Hostname: page-component-78c5997874-4rdpn Total loading time: 0 Render date: 2024-11-02T23:28:52.982Z Has data issue: false hasContentIssue false
  • English
  • Français

Brittle fracture in disordered materials: A spring network model

Published online by Cambridge University Press:  31 January 2011

W. A. Curtin  and
H. Scher
W. A. Curtin
Affiliation:
BP Research, 4440 Warrensville Center Road, Cleveland, Ohio 44128
H. Scher
Affiliation:
BP Research, 4440 Warrensville Center Road, Cleveland, Ohio 44128
Get access

Abstract

A model for investigating the influence of distributed disorder on the failure of brittle materials is introduced. The model assumes that microstructural features of a material can be represented by simple linear springs with a failure threshold, and that the entire material can be represented by a connected network of such springs. Distributed disorder is introduced by allowing spring-to-spring variations in spring characteristics such as the modulus and the failure strain. The conditions under which such a spring network model is valid for studying failure are discussed. The consequences of distributed residual stress disorder on macroscopic mechanical behavior are then studied using the network model, and a brittle to ductile-like transition in the stress-strain behavior is observed with increasing disorder. All the qualitative features of the network results can be described theoretically by a statistical analysis of this problem. Finally, notch tests are performed to evaluate the strength and toughness of the ductile-like materials as compared to the uniform (no disorder) material, and the ductile-like material is found to have (i) a larger work of fracture, (ii) comparable strength in the presence of processing flaws, and (iii) the possibility of larger toughness. Based on these results, the possibility of observing such ductile-like behavior in real composite materials is discussed.

Type
Articles
Information
Journal of Materials Research , Volume 5 , Issue 3 , March 1990 , pp. 535 - 553
Copyright
Copyright © Materials Research Society 1990

Access options

Get access to the full version of this content by using one of the access options below. (Log in options will check for institutional or personal access. Content may require purchase if you do not have access.)

References

REFERENCES

1Tvergaard, V. and Hutchinson, J.W., Am, J.. Ceram. Soc. 71, 157 (1988).CrossRefGoogle Scholar
2Ruhle, M., Claussen, N., and Heuer, A. H., Am, J.. Ceram. Soc. 69, 195 (1986); M. Ruhle, A.G. Evans, R.M. McMeeking, P.G. Charalambides, and J.W. Hutchinson, Acta Metall. 35, 2701 (1987).CrossRefGoogle Scholar
3Marshall, D. B., Am, J.. Ceram. Soc. 69, 173 (1986); A. H. Heuer, M. J. Readey, and R. Steinbrech, Mater. Sci. Engr. A 105/106, 83 (1988).Google Scholar
4Faber, K.T., Iwagoshi, T., and Ghosh, A., Am, J.. Ceram. Soc. 71, C-399 (1988).CrossRefGoogle Scholar
5Griffith, A. A., Philos. Trans. Soc. London A 221, 163 (1920).Google Scholar
6Lawn, B. R. and Wilshaw, T. R., Fracture of Brittle Solids (Cambridge University Press, Cambridge, England, 1975).Google Scholar
7Evans, A.G., Adv. in Ceram. 21, 171 (1976).Google Scholar
8Budiansky, B., Hutchinson, J.W., and Lambropoulos, J. C., Int. J. Solids Structures 19, 337 (1983).CrossRefGoogle Scholar
9Faber, K.T. and Evans, A.G., Acta Metall. 31, 565 (1983).CrossRefGoogle Scholar
10 A similar lattice model has been employed by Sahimi, M. and Goddard, J. D., Phys. Rev. Lett. 33, 7848 (1986) and very recently in Refs. 15, 16, and 19.Google Scholar
11 Note that the criterion for fracture used here is based on exceeding a critical maximum stress within a discrete element, with no consideration of the subsequent elastic energy gained upon the relaxation of the remaining network. Fracture does not appear to be based, therefore, on the equilibrium energy approach of Griffith. However, the Griffith criterion could be preserved on a smaller scale since the hypothetical tensile test performed oneach element does not actually specify how failure of the element occurred. Failure may have initiated by the Griffith mechanism, with essentially all of the elastic energy released within the elementitself (which is valid if the initial defect is much smallerthan the element size), and then the network relaxation energy is simply the excess energy released on further unstable propagation of the initial defect up to the element size.Google Scholar
12Dienes, G. J. and Paskin, A., J. Phys.Chem. Solids 48,1015 (1987); W. A. Curtin, unpublished research.CrossRefGoogle Scholar
13 The elastic properties of the triangular network of nonfailing springs with distributed moduli a have been studied extensively in recent years in connection with the elastic percolation problem [see S. Feng and P. B. Sen, Phys. Rev. Lett. 52, 216 (1984), for example].If a fraction 1-p of the springs are assigned modulus a = 0 and placed at random in the network, with the remaining fraction p of the springs having a nonzero α (usually all identical), the elastic moduli of these highly defective systems goes to zero at a critical value of p, estimated ai pc = 0.65, despite the fact that the network is entirely connected. Including additional spring-angle dependent forces into the network shifts this percolation point down to pc = 0.33, where the network does become disconnected. Thus, highly defective triangular networks with spring forcesonly do have some unusual mechanical properties, but they will not generally be manifested in the fracture problems studied here. There are other lattice effects appearing in low-defect networks and we discuss them briefly in the accompanying paper.Google Scholar
14 The algorithm used here for evolution of the material with increasing strain is entirely deterministic. That is, the distributed disorder present at the outset of the simulation (ε = 0) determines the subsequent behavior (at ε > 0) completely. The choice of which spring to break at any point is clear and unique, with no probabilistic aspect to it. Such a deterministic algorithm is appropriate in situations where the distributed disorder dominates any possible effects due to thermal fluctuations. This is most often the case, since the springs represent material regions large compared to atomic dimensions with failure occurring because of local high stress, stiffness, or weakness. A probabilistic algorithm [Y. Termonia and P. Meakin, Nature 320, 429 (1986)] should be used only when the mechanical properties of every region are essentially identical so that only fluctuation effects distinguish the different regions.+0)+completely.+The+choice+of+which+spring+to+break+at+any+point+is+clear+and+unique,+with+no+probabilistic+aspect+to+it.+Such+a+deterministic+algorithm+is+appropriate+in+situations+where+the+distributed+disorder+dominates+any+possible+effects+due+to+thermal+fluctuations.+This+is+most+often+the+case,+since+the+springs+represent+material+regions+large+compared+to+atomic+dimensions+with+failure+occurring+because+of+local+high+stress,+stiffness,+or+weakness.+A+probabilistic+algorithm+[Y.+Termonia+and+P.+Meakin,+Nature+320,+429+(1986)]+should+be+used+only+when+the+mechanical+properties+of+every+region+are+essentially+identical+so+that+only+fluctuation+effects+distinguish+the+different+regions.>Google Scholar
15 In the absence of inhomogeneities, the elements may represent any length scale. But once failure occurs at one element, completely brittle macroscopic failure usually follows since each subsequent element at the crack tip will exceed the failure stress required for cracking. This scale independence is also basically relevant for the case where only “cracks” (i.e., missing elements or elements with zero modulus) are present as the inhomogeneities, with brittle failure initiating at that element under the higheststress due to stress concentration from some most-critical defect configuration. Thus, the nice work of Srolovitz and Beale,16 based on an electrical analog studied by Duxbury et al.,17 using a network model of springs with a distribution of α = 0 defects only, need not be motivated by microstructural considerations. In contrast, the recent work of Kahng et al. on the electrical analog of a distribution of failure strains (a distribution of failure currents in a network of resistive fuse elements) is motivated incorrectly.18 Kahng et al. motivate a distribution of failure currents by conceptually performing a coarse-graining of the Duxbury et al. system of randomly blown fuses: blocks of the fuse network, containing within them various configurations of blown fuses, are regarded as single effective fuses, and block to block variations generate a distribution of critical currents in a “renormalized” or rescaled network of effective fuses. This hypothetical procedure misses a major point of the Duxbury et al. work; namely that the macroscopic failure is dependent on one most critical defect configuration no matter where it appears in the network nor which effective fuse contains it upon coarse-graining. The essentially arbitrary boundaries of the coarse-graining procedure do not arrest the most-critical defect once it begins to propagate. Stress concentration on a much smaller length scale than the arbitrary coarse-graining scale is the crucial feature, and it cannot be accounted for properly for coarse-graining. Kahng et al.'s results are, of course, interesting and potentially useful for considering the mechanical analog of a distribution of failure strains on an appropriately considered length scale. But to reiterate, in brittle failure the retention of the intrinsic disorder on the smallest relevant length scale is the key to the problem.Google Scholar
16Beale, P. D. and Srolovitz, D. J., Phys. Rev. B 37, 5500 (1988).CrossRefGoogle Scholar
17Duxbury, P. M., Leath, P. L., and Beale, P. D., Phys. Rev. B 36, 367 (1987).CrossRefGoogle Scholar
18Kahng, B., Batrouni, G. G., Redner, S., deArcangelis, L., and Herrmann, H. J., Phys. Rev. B 37, 7625 (1988).CrossRefGoogle Scholar
19 This behavior may be compared to that which prevails in the case of distributed breakdown strengths. There, all nonlinearity is due to accumulated damage with no release of residual strain and so at any point on the macroscopic σ-ε curve, the effective modulus is precisely σ/ε. Therefore, extensive nonlinear regimes in the distributed εc case must be accompanied by considerable modulus reductions.Google Scholar
20Herrmann, H. J., Hansen, A., and Roux, S., Phys. Rev. B 39, 637 (1989); L. deArcangelis and H. J. Herrmann, Phys. Rev. B 39, 2678 (1989).CrossRefGoogle Scholar
21Rice, J. R., URI-Winter Study Group, Santa Barbara, CA, 1989.Google Scholar