydrogen-bond clusters provide covalent-like thermal resistance, while the 3D-interconnected MXene framework restricts viscoelastic relaxation pathways. Using rheological master curves derived from the time-temperature superposition principle (Supplementary Fig. S32)50, we calculated the apparent activation energy (Ea) via linear fitting of the logarithmic shift factor (InαT) versus 1/T51. The structural CPU/MP composite exhibited an Ea of 82.7 kJ mol−1 (Fig. 4h), significantly higher than CPU (56.6 kJ mol−1), IPU (36.7 kJ mol−1), and closely aligned with dynamic covalent polymers52. This demonstrates that the multiscale interlinked high-order structure achieves temperature resilience comparable to dynamic covalent networks, effectively stabilizing the material network against thermal activation.

NIR-induced self-healing

Owing to non-covalent supramolecular hydrogen bonding, the multiscale interlinked high-order structures act as reversible crosslinkers, enabling dynamic self-healing in CPU/MP composites post-damage. A specimen (30 × 10 × 0.7 mm3) was severed, immediately pressed under mild manual pressure, and healed. Freshly fractured interfaces exhibited room-temperature self-adhesion, supporting 100 × its own weight (20 g vs. 0.2 g) after just 1 min contact (Fig. 5a and Supplementary Movie 4). However, CPU/MP showed slow healing kinetics at room temperature, sustaining only 3.5% strain after 24 h (Supplementary Fig. S33). The well-recognized limitations of extended healing times and low efficiency—compromising functionality during healing—restrict the applicability of self-healing materials in many scenarios53,54. Elevated temperature enhances polymer chain mobility, enabling temperature-regulated control of healing kinetics for tailored performance55,56. Scratch tests qualitatively assessed CPU/MP healing at elevated temperatures. Increasing the temperature from 100 to 140 °C significantly accelerated healing, fully restoring artificial scratches (Supplementary Fig. S34a). Healing time stabilized at 120 °C after an initial rapid decrease (Supplementary Fig. S34b), where a visible scratch became shallow and disappeared within 1 min (Fig. 5b). Furthermore, a larger hole defect (~ 100 μm diameter) and cross-cut injure (~100 μm width) were repaired within 1 min via local heating (Supplementary Fig. S35), while a complete cut achieved 96.8% healing efficiency under identical conditions (Supplementary Fig. S36). Effective polymer self-healing requires sufficient chain diffusion across damaged interfaces via a flow transition mechanism57,58,59. However, the bulk CPU/MP composite exhibited a flow transition relaxation time (τf) of several hours at 120 °C (Supplementary Fig. S37), which was orders of magnitude longer than the actual healing duration. This discrepancy may stem from the liquid-like behavior of freshly cut surfaces, where abundant free-end chains enable rapid healing through local interpenetration, as their relaxation time is significantly shorter than that of the bulk materials60,61. Atomic force microscopy (AFM) measurements at 120 °C support this mechanism, showing higher adhesion forces at fractured interfaces versus undamaged surfaces (Fig. 5c). However, a full understanding of the mechanism underlying the fast healing will require further investigations.

Fig. 5: NIR-induced self-healing and mechanism.
figure 5

a Photographs showing the instantaneous self-healing of the CPU/MP composite at 25 °C. b Optical microscopy images capturing the fast scratch-healing process of the CPU/MP composite at 120 °C. c AFM adhesion force for the surface layer and freshly fractured interface of CPU/MP specimens at 120 °C. d Schematic illustration of rapid thermal conduction in the 3D-interconnected MXene framework; the red arrow indicates the path of conductive heat transfer. e Schematic mechanism illustrating dynamic and reversible interfacial hydrogen bonding under NIR laser irradiation; the red and orange shadings denote the high and lower temperature zones, respectively, with the red arrow showing heat transfer across the interface. f Time-temperature infrared thermography images and corresponding photothermal curves of the CPU, CPU/M, and CPU/MP materials under 1.9 W cm−2 irradiation. g Puncture damage in the CPU/MP film caused by a needle, healed by NIR irradiation at 1.9 W cm−2 within 1 min. h Stress-strain curves of the original and healed CPU/MP samples. Source Data are provided as a Source Data file.

Full size image

Although thermal activation effectively accelerates healing, the required 120 °C temperature is often impractical in real-world settings. In contrast, light stimuli are particularly attractive because they can be applied remotely and instantaneously with high precision to the damaged regions, without affecting other areas62. As demonstrated in our previous work, MXene nanosheets exhibit optically induced resonance characteristics, allowing the conversion of near-infrared (NIR) light into thermal energy (Supplementary Fig. S38), which results in a localized temperature increase for rapid and precise NIR-triggered healing40. Consequently, the resulting CPU/MP composite can regulate its surface temperature on demand via NIR light irradiation. Intriguingly, the MXene nanosheets are assembled into a 3D-interconnected MXene framework within the CPU/MP composite, theoretically providing enhanced light-to-heat conversion efficiency and improved heat conduction (Fig. 5d and Supplementary Fig. S39). To validate this, the CPU/MP sample was irradiated with an intensity-modulated NIR laser (λ = 808 nm), and the photothermal conversion process was recorded using an infrared thermal camera (Supplementary Fig. S40). As shown in Fig. 5f, the central temperature of the CPU/MP rapidly increased within a few seconds under irradiation at 1.9 W cm−2, reaching a maximum local temperature of 125 °C, whereas the unstructured CPU/M control sample only reached 105 °C under the same conditions. These findings demonstrate that the CPU/MP exhibits a much higher photothermal heating rate than the CPU/M control sample due to its structured MXene framework. Moreover, the pristine CPU represented no evident temperature increase at the same experimental conditions. The surface temperature of CPU/MP can be precisely regulated by adjusting the output laser power (Supplementary Fig. S41), contributing to the outstanding NIR-responsive self-healing capabilities with high accuracy.

The abundant interfacial hydrogen bonds between the MXene-PA nanosheets and CPU chain hard segments enable the self-assembled 3D MXene framework to rapidly transfer NIR-induced heat to non-covalently linked polymer chains (Fig. 5e), which enhances chain mobility and segmental motions of dissociated hydrogen bonds for accelerating healing. To evaluate the NIR-responsive healing capability of the CPU/MP composite, we first introduced extreme puncture damage into a freestanding flexible film (100 μm thick), creating a hole defect. This damage was completely healed in 1 min through localized NIR irradiation (1.9 W cm−2) on the affected area (Fig. 5g). Subsequent assessment of NIR-triggered healing for complete cuts determined the self-healing efficiency (η), defined as the recovered integral area under stress-strain curves63. Following 1 min of NIR irradiation (1.9 W cm−2), the damaged CPU/MP specimen nearly fully restored its mechanical integrity, achieving η = 95.5% ± 1.4% (Fig. 5h). Furthermore, the fully healed CPU/MP composite exhibited a fatigue threshold of 8195.6 J m−2 after optimal 1 min healing (Supplementary Fig. S42). This rapid NIR-responsive healing rate combined with ultra-high fatigue threshold significantly surpasses values reported for self-healing materials requiring either thermal activation (e.g., the CPU in Supplementary Fig. S43) or prolonged healing times (typically exceeding 24 h).

Discussion

In summary, we engineer a viscoelastic self-healing composite material with high stiffness, high fatigue resistance, and enhanced thermomechanical stability by multiscale coupling of dynamic high-order structures via interfacial self-assembly. These structures outperform primary polymer networks in elastic energy storage, acting as high-energy constituents that arrest fatigue cracks through multiscale stress deconcentration. This elevates the CPU/MP fatigue threshold to 8226.3 J m−2, surpassing the Lake-Thomas limit. Simultaneously, the 3D-interconnected MXene framework coupled with a continuous hard phase enhances rigidity, achieving a Young’s modulus of 51.1 MPa—overcoming the stiffness-fatigue trade-off in polymer systems. The structure further ensures stable mechanical performance at 70 °C (fatigue threshold: 965.5 J m⁻²) and enables rapid NIR-triggered healing (< 1 min) via efficient photothermal conversion. Our study underscores the significance of tailoring multiscale structures in materials to achieve unprecedented mechanical and dynamic properties. These properties can be further enhanced by incorporating different nanosheets or nanoparticles to self-assemble into higher-order structures. In addition, the incorporation of such nanomaterials can introduce other functional attributes, such as electrical conductivity, thermal dissipation, magnetic actuation, and high dielectric constants, vastly expanding the possibilities for material design and application.

Methods

Materials

polycaprolactone (PCL, Mn = 2000 g mol−1), methylene-bis(4-cyclohexylisocyanate) (HMDI, 90%), dibutyltin dilaurate (DBTDL), 2,2-dimethylolbutanoic acid (DMBA), Ti3AlC2 (MAX) powder, and phytic acid (PA) were purchased from Shanghai Aladdin Biological Technology Co., LTD (China). Lithium fluoride (LiF ≥ 99.99%) and triethylamine were obtained from Sigma-Aldrich. HCl (37%), N, N-Dimethylformamide (DMF, 99.9%, extra dry), and acetone (AR) were purchased from Energy Chemical, China. All reagents were used as received without any further purification. The water utilized in all experiments was obtained from a Milli-Q water system, ensuring a resistivity of 18.0 MΩ·cm.

Synthesis of IPU elastomer

In a typical experiment, a mixture of PCL (1.5 mmol, 3 g) and HMDI (3.2 mmol, 0.879 g) were dissolved in 2 mL DMF with a Teflon magnetic stirrer in a nitrogen atmosphere at 80 °C. After allowing the reaction to proceed for 4 h, DMBA (1.5 mmol, 0.222 g) and DMF (5 mL) were added to the flask. Subsequently, the reaction temperature was lowered to 70 oC, and the reaction continued for an additional 24 h. Next, the reaction temperature was reduced to 55 °C, and acetone (10 mL) containing triethylamine (1.5 mmol, 0.152 g) was added to the flask to continue the reaction for 20 min. Following this, 10 mL of deionized water was slowly introduced into the flask. This addition resulted in a noticeable color change of the solution from colorless and transparent to white. The reaction continued for an additional 2 h. Subsequently, the resulting solution was extracted and placed in a dialysis bag for 48 h to obtain a waterborne IPU emulsion. Finally, the IPU emulsion was poured into a PTFE mold and placed in an oven at 50 oC for one day to obtain the final IPU elastomer.

Synthesis of CPU elastomer

In a typical experiment, a mixture of PCL (1.5 mmol, 3 g) and HMDI (3.8 mmol, 0.879 g) were dissolved in 2 mL DMF with a Teflon magnetic stirrer in a nitrogen atmosphere at 80 oC. After reaction for 4 hours, DMBA (1.5 mmol, 0.222 g), HMA (0.6 mmol, 0.1014 g), and DMF (5 mL) were added to the flask. Subsequently, the reaction temperature was lowered to 70 oC, and the reaction continued for an additional 24 h. Next, the reaction temperature was reduced to 55 °C, and acetone (10 mL) containing triethylamine (1.5 mmol, 0.152 g) was added to the flask to continue the reaction for 20 min. Following this, 10 mL of deionized water was slowly introduced into the flask, and the reaction continued for an additional two hours. Afterwards, the resulting solution was extracted and placed in a dialysis bag for 48 hours to obtain a waterborne CPU emulsion. Finally, the CPU emulsion was poured into a PTFE mold and placed in an oven at 50 oC for one day to obtain the final CPU elastomer.

Preparation of MXene and MXene-PA nanosheets

Two grams of MAX was slowly added to a mixture of 2 g of LiF and 40 mL of a hydrochloric acid (HCl) solution (9 mmol) at 35 °C under magnetic stirring for 24 h. Following this, the resulting suspension was washed five times with deionized water and centrifuged at 3500 rpm until the pH of the solution approached 7. Next, 40 mL of ethanol was added to the solution to facilitate the intercalation of MXene, accompanied by 1 h of ultrasonic treatment. After this step, the solution was centrifuged at 10,000 rpm for 10 min to obtain the bottom sediment. Subsequently, the MXene sediment was mixed with 20 mL of deionized water and subjected to vigorous shaking for 3 min, followed by ultrasonic treatment at 800 W for 20 min. After centrifugation for an additional 3 minutes at 3500 rpm, MXene dispersion was obtained, yielding a concentration of 4 mg mL−1.

For the preparation of MXene-PA nanosheet, phytic acid (PA) powder was added to the as-prepared Ti₃C₂Tx nanosheet solution at a concentration of 4 mg mL−1, with a weight ratio of PA to MXene set at 1:10. The resulting suspensions were then subjected to sonication at 800 W for 1 h.

Fabrication of CPU/MP composite

Firstly, 20 g of the CPU emulsion (with a solid content of 20%) was added to a PTFE beaker. Gradually, 60 mL of the MXene-PA solution, which has a concentration of 4 mg mL−1, was incorporated into the beaker over a period of 30 minutes while stirring continuously. After that, the mixture was stirred for an additional 5 hours. Once this stirring phase is completed, the resulting solution was poured into a PTFE mold. Finally, the product was left to dry for 48 h at 50 °C to obtain the CPU/MP composite.

Fabrication of control CPU/M composite

Firstly, 20 g of the CPU emulsion (with a solid content of 20%) was added to a PTFE beaker. Gradually, 60 mL of the MXene solution, which has a concentration of 4 mg mL−1, was incorporated into the beaker over a period of 30 min while stirring continuously. After the addition is complete, the mixture was stirred for an additional 5 h. Once this stirring phase is completed, the resulting solution was poured into a PTFE mold. Finally, the mold was left to dry for 48 h at 50 °C to obtain the control CPU/M composite.

General characterization

The 1H Nuclear Magnetic Resonance (NMR) spectra were recorded on a Bruker AVANCE III 500 MHz spectrometer at room temperature, using tetramethylsilane (TMS) as an internal reference. Attenuated Total Reflection Fourier-transformed Infrared Spectroscopy (ATR-FTIR) spectra were obtained with a Bruker Tensor 27 spectrometer equipped with a Specac Golden Gate ATR heating cell, covering the range of 4000 to 600 cm−1. UV-visible transmittance spectra were measured using a Thermo Fisher E220 spectrophotometer. X-ray diffraction (XRD) measurements were performed on a Bruker D8 Advance system. Transmission electron microscopy (TEM) was measured on a JEOL-100CX high-resolution transmission electron microscopy at an acceleration voltage of 200 kV. Thermogravimetric analysis (TGA) was conducted using a TA TGA-550 instrument, with a heating rate of 20 oC min−1 in a nitrogen atmosphere from 50 to 800 oC. Differential scanning calorimetry (DSC) measurements were carried out on a TA DSC-25 differential scanning calorimeter at the same heating rate of 10 oC min−1. Dynamic light scattering (DLS) measurements were performed on a Malvern Zetasizer Nano ZS instrument. X-ray photoelectron spectroscopy (XPS) was conducted using a PHI Quantera II X-ray photoelectron spectrometer operating at 15 kV. Infrared thermal images and temperature variations under 808 nm near-infrared (NIR) light irradiation were recorded using an A615 Infrared Thermal Imager. Atomic Force Microscope (AFM) images were captured in tapping mode using a Bruker Multimode 8. Small-angle X-ray scattering (SAXS) measurements were conducted on a Bruker NanoSTAR. Optical microscopy images were recorded with a Jiangnan MV3000 optical microscope. Gel permeation chromatography (GPC) analysis was performed in tetrahydrofuran (THF) using PolyPore columns from Agilent.

Mechanical characterization

Tensile tests were conducted using a universal testing machine (Shimadzu AGS-X) at room temperature (25 °C) with a strain rate of 100 mm min−1. The rectangular specimens for the tensile tests had dimensions of 10 mm gauge length, 5 mm width, and a thickness ranging from 0.5 to 1.2 mm. The Young’s modulus (Y) was determined by calculating the fitted slope of the initial linear portion of the stress-strain curve. The work of fracture (W) was defined as the area under the stress-strain curve, representing the total energy absorbed by the material up to the point of fracture.

To measure the Γth, we adopt the single-notched method. Cyclic tensile tests were conducted on both notched and unnotched rectangular samples, which measured 10 mm in gauge length, 5 mm in width, and had a thickness ranging from 0.5 to 1.2 mm. First, cyclic tensile tests were conducted on unnotched rectangular samples under a specified stretch λA (λ = strain/100 + 1). After N cycles of stretching, a stable stress-strain curve is obtained. From this stable stress-strain curve, the strain energy density (W) can be derived using the following formula:

$$W\left({{{\rm{\lambda }}}}_{{{\rm{A}}}},{{\rm{N}}}\right)={\int }_{{1}}^{{{{\rm{\lambda }}}}_{{{\rm{A}}}}}{{\rm{SD}}}{{\rm{\lambda }}}$$
(1)

where S and λ are the measured stress and stretch, respectively. The same cyclic stretch λA was applied to the cracked sample, which featured a pre-cut crack with a length of c = 1 mm. The total length of the cut in the cracked sample c(N) was measured in its undeformed state after N cycles of cyclic stretching. The applied energy release rate G in the notched sample during the Nth cycle of applied stretch λA can be calculated as follows:

$$G\left({{{\rm{\lambda }}}}_{{{\rm{A}}}},{{\rm{N}}}\right)=2{{\rm{k}}}\left({{{\rm{\lambda }}}}_{{{\rm{A}}}}\right) \cdot c(N) \cdot W\left({{{\rm{\lambda }}}}_{{{\rm{A}}}},N\right)$$
(2)

where k is a varying function of the applied stretch expressed as \(k\) = 3·λ−1/2; W denotes the strain energy density determined from the integral area of the stress-strain curve of an uncracked sample over N cycles of applied stretch λA. By systematically varying the applied stretch, a plot of crack extension per cycle versus the applied energy release rate can be obtained. By linearly extrapolating the curve of dc/dN versus G to the intercept with the abscissa, we can approximately determine the critical energy release rate Gc. This value signifies the threshold below which the fatigue crack will not propagate under an infinite number of loading cycles. By definition, the fatigue threshold Γth is equivalent to the critical energy release rate Gc. To assess the fatigue threshold of CPU/MP samples at elevated temperatures, we conducted cyclic tensile tests using a universal testing machine equipped with a constant temperature chamber. The methodology for calculating the fatigue threshold remains consistent with the above approach.

DMA tests were conducted using a TA Q800 instrument using a film tension clamp. The length of the specimens between the tensile clamps for DMA measurements was determined by the instrument itself, while the width and thickness were measured using a standard Vernier caliper. DMA temperature sweep experiments were conducted at a frequency of 1 Hz, with a scanning temperature range from − 70 °C to 50 °C and from 30 °C to 210 °C at a heating rate of 5 °C min−1, respectively. In addition, creep tests were carried out at a frequency of 1 Hz to evaluate the material response to applied stress (5, 10, 20 kPa) over time and its ability to recover once the stress is removed. The testing temperatures are 65 °C and 120 °C, respectively. Bulk rheological measurements were carried out on a TA DHR-1 Rheometer using a 20 mm parallel steel plate, with frequency sweeps performed at a rotational strain amplitude of 0.1%, varying the frequency from 0.1 rad s−1 to 100 rad s−1 in a temperature range from 100 °C to 220 °C.

Self-healing characterization

For the self-healing tests of the mechanical properties of CPU/MP, each specimen was first cut into two completely separate pieces using a razor blade. The freshly cut faces were then gently pressed together and allowed to heal at room temperature, elevated temperature, or under the irradiation of near-infrared (NIR) light, respectively. Subsequently, the self-healed specimens, after varying healing durations, were subjected to tensile tests at 25 °C with a strain rate of 100 mm min−1. The healing efficiency (η) was calculated based on the ratio of the integral areas under the stress-strain curves of the self-healed specimens compared to those of the intact specimens. This evaluation involved testing five specimens for each healing duration to determine the healing efficiency. The same self-healing conditions were applied uniformly across all tests, including those involving defects such as scratches and punctures.

Molecular dynamic simulation

Molecular dynamics simulation was carried out by BIOVIA Materials Studios 2017R2 (Accelrys Software Inc.). Computational results were obtained by using Dassault Systèmes BIOVIA software programs. BIOVIA Materials Studio was used to perform the calculations and to generate the graphical results. The Universal force field was adopted during the whole Forcite module computing processes at 298 K. In the MXene-MXene and MXene-PA-MXene-PA systems, the former consisted of two MXene nanoplates, whereas the latter incorporated two MXene nanoplates along with two PA molecules. The binding energy of MXene-MXene systems was evaluated by the following equation:

$${E}_{{binding}}={E}_{({MXene}-{Mxene})}-{E}_{{Mxene}}-{E}_{{MXene}}$$
(3)

Similarly, the binding energy of MXene-PA-MXene-PA systems can be represented in the same way.

Subsequently, we measured the cohesive energy density (cohesive energy per unit volume for the polymer-based materials) of the CPU/MP system and the CPU/M system to assess whether the incorporation of PA significantly enhances the internal energy of the composite. The CPU/MP system comprised five CPU polymer chains, specifically two PCL polymer chains and one hard segment containing UPy motif, in addition to two MXene nanosheets and two PA molecules. In contrast, CPU/M system consists of five CPU polymer chains, including two PCL polymer chains and one hard segment containing the UPy motif, along with two MXene nanosheets. The cohesive energy density of systems with periodic boundary conditions was calculated after 5 ns-dynamic equilibrium, utilizing NVT and NPT ensembles alternatively. The van der Waals and electrostatic interactions were handled by Ewald summation methods with an accuracy of 0.001 kcal mol−1. Finally, we utilized a Perl script to stretch the model and captured the corresponding frame when the strain reached 10%. Subsequently, compression was applied until the volume returned to its original state, followed by a tensile test. This process was repeated five times. Cohesive energy density (CED) and the H-bond counts were obtained from forcite analyses. Each value of Young’s modulus was achieved through the Constant strain method of Mechanical Properties.