Nishan
Parvez
and
Catalin R.
Picu
*
Department of Mechanical, Aerospace and Nuclear Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180, USA. E-mail: picuc@rpi.edu; Tel: 1 518 276 2195
First published on 29th November 2022
Network materials with stochastic structure are ubiquitous in biology and engineering, which drives the current interest in establishing relations between their structure and mechanical behavior. In this work we focus on the effect of connectivity defined by the number of fibers emerging from a crosslink, z, and compare networks with identical (z-homogeneous) and distinct (z-heterogeneous) z at the crosslinks. We observe that the functional form of strain stiffening is z-independent, and that the central z-dependent parameter is the small strain stiffness, E0. We confirm previous results indicating that the functional form of E0(z) is a power function with 3 regimes and observe that this applies to a broad range of z. However, the scaling exponents are different in the z-homogeneous and z-heterogeneous cases. We confirm that increasing z across the Maxwell's central force isostatic point leads to a transition from bending to axial energy storage. However, we observe that this does not necessarily imply that deformation becomes affine in the large z limit. In fact, networks of fibers with low bending stiffness retain a relaxation mode based on the rotational degree of freedom of the crosslinks which allows E0 in the large z limit to be smaller than the affine model prediction. We also conclude that in the z-heterogeneous case, the mean connectivity is sufficient to evaluate the effect of connectivity on E0 and that higher moments of the distribution of z are less important.
The small strain modulus, E0, of network materials composed from fibers with non-zero bending and axial stiffness is controlled by the fiber density, ρ (total length of fiber per unit volume), the relative importance of the bending and axial stiffnesses, characterized by parameter (EfIf and EfAf are the bending and axial rigidities of fibers), and the connectivity index, z, which represents the number of filament segments emerging from a crosslink. It has been shown that low density networks of floppy filaments (small lb) deform non-affinely and store most of the strain energy in the bending mode of the fibers, while dense and densely crosslinked networks, with filaments rigid in bending, deform almost affinely and store the strain energy in the axial mode of fibers.4,15,16 This non-affine-to-affine transition reported for network with low z may be characterized using the structural parameter w = log10ρx−1l2b, large values of w corresponding to the affine deformation regime. The exponent x takes the value of 2 for cellular networks like the open cell foams and Voronoi structures,17 3 for fibrous networks in 3D18 and collagen networks,19,20 and takes larger values in 2D.21,22
Fiber networks may be unstable (vanishing stiffness) at small strains in certain conditions. Maxwell determined that a central force network is unstable (floppy) when the number of constraints the structure is subjected to is smaller than the number of degrees of freedom.23 This may be written in terms of the connectivity index as z < zc = 2d, where zc represents the connectivity at the central force isostaticity point (CFIP) and d is the dimensionality of the embedding space (we note that for a finite size model, the Maxwell counting argument leads to zc = 2d − 6(d − 1)/Nx, where Nx is the number of crosslinks in the model; the models used here have Nx > 20000 and hence zc is closely approximated by 2d). Virtually all engineering and biological networks have connectivity much smaller than zc = 6; in fibrous networks a crosslink is formed by two fibers in contact and hence z = 4, cellular networks of Voronoi type in 3D have z = 4, and crosslinks in collagen-based connective tissue are formed by the separation of fiber bundles into (typically two) sub-bundles and hence z = 3. Such networks have finite small strain stiffness due to the stabilizing effect of the fiber bending stiffness and/or the presence of pre-stress,24–27 as captured by the updated stability criterion of Calladine.28 Therefore, if filaments have finite bending stiffness and the crosslinks transmit forces and bending moments, the network has non-vanishing stiffness when z ≤ zc.
Interestingly, it has been shown that in networks with non-zero axial and bending stiffness, CFIP remains a critical point associated with diverging strain amplitude fluctuations and associated correlation length.24,29,30 Specifically, while the emergence of non-zero stiffness as the network density increases takes place at the stiffness percolation threshold, which corresponds to zp ≈ 2.6 for these networks,13,27,31 the non-affinity measure diverges at zc. In the hypostatic range zp < z < zc the behavior is of the type described in the preceding paragraph, i.e., it is bending dominated and non-affine when w is small and becomes axially dominated and approximately affine as w increases. In the hyperstatic regime z > zc, the axial deformation mode becomes energetically dominant.24,30
Most studies investigating the effect of connectivity on network behavior were performed with lattice-based models in which struts were removed with specified probability to adjust the mean z of the network.24,29,30 This procedure modifies ρ at the same time and renders z spatially non-uniform. While in such networks the connectivity parameter is described by a distribution, it is generally conjectured that all above arguments hold provided the mean connectivity, , is used as parameter in place of z.
In this work we revisit the effect of z on the small strain stiffness of networks by comparing structures in which all crosslinks have the same z, which we denote as z-homogeneous, with z-heterogeneous networks constructed by modifying a fraction of the crosslinks of an initially z-homogeneous network with z = 4 to z = 8. This fraction is varied to adjust . Here and in the following, z and
always refer to z-homogeneous and z-heterogeneous cases respectively. In both cases we investigate hypo- and hyperstatic cases and seek the relation between E0, w and z (or
in z-heterogeneous networks). We recover the behavior previously reported for z-heterogeneous networks24,29,30 and confirm that
is indeed a sufficient descriptor for such structures. However, we observe that the scaling exponents relating z (or
) to E0 are strongly dependent on the type of structure considered. Further, we determine that the effects of z and w on the degree of deformation non-affinity are not equivalent. A network of large w is essentially insensitive to increasing z since its deformation is already approximately affine. However, a network of small w is not rendered affine by increasing z. To demonstrate this, we introduce a measure of the rotational non-affinity of the crosslinks and use it along with the commonly used measure of translational non-affinity. We determine that in the small w range, increasing z (or
) into the hyperstatic range decreases the translational non-affinity, but the rotational non-affinity remains large. This indicates that hyperstatic networks relax relative to the perfect affine deformation and hence their stiffness is smaller than the stiffness of the purely affine regime reached at large w. We also investigate the large deformation response of networks with low w for both hypo- and hyperstatic conditions and confirm that the functional form of strain stiffening is independent of z,24,29 but high z structures, which have much larger E0 than the low z structures of same w, undergo a structural instability before the onset of the non-linear, stiffening regime.
The types of networks and models considered are presented in Section 2, the relation between stiffness and network parameters is discussed in Section 3.1, followed by an evaluation of the degree of non-affinity and energy partition (Section 3.2) and of the large deformation response (Section 3.3).
To generate z-homogeneous networks with z > 4, fibers are added to the initial network by fulfilling two constraints: (i) the crosslinks connected by an added fiber are separated in the graph space of the original network by no more than 3 edges, and (ii) the length of added fibers should be smaller than 2.5lc in real space. Further, fibers are added to connect nearest neighbor crosslinks first. These conditions ensure that the mean segment length of the resulting network is within 6% to that of the original network, lc. This method prevents the creation of very long fibers which may trigger non-local effects and allows separating the effect of connectivity from that of non-local interactions. Fibers are added iteratively to reach the target coordination number, after which the modified network is trimmed to the intended size, L, such to ensure that the boundary regions have density comparable to the interior. This procedure leads to z-homogeneous networks in which at least 90% of the crosslinks have exactly the desired z for all target connectivity values considered (z = 4…10). The interior crosslinks (not located along the boundary of the model) have mean connectivity marginally different from the target. For example, with target z = 6, the network we generate has z = 5.97 for the interior crosslinks. Fig. 1a shows a 2D schematic of the procedure used to select crosslinks to be connected by additional fibers and Fig. 1b shows a sectioned 3D network with z = 7 produced by this procedure.
The same procedure is applied to generate z-heterogeneous networks. z is increased to a value of 8 at a target fraction, f, of spatially uncorrelated crosslinks selected through uniform random sampling. f varies from 0.05 to 0.90. Hence, z-heterogeneous networks have f fraction of crosslinks with z = 8 and 1 − f fraction of crosslinks with z < 8. This creates heterogeneity within the network which is described by a characteristic length scale equal to the mean distance between crosslinks with z = 8, approximated by , where Nx is the total number of crosslinks in a model of edge length L. To avoid size effects, models with L\ls ≈ 35 are used for z-heterogeneous networks with f > 0.10 which translates to 40 < L/lc < 85 for these models. For all homogeneous networks considered, L/lc ≈ 35.
Fibers are considered athermal and are represented as beams. The crosslinks transmit both forces and moments and are rigid welds (the angle between fibers is not allowed to change during deformation). The model is discretized using beam finite elements32 (Timoshenko beam element, B32 in Abaqus) and the commercial package Abaqus Standard (2022)33 is used to obtain the solution.
The boundary conditions represent uniaxial tension. Displacements are prescribed for one face of the cubic model in the direction normal to that face (loading direction), while zero normal displacements are prescribed to the nodes on the opposite face. The other degrees of freedom of the loaded faces are left free. The lateral model faces are kept planar. To this end, the nodes on each of the lateral faces are kinematically coupled and remain coplanar, although the respective plane is free to move in its normal direction such that tractions on these faces vanish, in average, and the model is free to contract in the direction transverse to loading. The rotational degrees of freedom of the nodes on the lateral faces are left free.
For the results presented in this study, approximately 600 realizations are used, with 6 samples for each homogeneous and 3 for each heterogeneous network specification (given z and lb). The homogeneous network models have between 50000 and 122
000 fibers, function of the average connectivity. The total number of degrees of freedom per model ranges from 500
000 to 1
000
000. For reasons outlined in the previous paragraphs, z-heterogeneous networks of average connectivity
are much larger than z-homogeneous networks with z =
. The nominal stress is used in all calculations and only one component of this tensor (normal stress in the loading direction, S) is non-zero.
Fig. 2a shows the master plot for z-homogeneous networks with z ranging from 4 to 10. Hypostatic networks exhibit the two regimes I and III described in the previous paragraph. The curves corresponding to networks of increasing z shift gradually to the left in the z < zc = 6 range.
![]() | ||
Fig. 2 (a) Normalized stiffness, E* = E0/ρEfAfvs. the structural parameter w for z-homogenous (filled symbols) and z-heterogeneous (open circles) networks. (b) Collapse of the data in panel (a) based on eqn (1). The legend in (a) applies to both panels. The three regimes are highlighted in (a) by color shades and labeled as defined in text. |
Networks with z > zc behave differently in the low w range: the stiffness is z-dependent, but essentially w-independent (regime II). As z increases above zc, E0 exhibits a large jump from regime I to II. The asymptote of E0 at constant w as z increases is lower than the affine limit of the stiffness reached for w > −1, in regime III. This implies that, if w is sufficiently small, the network retains a relaxation mechanism which is inactive in regime III; this is discussed further in Section 3.2. It is instructive to recall the expectation for E0 in the affine limit of regime III: E0 ∼ ρEfAf. This relation results by observing that fibers are loaded axially, and deformation is approximately affine. The stiffness depends on the total fiber length per unit volume and not on how fibers are connected. This justifies the z-independence of the curves in regime III.
The data in Fig. 2a is collapsed using the relation:24,29
E*|Δz|−h = Φi(ρlb2|Δz|−g) = Φi(10w|Δz|−g), | (1) |
Three sets of z-heterogeneous networks are constructed as described in Section 2. Each set has specific lb and w values and the mean connectivity varies within each set from close to 4 to close to 8 as the fraction f of crosslinks with z = 8 increases from 0.05 to 0.9. Results for these systems are shown in Fig. 2 with open circles. Data for given set does not align perfectly along a vertical line in Fig. 2a due to the minor variation of the density with increasing fraction f. No jump is observed in this case as
increases past zc = 6. The data collapse based on eqn (1) leads to exponents h = 3 and g = 4, which are different from those obtained for the z-homogeneous networks. The collapse of such data for z-heterogeneous networks was reported in the previous literature24,29,30 and the present results are in agreement with the respective reports, although the values of h and g are different. This points to the dependence of these exponents on the structure and possibly the size of the networks considered.
![]() | ||
Fig. 3 Fraction of axial energy in z-homogeneous (filled symbols) and z-heterogeneous (open symbols) networks of selected w subjected to small deformations. |
This z-controlled energy storage transition was observed before, and it was interpreted as an indication that deformation becomes affine with increasing connectivity.4 This interpretation emerges by analogy with the similar bending-to-axial transition controlled by w which, indeed, is associated with a gradual reduction of the degree of non-affinity. However, this analogy does not apply to the z-controlled transition. A suggestion that this might be the case is provided by the observation made in Section 3.1 in relation to Fig. 2a, that the asymptotic values of E0 in the low w regime as z increases (regime II) is lower than the asymptotic value obtained in regime III as w increases. To clarify the nature of the relaxation mode leading to this effect, two non-affinity measures are evaluated here characterizing translational and rotational degrees of freedom of the crosslinks. The translational non-affinity is evaluated as:
Γt = 〈|u−ua|〉/lc, | (2) |
The second measure introduced here quantifies the magnitude of the crosslink rotation and is evaluated as:
Γr = 〈θi〉i, | (3) |
Fig. 4 shows the variation of Γt and Γr with w and z in the small deformation range for all homogeneous networks considered. Γt exhibits features previously reported:36,37 (i) in the hypostatic range, the non-affinity increases as w decreases and the rate of increase is smaller in the low w range, (ii) Γt exhibits a peak at z ≈ zc in the low w range, and (iii) Γt is independent of z in the large w range. This correlates well with the shift from the bending to the axial deformation mode controlled either by w or by z (at low w). We note that in the previous literature,5,13,24 a peak of Γt for non-affine networks is reported to occur exactly at the CFIP threshold, while in the present data the peak is shifted to slightly larger values. This apparent discrepancy emerges since here we report z as the network internal connectivity, ignoring surfaces. Accounting for the low connectivity of surface crosslinks would reduce slightly the effective z but would also render the z of z-homogeneous networks a fractional number, which is not representative for the present networks.
Γ r leads to additional conclusions: (i) Γr increases as w decreases, (ii) Γr increases as z increases across the transition defined by zc and remains large in the hyperstatic regime, and (iii) Γr is independent of z in the large w range. Therefore, in the hyperstatic, low w regime II of Fig. 2a the network relaxes by crosslink rotation even though the translational non-affinity is inhibited by the high connectivity. Interestingly, this relaxation mode which necessarily involves fiber bending, does not lead to bending energy dominance. Rotational non-affinity is sufficient to allow the reduction of E0 relative to the prediction of the affine model, but it is not large enough to change the energy storage mode of the network.
To gain further insight into the mechanics underlying the rotational non-affinity measure of eqn (3), refer to Fig. 4c. This 2D schematic shows a generic fiber of crosslinks M and N in the undeformed (dashed lines) and deformed states (continuous lines). The deformed configuration is shifted such to remove the translation of crosslink M and hence, the displacement of N is uN − uM. Let us assume that the deformation is translationally affine, i.e.uN − uM = (F − I)rMN, where F is the deformation gradient of the respective macroscale deformation and rMN is the position vector of N relative to M. Two possibilities exist: (i) the deformation is affine only at (and above) the scale of the crosslinks while the points along fiber MN may take any position, which is not constrained by F, and (ii) the deformation is strictly affine at all scales, which implies that all points of MN move as defined by F and MN remains straight. In case (i), the strain energy of the network subjected to affine crosslink displacements may be reduced by fiber bending and crosslink rotation. The softer the bending mode (lower w), the more pronounced this relaxation is. This explains the observation that rotational non-affinity is pronounced in the low w range of hyperstatic networks in which the high z values force a reduction of the translational non-affinity. In case (ii), crosslink rotation is entirely defined by the affine displacements (Fig. 4c) and hence in the large w regime, where the bending mode is inhibited, the rotational and translational non-affinity measures vary in proportion.
Further insight into the effect of the connectivity on energy partition and the associated stiffening may be obtained by examining the z-heterogeneous networks. As described in section 2, these structures are generated by starting with a Voronoi network with z = 4 at all crosslinks and transforming a fraction f of the total number of crosslinks to z = 8 via enriching the local neighborhood of these crosslinks with additional fibers. In the dilute limit, e.g., for f = 0.05, the crosslinks with z = 8 are isolated within the network with z ≈ 4, i.e., almost all fibers emerging from a z = 8 crosslink have z = 5 crosslink at the other end. Fig. 5a shows the average energy partition for the neighborhood of crosslinks with z = 8 and with z = 4 in networks with various f and , and with w ≈ −7.5. Interestingly, all fibers deform in the bending mode in the dilute limit, up to f ≈ 0.2, regardless of the connectivity at the crosslinks. For 0.2 < f < 0.4, fibers associated with z = 8 crosslinks deform strongly in the axial mode, while those bounded at least at one end by crosslinks with z = 4 deform in the softer bending mode. As f increases (f > 0.4), the probability for a fiber emerging from a z = 8 crosslink to have another z
4 crosslink at the other end increases. At the same time, the energy partition of the neighborhood of crosslinks with z = 8 (considering only the fibers emerging from such node) shifts to axial. Two reasons can be identified for this behavior in relation to Fig. 5b. First, fibers with higher connectivity nodes tend to be axially dominated. Secondly, the rapid change near f ≈ 0.30 for all fibers suggests that stiffening in the broader neighborhood (mean field) has also an effect. This is suggested by the data in Fig. 5b which shows that the energy partition of a fiber with given z at the two crosslinks shifts to axial as f increases, i.e., it depends on the broader neighborhood and not just on its own connectivity.
![]() | ||
Fig. 5 (a) Average fraction of axial energy of the neighborhood of crosslinks with z = 8 and z = 4. (b) Variation of the fraction of axial energy with f for fibers with specified z at the two ends. |
Fig. 6b shows the evolution of energy partition during large deformations for the networks in Fig. 6a. In the small strain limit (λ → 1) the fraction of energy stored in the axial mode increases as z increases from the hypo- to the hyperstatic regime, in agreement with the data in Fig. 3. The hyperstatic network shifts from the axial to the bending mode as it goes through the intermediate state labeled by I in Fig. 6c. This variation of the energy partition supports the suggestion made above that large structural re-organization/alignment takes place as the network moves from regime A to regime B. Both hypo- and hyperstatic networks deform in the bending mode up to a stretch of ≈1.15, after which the axial fraction increases. The increase of the axial fraction at stretches above 1.2 is due to the emergence of stress paths, as typically observed in hypostatic networks.38,39 Interestingly, the hyperstatic networks gain sufficient kinematic freedom upon the instability to undergo the structural re-organization required to produce stress paths.
This journal is © The Royal Society of Chemistry 2023 |