Вы находитесь на странице: 1из 43

J. Phys. Chem. Solids Vol. 48, No. Printed in Great Britain.

II,

pp. 111~11.57,

1987

CK22-3697/87 iQ 1987 Pergamon

$3.00 + 0.00 Journals Ltd.

FRACTURE

OF CERAMICS
D. R. CLARKE

AND

GLASSES

Thomas J. Watson Research Center, IBM, Yorktown Heights, NY 10598, U.S.A.

and K. T. FABER Department of Ceramic Engineering, Ohio State University, Columbus, OH 43210, U.S.A.
Ah&act-The rich variety of fracture behavior exhibited by glasses and ceramic materials is reviewed with particular emphasis on the understanding gained through the use of deliberately introduced, controlled cracks. After a brief summary of the mechanics of indentation cracks four major topics are discussed, the structure of crack tips, environment assisted crack growth, high temperature fracture and the toughening of ceramics. Resolution of the sharp vs blunt crack dilemma is presented together with recent microscopy observations of crack tips in brittle solids. In describing fracture in polycrystalline ceramics we explore some of the complexities beyond the simple Griffith behavior relating strength to flaw size, and show how the scale of the microstructure with respect to the crack length atfects the observed toughness. It is shown that the interaction of a crack with the microstructure provides a unifying theme for interpreting much of the current work in the literature and leads to important concepts discussed here, such as the discrete-continuum transition, R-curve behavior, toughening due to crack deflection and crack bridging, transformation toughening and stress-induced microcrack toughening.
Keyword:

Fracture, ceramics, glasses, indentation, cracks, R-curve, toughening, microstructure.

I. INTRODUCTION

In the last decade a considerable effort has been devoted around the world to the development of new and improved ceramics and glasses. Much of this endeavor has been directed to improving the strength, fracture toughness and mechanical reliability of these materials for purely practical purposes. With it has come a considerable advancement in the understanding of the fracture of brittle materials particularly from the perspective of the physics and chemistry of fracture. Because brittle fracture is so inextricably bound up with the materials aspects, and because the development of the central ideas has been rapidly adopted in the microstructural design of tougher ceramics, the focus of this contribution is on aspects of brittle fracture that have proven central to an understanding of the fracture of the wide range of ceramic materials. Ceramics, including glasses, encompass an enormous range of different materials. In the last 10 or 15 yr alone, a number of important new classes of ceramic have been developed many of which exhibit rather individual fracture behavior. This includes the transformation toughened zirconia-based ceramics [l], the refractory silicon nitride and Sialon ceramics [2], the fl-alumina and related solid electrolytes [3], the polyphase nuclear waste ceramics [4], the electronic substrate and packaging ceramics [S], glassceramics [6], fiber reinforced ceramics [7], and most recently, the new high T,, superconducting oxide

ceramics [8]. All of these materials are polycrystalline and have distinguishing microstructures which can be altered by processing. It is possible, and increasingly practical, to design a complex microstructure for the specific purpose of increasing the fracture resistance of the material. In consequence, one of the recurrent themes of this contribution is how the microstructure renders the materials elastically (and, in some cases, plastically or viscously) inhomogeneous over some characteristic length scale, and how this affects the fracture toughness of the material and its crack length dependence. Much of the new insight into the brittle fracture process, and how the microstructure effects crack propagation, has come from the use of controlled flaws [9, lo]. Progress in understanding fracture had been plagued for many years by the statistical nature of strength resulting from the intrinsic flaw population of a polycrystalline material. Starting with the work of Weibull [l 11, and making use of the developments in the mathematics of extreme value statistics, considerable progress has been made in incorporating this statistical nature into the characterization and design with brittle materials. And while this approach has had impressive success, it has added little to our understanding of the physics of the fracture process itself. This has changed with the use of a deliberately introduced crack, of well-characterized dimensions, into experiments in which the behavior of the crack could be monitored under controlled test conditions [9]. In such experiments, the flaw, by

1115

1116

D. R. CLARKEand K. T. FABER U, of the system, including the crack, consisted of two components, that produced by the applied loading, U, and the work to create fresh surface, U,. The applied loading energy represents the potential energy of the system to do work, to extend the crack, and is given by U, = (U, - W), where LIE is the elastic strain energy stored in the system and W is the work done by the system. For the geometry of a through crack in a thin sheet of material of unit thickness, the individual terms can be simply stated: the term Us is equal to the area of the crack multiplied by the surface energy per unit area, y; for a linearly elastic material, the work done by the loading system, W, is twice, but of opposite sign, to the elastic strain energy. The elastic strain energy is more difficult to calculate, but by arguing that a crack could be represented by an elliptical hole, Griffith was able to utilize an earlier result due to Inglis, who had calculated the distribution of stresses and strains around an elliptical-shaped hole in a loaded sheet. With these terms, Griffith was able to express the total energy of the cracked plate in terms of the crack length and the materials properties:
U = -nc2a2/E + 417.

virtue of its size, location and residual stress field, dominates the fracture of the material and, thereby, circumvents many of the statistical problems in fracture. The most popular method of creating the controlled flaw has proven to be the indentation method, in which a crack is introduced into a material at a predetermined location by loading a pyramidal diamond into the surface with a known load, producing two orthogonal semi-circular cracks. The use of indentation flaws has also enabled some of the fundamental aspects of brittle fracture in single crystals to be studied experimentally due to the lack of plasticity. In surveying the fracture of ceramic materials, we have attempted to address a number of the major concerns that have guided research as the structural applications of ceramics have been explored. One such concern has been the large variability in strength typically exhibited by a batch of materials. This is now recognized to be largely a result of inhomogeneities and flaws introduced by processing and subsequent machining and handling. A second concern relates to the long-term reliability of glasses and ceramics, and is usually attributed to environmentally assisted crack growth. For this reason, particular attention is paid to slow crack growth in this review. A third issue is the lack of resistance to fracture typical of most ceramics, and so much of the research activity is devoted to understanding the fracture processes and how to exploit the understanding to impart greater fracture toughness. Space does not, unfortunately, permit a mathematical development of all the ideas involved so we have concentrated on describing the concepts themselves, giving reference to the appropriate literature. Wherever possible we have selected observations that provide experimental support for the mechanics discussed. II. CONCEPTS Although the mechanics of fracture are extensively discussed elsewhere in this special issue by Hutchinson [12], it is instructuve to review a number of the basic ideas here. Of the concepts involved, the key ones of interest for this contribution are: conditions for crack extension in both a uniform and nonuniform loading, crack driving forces, steady-state crack growth, and R-curve behavior. 11.1 Conditions for crack extension 11.1.1 Uniform loading. Our starting point, as indeed it was for Griffith in 1920 [13], is the application of the first law of thermodynamics to the virtual extension of a pre-existing crack. The system considered by Griffith was closed, consisting of a crack of length c in an elastically homogeneous body subject to a spatially uniform biaxial tensile stress field, 0, by an externally applied load. In establishing the conditions for crack extension, Griffith explicitly considered, for the first time, that the total energy,

(1)

As in other examples of stability theory, the system is at equilibrium when its overall energy has a stationary value with respect to any virtual change in the crack length, c: dU/dc = dU,/dc + dUJdc = 0. (2)

Applying this condition led Griffith to the statement of equilibrium that bears his name: c7= (2Ey/xc)2. (3)

As mentioned above, Griffith used the stress solution for an elliptical hole to model that of a crack. As we will see later, the notion of a crack as an ellipsesurely just an available stress analysis solution invoked by Griffith to provide a tractable solutionhas conditioned several generations of materials scientists to think that a crack has a well-defined tip radius. The energy arguments of Griffith, as well as subsequent models, have only one physical dimension in them-the length of the crack. The dependence of strength on the flaw size was confirmed experimentally by Griffith, and has subsequently become deeply ingrained in the protocol of the materials science community as a way of determining the critical flaw size introduced by processing or machining. It is no exaggeration to say that thousands of papers have been published seeking to back-calculate the critical flaw size from strength data. It is also interesting to note that in many cases there is an implicit assumption of what might be termed elastic similitude; a processing defect,

Fracture of ceramics and glasses

1117

whether a crack or not, could act as the origin of fracture and therefore could be modelled as a crack. Although many papers have demonstrated the applicability of the Griffith relationship, one in particular is of note. Engel and Hubner [14] systematically altered the flaw size (of a cemented carbide) by subsequent hot isostatic pressing, thereby keeping the fracture toughness constant. They found that the strength was indeed controlled by the largest microstructural defect and that the strength was inversely related to the flaw size as predicted by the Griffith relationship. This was found to hold irrespective of whether the flaw was a fissure, an unusually large WC grain, or a pocket or cobalt. As we will see later (Section III) the Griffith relationship is indeed a useful concept provided that there is indeed a dominant flaw, its size is significantly larger than the scale of the microstructure, and that no residual stresses are present. The concept of a statistical variation in fracture strength and its dependence on the microstructure also has its origin in the Griffith relationship. The assumption made is that any material contains a population density of flaws of varying size, and that an observed variation in strength is attributable to a variation from one sample to the next of the size of the largest flaw. Thus, from a practical point of view any attempts to increase the reliability (for instance the Weibull modulus) of a material have been directed towards improving the reproducibility of processing in order to produce a tighter distribution of flaw sizes. To continue with the energetics of crack growth, the arguments introduced by Griffith can be generalized to any system irrespective of the crack and loading geometry. Defining the quantities G = -dU,/dc, (4)

re-expressed in terms of this crack driving force. (The crack driving force is equivalent to the force per unit length acting on a dislocation.) Another widely used function to express the forces acting on a crack, and one that is used extensively in this contribution, is the stress intensity factor, K, again introduced by Irwin. For elastic materials under plane stress conditions, the stress intensity factor is related to the strain energy release rate by G = K=/E One of the principal advantages of the stress intensity formalism, and one that has made it particularly useful in discussing indentation fracture mechanics, is that (for a given mode of fracture) the stress intensities are additive quantities whereas the strain energy release rates are not. Equation (3) represents the necessary conditions for crack extension and failure. The sufficient condition for equilibrium crack extension is that the second derivative of the energy be less than zero. In terms of the forces on the crack, the sufficient condition derives when the following inequality is satisfied: dG/dc > dR/dc. (6)

where G is the strain energy release rate and the resistance force, R, to fracture is

R = dU,/dc,

(5)

enables this equilibrium condition to be stated in terms of the (thermodynamic) forces acting on the crack:

In the case of a homogeneous material, the crack resistance is expected to be independent of the crack area. This reduces the condition for crack extension to the simpler one that the strain energy release rate decreases as the crack extends. However, if the material is not homogeneous, a variation in crack resistance force with length, termed R-curve behavior, may result. In terms of the stress intensity factor, the instability condition may be expressed as dK/dc > 0. More recently, a number of investigators [16-181 have argued that the irreversibility in crack extension imposes an additional constraint on the conditions for the propagation of a crack. A most thorough analysis, one within the framework of irreversible thermodynamics, has been introduced by Rice [16]. He shows that for a non-negative entropy production associated with the extension of a crack the following inequality holds: (G - 2y)v 2 0, (7)

G=G,=R.
The concept of the force balance acting on a crack front was introduced by Irwin [15], who showed that the strain energy release rate corresponded to a thermodynamic force acting per unit length of crack front,t and that previously derived equations for mode I and mode II loaded cracks could be explicitly t Other generalized energy functions have been introduced to incorporate effects such as inelastic and creep behavior, for instance the J integral and the C integral. where v is the crack speed. The 2y term in eqn (7) properly refers to the reversible work of separating the fracturing surfaces per unit area of surface. This is the same as the true surface energy under conditions of chemical equilibrium with an adsorbing species in a bulk phase at a constant chemical potential. In general, this may be identified with the Dupre energy of adhesion for the separation of two dissimilar materials, or the grain boundary energy for fracture along grain boundaries. The condition for crack growth of eqn (7), although apparently not

1118

D. R. CLARKE and K. T. FABER sharp indentor used to create the hardness impression [24]. The plastic strain mismatch of the indented volume with the surrounding elastic matrix acts to wedge open the two semi-circular radial cracks. The indentation crack may then be modeled as a crack subjected to a center loaded point force for which the stress intensity [25] may be written as
K, = ~P/c~, (10)

widely appreciated in the ceramics and glass literature, is of particular relevance to the discussion of environmentally affected crack growth (see Section V) since it relates both the crack velocity and the energy to create new surface in the environment in which crack growth is occurring. Experimental verification of the form of eqn (7) comes from the recent work of Maugis (reviewed in Ref. [17]), who has studied a related fracture problem, that of the contact between a sphere and a flat surface, albeit one in which the forces involved (primarily van der Waals forces) are considerably smaller than those associated with the fracture of structural materials. Maugis measured the rate at which the area of contact between a polyurethane sphere and a glass substrate changes as a function of applied load. The advantage of this configuration is that it allows the direct observation of the growth (and healing) of an (external) interface crack. (The deformation of the sphere near the contact plane has also been used to study the effect of surface and intermolecular forces on the crack tip shape [19-231). Maugis found that the condition for propagation at velocity, v, can be expressed as (G - 2~) = v#J@,, v), (8)

where the subscript r is used to denote the residual crack driving force remaining after the removal of the indentation load P. The parameter x is a constant for any given combination of material and indentor. [24]. An important property of the indentation crack, in the absence of any applied stress, is that it is inherently stable since the crack driving force decreases as the crack extends (dK/dc < 0). When an external stress is applied to the material, the indentation crack is subject to the combined effects of the applied stress and its residual stress field. Making use of the additive property of stress intensity factors, the overall crack driving force may be written as
K = K, + K, = J/a,,cj2 + xP/c312. (11)

where 2y is the interfacial surface energy, in this case equal to the Dupre work of adhesion, and a, is the Williams, Landel and Ferry (WLF) shift factor relating the spectrum of characteristic viscoelastic losses to frequency. Maugis shows that eqn (8) can equivalently be written, for small crack velocities, as (G - 2y)~~= 2ya(T). (9)

Not only does Maugiss work provide an experimental verification of the form of eqn (7), but is also provides a quantitative prediction. In separate experiments, Maugis independently measured the surface energy and the propagation losses, thus confirming the value of the individual terms involved. 11.1.2 The indentation crack&-an example of nonunzfirm loading. As mentioned in the Introduction the use of indentation induced cracks has enabled systematic studies to be made of some of the physics of fracture in brittle materials. Because of its widespread application to fracture problems in ceramics we consider here the conditions for extension of an indentation crack. The indentation crack in the surface of a homogeneous material subject to an applied external stress represents a simple instance of nonuniform loading. The non-uniformity arises from the fact that, in addition to the uniform applied stress, the indentation crack has an associated residual stress field. This can be clearly seen in a transparent glass by observing the indentation in transmitted light through crossed polarizers (Fig. 1). The origin of the indentation crack stress field lies in the elastic/plastic accommodation by the surrounding material of the

Examination of this equation confirms the expected trends. For short cracks, the influence of the applied stress is small compared with the local residual stress and so the crack system is stable. By contrast, at large crack lengths, the applied stress field dominates and the system is unstable. The crack system passes from a stable to an unstable condition at an intermediate point at which dK/dc = 0. Applying this condition, and the equilibrium fracture condition, K = KC, leads to equilibrium failure crack length, c,,,, and applied stress (a,): c,,,= [4xP/KJ2 a,,, = 0.75 K,/$c!,!~.
(12)

These equations demonstrate that a contact flaw having a residual stress field can grow in a stable, equilibrium manner on the application of an applied stress until a critical stress is reached. This behavior is quite different from the Griffith, or uniformly loaded, crack where unstable growth occurs spontaneously from the initial crack length. The conditions for crack extension from the above equations are plotted in Fig. 2 for a comparison of the equilibrium failure paths for indentation and Griffith flaws. Consider the application of an applied stress to a flaw of initial length, c,. If the flaw does not have an associated residual stress field (or if it has been annealed away), it will suddenly become unstable and grow catastrophically when the applied stress reaches a value of aO. Alternatively, for an indentation flaw of the same size, the crack would

Fracture

of ceramics

and glasses

1119

Fig. 1. Vickers indentation flaw in soda-lime glass. (a) In reflected light the radial traces of the radial cracks can be seen emanating from the corners of the contact impression. (b) In transmitted light through crossed polarizers the residual stress field is visible in reverse contrast. (Courtesy of R. F. Cook.)

grow stably along the solid curve until it too becomes unstable at a length, c,,,, and corresponding applied stress, a,. In the majority of glasses, at least, subcritical crack growth (see Section V) of the initial flaw will occur before any loading can be applied. This is depicted by the extension of the flaw to a length, ch, in Fig. 2. For such an environmentally affected indentation flaw application of an external stress does not extend the crack until the equilibrium (solid) curve is reached. With further increase in stress, the crack grows stably along the curve 1 until the point of instability is reached and unstable crack extension occurs at the same value, CT,,as before. The effect of
PCS. WI-K

subcritical crack growth on a Griffith flaw is quantitatively quite different; no growth of the Griffith flaw takes place with increasing applied stress until it reaches a value of aA at which point the material fails spontaneously (curve 2). The fact that the failure stress is lower when subcritical crack growth occurs is a consequence of the strong dependence on strength on initial flaw size. It is interesting to note that the analysis reproduced here makes the specific predictions that, in the absence of subcritical crack growth, the applied stress at which the Griffith flaw extends spontaneously is 2.12 times the value at which the indentation flaw becomes

1120

D. R. CLARKE and K. T. FABER

then to measure the failure strength. In such an experiment, it is not necessary to measure the indentation crack length, which in opaque materials can be a particularly difficult task. A plot of strength vs indentation load (a strength degradation plot) gives a straight line with slope (-l/3) for a continuum material. II.2 Steady-state conditions Invoking the condition of steady-state propagation of a crack allows, as will be seen below, a separation of the effects that occur at the crack tip from the rest of the crack. This is of importance in establishing the role of a variety of phenomena, ranging from intermolecular and surface forces to crack bridging fibers, on fracture. The assumption of steady state is tantamount to operationally assuming that there exists a crack-tip core region that moves along with the crack and where the details of the atomic arrangements and forces are invariant with crack position and hence can be neglected when determining conditions of steady state. Separation of these terms can be accomplished by writing the surface work term, U,, as u, = 2yc + r, (14)

CO

'0

Qm

CRACK LENGTH, cr

Fig. 2. Schematic diagram of equilibrium failure paths for an indentation flaw (1) and a Griffith flaw (2). The initial flaw size is c, but can grow by sub-critical crack growth to a length of CA.Note that the indentation flaw can grow in a stable manner until a critical length c, is reached. By contrast, the Griffith flaw (a residual stress-free flaw) fails spontaneously once a stress oh is reached.

where r is the energy associated with the core. Substituting into eqn (2) for the equilibrium condition for crack growth gives -dU/dc = -dU,/dc - d(2yc + r)/dc. (15)

unstable. Likewise, the initial flaw can extend to 2.52 times its size before it is unstable. This is a remarkable result since it indicates that there exists no unique crack length or flaw size below which failure will not originate. Therefore, processing must aim to produce materials with no flaws. [In polycrystalline materials, the continuum assumptions are not valid at small flaw sizes with respect to the grain size, and additional terms must be introduced into the driving force analysis of eqn (1 l).] The prediction of a universal, equilibrium failure curve for indentation flaws has been tested in a number of investigations in which length of an indentation crack is measured as a function of the applied stress. The results of one such study on glass are reproduced in Fig. 3. The foregoing analysis sets the stage for using indentation cracks to investigate the physics and chemistry of fracture. The analysis also provides a method for measuring the fracture toughness of brittle materials. One method relies on a rearrangement of eqn (12): ,J, = AK:/3 J - I, (13)

The crack length dependence of the term r is unknown. However, if the assumption is made that in steady state the term r is independent of crack length, the second term on the right will be constant 2y and the equation reduces to the Griffith equilibrium condition: G = G, - 2~. (16)

150z
L

ii
e i
f

Qm-------

OQ
00 0

IO,
cm

100

00

co
0

50

-/IO

c? 0

8 o~PREOICTION

where the constant A depends on the material. This equation forms the basis of an experimentally convenient way of measuring fracture toughness, and for assessing the flaw size dependence of strength [26]. A typical experiment is to create an indentation crack (with an easily measured point contact load, P) and

20

30
CRACK LENGTH

40

1
50

Fig.

3. Experimentally determined failure path for indentation flaws in glass. The solid curve corresponds to the theory of this section. (Redrawn after Marshall.)

Fracture of ceramics and glasses Thus, the detailed knowledge of any short-range (compared with the crack length) crack tip interactions can be neglected if steady state pertains. The concept of steady state has been invoked by Cook [27] in order to show that the strain energy release rate at the threshold stress intensity for slow crack growth (Section V) is a direct measure of the equilibrium surface energy. From his measurements of the threshold for the propagation of cracks in sapphire in water, an estimate of the surface energy for sapphire in water is calculated to be 1.42 J m-. The assumption of steady state, and with it the possibility of neglecting the details of the actual processes involved covers not only the detailed physics of lattice trapping? and relaxation phenomena but also the contributions lumped together under the term disjoining pressures1 used to describe the short-range interactions between crack surfaces in close proximity when fracture occurs in a liquid environment. There are a number of precedents for this type of propagating core assumption in different fields. In considering lattice trapping of a crack, Thomson [3 I] partitions the energy in such a manner that the atoms at the crack tip are in a core and that their energy can be neglected in the change in overall energy of a long crack. Perhaps the most widely used analogy is in the field of dislocation mechanics. The energy of the dislocation core is neglected compared with changes in the elastic strain energy when calculating the energies of dislocation motion, alternative dislocation configurations and dislocation interactions. Another analogy is in the field of wetting, As de Gennes has noted [32], Thomas Young realized that the equilibrium contact angle for a partially wetting drop can be derived from the far-field surface energies, without consideration of any surface force effects at the core of the contact line of the drop. In a situation paralleling that of fracture, de Gennes shows that the spreading of a liquid drop over a solid surface in vacuum (dry spreading) can be well described by neglecting any energy dissipation in the core region local to the range of any surface forces, at the contact line of the spreading drop, and that the driving force for spreading depends only on the excess energy above that necessary to maintain drop equilibrium. II.3 R-Curve behavior The concept of a crack resistance that increases with the length of the crack-R-curve behavior-is well established in the fracture mechanics literature since it has been recognized for many years that metal structures under plane stress conditions can

1121

t The crack analogy of the trapping of a dislocation in a Peirels barrier, discussed by Thomson [28] in a companion article in this issue. $ This includes, amongst many others, screened van der Waals interactions, solvation forces, electrostatic interactions and structural forces [29,30].

exhibit such a crack length dependence on fracture toughness. Until recently, the concept did not seem to apply to ceramic systems since there were no reported instances of such behavior. However, measurements of the fracture toughness of transformation toughened zirconia ceramics and of fiber reinforced ceramics indicate that these materials demonstrate R-curve behavior. Furthermore, R-curve behavior is a predicted feature of even the simplest theoretical models based on the formation of either a process zone (Section VII.2) or a crack bridging zone (Section VII.3). Whether or not a material exhibits R-curve behavior is of utmost practical importance, since the material will then have a flaw tolerance and, if the R-curve is steep enough, stable crack growth prior to failure. Ideally, one would like an R-curve that extends over tens or hundreds of pm so that stable cracks can grow sufficiently large for them to be detected by a variety of NDE techniques before failure occurs. The mechanistic origin of R-curve behavior is emphasized in recent work on fiber toughening, ligaments, etc. The underlying reason for the existence of an R-curve may, however, be more general. There is reason to believe that the origin of the R-curve and its extent lies in the scale over which a crack must sample the statistical variability in fracture resistance through the microstructure. When the crack is small, it encounters changes in local crack resistance forces over significant distances compared with its length. As the crack extends, it has an increased probability of encountering those microstructural features having the largest resistance to fracture, hence the overall crack resistance increases. At a still larger length, the probability of interacting with the regions of largest resistance becomes unity and the crack resistance reaches the highest value. It is easy to appreciate that this also marks the transition to continuum or quasi-homogeneous behavior. The effective crack resistance will be unaffected by local, small-scale variations for still larger crack lengths. Thus, the length of the crack over which the crack resistance increases to a plateau level will increase with the variability of microstructural crack resistance. Conversely, for a very narrow distribution of fracture resistance, such as equispaced obstacles of uniform resistance in a homogeneous matrix, the R-curve will be very abrupt. There is some support for the above conclusions in the literature. One is work on the strength of polycrystalline materials, described in the following section, in which an apparent R-curve results as the flaw size is increased from below to well above the grain size [33,34]. Second, a connection between R-curve behavior and the statistical distribution in grain size of a microstructure is suggested by a recent publication by Swain [35] on a polycrystalline alumina. This work indicates that the origin of the R-curve is the bridging of the crack by especially large and intact grains, having the largest thermal

1122

D. R. CLARKE and K. T.

FABER

expansion strains, that have not yet been strained to failure. An interesting consequence of R-curve behavior is the apparent increase in reliability demonstrated by these materials [36,37]. The Griffith equation [eqn (3)] stipulates the range of strengths for a given range of flaw sizes. The strength-flaw size relation for materials demonstrating R-curve behavior is much shallower than the Griffith relationship, because toughness scales with flaw size according to a power law K, = AC, (17)

III. DISCRETE

TO CONTINUUM

BEHAVIOR

where n describes the shape of the R-curve. This implies [36] that the measured strength of the material now is of the form a = a/7zc.5-n.
(18)

As the power law exponent n approaches 0.5, the strength converges to a singular value regardless of the flaw size.
Summary remarks

The foregoing sections are largely a recapitulation of standard theory. In terms of our discussion of the fracture of ceramics and glasses, they have been included because for many years, the fracture toughness of these materials has been considered to be a material invariant property, i.e. one that can be used to characterize the materials resistance to fracture. This approach, understandably, has had a very broad appeal since a single number could be used for design purposes. Then, the attainable strength would be dependent only on the success in limiting the size of any processing, machining or handling flaw. However, this material invariance is only valid for elastically homogeneous materials, and is not applicable for inhomogeneous materials. The former situation pertains, for instance, in the majority of glasses. The lack of such a material invariance in most other ceramics makes it much more difficult to specify the fracture resistance of a ceramic, but on the other hand opens up the opportunity for increased fracture toughness, or, more accurately, fracture resistance by materials design. Much of the current interest in design of fractureresistant materials can be considered in terms of a transition from local to continuum behavior, and in particular in designing materials with a range of scale. From this perspective, the alternative to designing a homogeneous material having a high intrinsic fracture resistance is to construct a heterogeneous microstructure in which the fracture resistance scales with the size of the microstructural components. In some senses, the most desirable would be a fractal micr0structure.t ?This suggestion was first made by T. M. Shaw.

In an elastically homogeneous material, such as glass or a single crystal, the resistance to crack propagation is independent of crack length and location of the crack. As a result the fracture toughness is independent of the size of the crack. This has been demonstrated by many workers over the years, and is given clear expression in strength degradation plots obtained from glasses. However, as introduced in the previous section, if the material is inhomogeneous, the crack resistance force [eqn (511might be expected to vary with crack size and a scale-dependent fracture toughness will result. Evidence for such an effect can be found in the numerous experimental reports that the strength of a ceramic is found to depend on its microstructure and grain size. It is also found that there is a tendency for the measured strength to be smaller than expected from the Griffith relation as the flaw size is smaller. There is, however, no consensus of how the strength does vary quantitatively with grain size, and at what point a transition to continuum behavior will occur. In part, the confusion of the data stems from the fact that the strength studies have been performed on materials containing naturally occurring flaws rather than from controlled flaws. The first clear-cut example of discrete to continuum behavior in brittle materials is taken from the work of Cook et al. [33,34]. Cook et al. studied possibly the simplest inhomogeneous material, a series of polycrystalline aluminum oxides in which the elastic inhomogeneities are grain boundaries and anisotropic grains. In contrast to the previous studies, the indentation method was used to introduce a known flaw size and the strength measured. From this data, they constructed strength degradation plots such as Fig. 4. At large indentation loads, when the flaw size is large compared with the grain size, the strength degradation plot is a straight line of slope (- l/3), as expected for continuum behavior. This characteristic holds down to the smallest loads (and hence flaws) for sapphire, but not for the polycrystalline aluminas. In the latter, the measured strength falls significantly short of the expected value and the deviation is larger the smaller the flaw size. For purposes of comparison, the grain size of these materials lies between about 10 and 35pm. The apparent decrease in fracture toughness at small flaw sizes, a form of R-curve behavior, is shown in Fig. 5, for the variety of polycrystalline aluminas studied. A similar transition from small-scale to continuum toughness has been reported for other polycrystalline materials including barium titanates and glassceramics [38]. Cook et al. account for the observed dependence on the scale of the microstructure by introducing an empirical microstructural driving force, &, into the net stress intensity factor acting on the equilibrium indentation crack [cf. eqn (ll)]:

Fracture of ceramics and glasses

1123

(dK/dc = 0) to eqn (20) gives the critical flaw size as c,,,= [4x(P + P*)lK,123,
(23)

which in turn leads to a value for the critical stress:


a,,, = 3K;3/443JI~3(P + P*). In the continuum (24)

0.J

1om3

1o-2
NORMALIZED

10-l

1
LOAD,

10
P/P

lo2

INDENTATION

limit, the fracture toughness can be expressed in terms of the critical stress to propagate a flaw introduced by the contact load as in eqn (13). As the flaw size diminishes to approach, and fall below, the discrete-continuum transition, the apparent toughness also falls, and the eqn (13) has to be modified by including the explicit dependence of the critical stress [eqn (24)]. This leads to an expression for the apparent toughness: K = K,[P/(P + P*)], (25)

Fig. 4. Normalized plots of strength vs indentation load for a series of calcium oxide and magnesium oxide co-doped aluminum oxide polycrystalline ceramics (Courtesy of R. F. Cook.)

K=K,+K,+K,,,

(19)

such that
K = @J,c* + (xP + pQ)c = Kc. (20)

Since Cook invoked this term, further work indicates that its physical origin is the constraint on crack opening provided by unbroken grains behind the crack front [39,40]. This phenomenon of crack bridging is discussed in some detail later, but is easy to visualize during the fracture of a polycrystalline material in which the cracks grow along the grain boundaries. Behind the crack tip, interlocking grains serve as ligaments. For small cracks the probability of intersecting a ligment is small, and the behavior is determined by the discrete microstructure. In the continuum limit of large cracks, xP $ pQ, the maximum applied stress reduces to the value of eqn (11) in Section 11.1.2 to give the Pm/ dependence described previously. In the limit of small cracks, xP < pQ, the strength is independent of the applied load and only dependent on the microstructural term:
u, = 3K;43/443$ @Q) = u$.

which is the theoretical curve plotted through the data of Fig. 5. A further transition is expected to occur at very small flaw sizes, when the flaw propagates to the point of instability entirely within a locally homogeneous material. In polycrystalline materials, such as those investigated by Cook et al., this would correspond to flaws located in, and propagating along, grain boundaries. In such materials, the growth of a very small flaw might encounter two sequential Rcurves. The flaw would grow stably in the grain boundary until it became unstable. Then, it would experience the varying crack resistance associated with the discrete-continuum transition, decelerate, grow stably along the new R-curve and finally become unstable again, causing the material to fail.

(21)

1
. B z I 0.2 I 10 _? -7 -1 vcc3

This formalism allows for the transition from discrete (termed by Cook et al. microstructural) to continuum behavior to be delineated by the condition at which the microstructural determined strength is equal to the continuum strength, viz.
xP = xP* = pQ. (22)

lo-

10

10

1
lo2

NORMALIZED INDENTATION

LOAD. P/P*

Applying

the

condition

of

crack

instability

Fig. 5. Normalized plot of apparent toughness vs indentation load for the same aluminum oxide ceramics as in Fig. 4. (Courtesy of R. F. Cook.)

1124

D. R. CLARKE and K. T. FABER IV. THE ATOMIC STRUCTURE OF CRACK TIPS IN BRITTLE SOLIDS

A fundamental question in the fracture of materials concerns the relationship of the atomic (and electronic) structure of a crack tip to the applied stress intensity and their dependence on the interatomic force laws. Answers to this question are required for a number of reasons, not the least of which is to provide a deeper understanding of the fracture process. In addition, with the increasing level of sophistication of numerical simulations, including molecular dynamics and first-principles electronic structure calculations, an experimental check of the simulations is required. The lack of plasticity exhibited by brittle solids suggests that these materials will provide the best experimental opportunity to observe the actual atomic structure and test any theoretical model. The associated issue of whether cracks are sharp or blunt also has been questioned because of an accumulation of experimental evidence (particularly on glasses) which suggests the need to invoke the idea of crack blunting. For the case of glasses this has become something of a controversy, which we will return to in the following section, but it is perhaps useful to elaborate on the progress that has been made over the years in describing the atomic structure at the crack tip.

Models for the crack tip Whilst the emphasis of study has (correctly from an engineering point of view) been on the conditions for crack extension, the details of the atomic structure have been of secondary importance. In a sense this lack of attention has been substantiated by the result, proven for a linear elastic model by Willis [41], and later for the more general nonlinear elastic model by Rice [42], that contrasting descriptions of blunt and sharp cracks described below both lead to the same conditions for crack extension. It remains an open question if the situation is similar to that found in simulations of grain boundaries where the energy of a grain boundary appears to be relatively insensitive to the exact details of the interatomic potentials used, whereas the detailed atomic positions are very dependent [43].t The two contrasting models for the atomic structure of a crack tip in brittle solids that have evolved in the literature may be simply summarized. The first is now attributed to Griffith, although it strictly has

t In this and subsequent discussions, the issue is quite distinct from that considered by Kelly er al. [44], and later Rice and Thomson [45], who investigated the stability of a crack relative to the generation of a dislocation out of the crack with a Burgers vector which blunts the crack. $ The Barenblatt model was later reformulated in terms of continuous distributions of dislocations thereby providing a consistent and rigorous continuum description of a crack [48].

its origin in the work of Inglis, and is referred to here as the Griffith-Inglis model. It is the extension of the assumption that a crack can be considered to be an elliptical hole. This leads to an abrupt 90 bifurcation of the plane of atoms at the end of the crack, which could be termed a blunt crack tip, and a specific crack shape that depends only on the applied stress intensity and the elastic modulus of the material. It is a continuum description and assumes that the material behaves in a linear elastic manner on all length scales down to the very crack tip. Apart from the well-known problem that the Griffith-Inglis model leads to a singularity in the stress field at the tip, it assumes that the atomic cohesive forces are equivalent to a hard-wall interaction. There is thus no length scale associated with the cohesive forces, which is inconsistent with the existence of the surface energy used in the Griffith equation. The contrasting model, one in which the crack surfaces separate continuously, stems from the work of Elliott [46] who noted that the tip of the crack cannot have an elliptical geometry. Rather, it consists of three distinct regions. Well ahead of the crack the atomic bonds were subject to small strains and so the displacements are linear elastic. Well behind the crack the bonds are stretched far apart; the atoms can exert little force across the crack and the traction-free surface assumption of linear elastic fracture mechanics remains essentially valid. In between, there is a third region where the cohesive forces are well beyond their linear elastic range but nevertheless remain significant to provide a degree of bonding across the crack faces. This idea of cohesive zone was extended by Barenblatt [47] in describing a linear continuum mechanics theory in which arbitrary cohesive forces acted between the crack faces in the cohesive zone.$ In the Barenblatt model, the width of the cohesive regions depends upon the details of the interatomic force laws, but is assumed to be independent of the applied stress on the crack. A very recent paper [49] shows that some of the inadequacies of the Griffith-Inglis and the Barenblatt models can be resolved by a consistent treatment of the intermolecular and surface forces in the vicinity of the crack tip. The predicted crack profile coincides with that of the Griffith-Inglis crack away from the crack tip and has a form similar to that of the Barenblatt cohesive zone at the tip. Discrete calculations performed on the computer with a variety of interatomic force laws have all substantiated the existence of a cohesive zone and a general crack shape consistent with that predicted by Barenblatt. This is not altogether surprising: in the majority of the simulations the calculations are rendered manageable by assuming that the atoms in the immediate vicinity of the crack tip, which are allowed to move individually in response to the local interatomic potential, are embedded in a linear elastic continuum; a refined but nevertheless conceptually similar calculation to that performed by Barenblatt.

Fracture of ceramics and glasses

1125

Fig. 6. Lattice fringe image of the crack-tip region in a Mg-Sialon ceramic. The crack, C, is seen edge on with the tip at the location T.

The extent of the cohesive zone, where it has been calculated, has been found to be quite restricted in size for reasonable force laws; for instance, in the work of Gehlen and Kanninen [50,5 11, the zone size is only two or three atomic spacings. The advantage of these computations is that they can provide detailed predictions of the atomic positions around the crack tip. Hitherto, the calculations have been primarily performed to calculate the conditions for Z&, but the atomic positions as a function of applied stress intensity up to KI, will undoubtedly be needed for comparison with experiment.

Direct observations
In contrast to the voluminous literature devoted to linear elastic fracture mechanics, the modes of fracture in different materials and the mechanisms of imparting fracture resistance to materials, there have been surprisingly few attempts to observe directly the structure of crack tips in solids. For many decades, until the widespread use of transmission electron microscopy, this was understandable since the crack

7 This contrasts with the direct observation by Ohr [54] of dislocation emission from crack tips, and the existence of a dislocation free zone, in crack propagation in ductile materials. This is described in a companion contribution in
this volume.

tip, and the enclave in which linear elasticity is thought to breakdown, was beyond the capabilities of most imaging techniques. There was thus little opportunity to put to the test which of the contrasting models best described the atomic arrangement at the crack tip. The experiments of Hockey [52,53], in which he used diffraction contrast imaging in the transmission electron microscope, marked the first direct observations of crack tips in brittle solids. These studies, on Si, Sic and Al,OJ, provided the first confnmation of the absence of remnant dislocation plasticity in such materials at room temperature.? Since the time of Hockeys original experiments, newer generations of electron microscopes with higher resolution have been developed and imaging technique has improved. Together, they have made it possible to directly observe the lattice structure of most materials including the majority of semiconductors and oxides. This had enabled the techniques of lattice fringe imaging, that have proved so useful in other areas of materials physics, to be finally being brought to bear on the observation of the crack tip. Figures 6 and 7 are examples of the type of direct, high-resolution observation of the crystal lattice in the immediate vicinity of crack tips that is now attainable using the transmission electron microscope. Figure 6 is such a lattice fringe image of a

1126

D. R. CLARKEand K. T.

FABER

Fig. 7. Crack tip in silicon. The crack is seen edge on with its tip at the position T. The individual fringes correspond to the (11 I) planes in silicon. As in the previous figure, distortions of the planes near the tip can be seen by viewing the micrograph obliquely.

crack tip in a Sialon ceramic-a Al,Mg,O-substituted silicon nitride [29]. The crack is viewed so that the crack plane, C, is seen edge on. The tip, T, is clearly identifiable. The lines correspond to the individual lattice planes of the crystal lattice. The diffraction contrast (the bright and dark contrast) around the tip provides direct evidence that the crack is under load; if it had not been there would be no such local variation in contrast. Figure 7 is a still higher resolution lattice fringe image, in this case of a (111) cleavage crack in silicon. The crossed lattice fringes correspond to the two (111) lattice planes having a spacing of 3 8. This photomicrograph is drawn from a recent investigation of cracks produced by indentation, and the brittl+ductile transition in silicon by one of the authors (DRC). Analysis of this and other images is still being undertaken, but they provide the first opportunity to test the contrasting continuum elasticity models and the discrete computations, such

as those of Sinclair and Lawn [55,56], performed for the same (111) cleavage crack in silicon. The experimental method of making these observations is similar to that first used by Hockey. Stable cracks are introduced into the material by indentation and then thinned by ion-bonbardment until sufficiently thin (generally a few hundred angstroms) to be transparent in the electron microscope. There are formidable experimental difficulties in obtaining atomic resolution images of a crack tip, not least of which is that as all the materials of interest are brittle the samples are exceedingly fragile and liable to break at any stage prior to actual observation in the microscope. One other set of experiments provides direct information about the fracture process, although the results have yet to be interpreted in terms of what is presently known about the structure of the crack tip. Haneman et al. [57] propagated cleavage cracks in

Fracture of ceramics and glasses single crystals of germanium in ultra-high-vacuum, and then allowed them to close up by reducing the load, measuring the resistance and photovoltage generated when the crack was illuminated. This seems to have escaped general attention, but its contents (on electronic states associated with the crack surfaces) is latent with possibilities of relating the detailed mechanisms of fracture to changes in the electronic structure and bonding.

1127

consistent with the role of attractive surface forces modifying the shape of the crack. This would indicate the importance of the surface forces and the more realistic nature of the cohesive zone models of the crack tip. Although the forces acting in these cases of adhesive contact are very small compared with the bonding forces that must be overcome during brittle fracture, the measurements of Horn et al. provide the first real test, other than those of electron microscopy, of the shape of cracks. IV. 1. Sharp vs blunt crack debate As suggested in Section II, the concept that the fracture stress of a material can be related to the radius of curvature of a crack, and that the crack tip radius was a meaningful parameter down at the atomic dimensions of the tip, had been the cause of a great deal of confusion in understanding brittle fracture. This literal interpretation was given experimental credence more than 25 yr ago in a study on aging effects on the mechanical behavior of glasses. Mould [61] examined the bending strength of abraded glass samples after aging in water or water vapor from 10 min to 100 days. By comparison to the untreated samples, Mould found that the samples that had been exposed to moisture were 3@-60% stronger. Moulds interpretation (along with that of numerous subsequent investigators) brought prominence to the notion of crack blunting in glasses by proposing that the strength increase was attributable to an increase in crack tip radius. Using Inglis solution, the strength of a material is described by its crack tip radius:
(26)

Other evidence
Some considerable insight into the shape of a crack comes from the experiments on the adhesion of solids, particularly spheres in contact with a rigid plane [17, 19231. In the experiments performed, the fracture plane is well defined and the low compliance compared with that of the majority of brittle solids make the effects of surface and intermolecular forces more evident. As in the true crack case, there are two contrasting pictures of the detailed shape of the contact region. One, the so-called JKR (Johnson, Kendall and Roberts) theory [19,20], assumes that there is an attractive force of infinitely short range between the surface and predicts an abrupt 90 contact as in the Griffith-Inglis model. The other, the Hertz model, generalized for the case of a finite range of attractive force by (DMT-Derjaguin, Muller and Toporov) [21,22], predicts a continuously varying contact, much as in the Barenblatt model. As in the crack problem mentioned above, the solution is difficult since the stress distribution depends on the surface interaction which depends on the exact profile of the deformed surfaces, which can only be determined once the stress distribution is known. This requires the solution of a nonlinear integral equation, once the surface force as a function of crack face separation is known (see Section VII.3.2). The two theories, JKR and DMT, have recently been put to experimental test by Horn et al. [58] using the Israelachvili apparatus [59,60] to measure both the contact radius and the surface profiles of two contacting mica sheets. With this apparatus, the investigators could measure the distance between the surfaces, adjacent to the point of contact, to within about 1 nm. Although the lateral resolution is limited to about 30 pm of the contact point (the crack-tip), the technique provides unparalleled measurement of the shape of contact. For non-adhesive contact (contact in a 1 M KC1 solution in which the net force is known from previous work to be a repulsive hydration force) the authors find excellent agreement with the Hertz theory, whereas for adhesive contact (contact in dry nitrogen gas so that the force of adhesion is the van der Waals attraction) they find detailed discrepancies with the JKR theory but broad overall agreement. The large stresses in the adhesive case distorted the mica sheets sufficiently to prevent an unassailable comparison with the JKR theory, but the observation of a larger contact zone and a shallower profile than in the non-adhesive case are

where ulheo,is the theoretical strength of the unflawed material. Dissolution of a few molecular layers could easily account for Moulds experimental results. No other interpretation for Moulds results seemed plausible until the issue was recently revisited from the perspective of the stresses acting on indentation flaws. According to indentation fracture mechanics, the net stress intensity acting on flaws produced by contact, such as point contact, line contact or abrasion, consists of a residual stress field and any applied stress field [eqn (11) for the case of a point contact]. Thus, an alternative explanation for Moulds findings would be that the aging treatments decreased the magnitude of the contact residual stress field, thereby leading to an effective increase in strength, no recourse to the detailed shape of the crack tip being necessary. To test this alternative, Lawn et al. [62] repeated Moulds experiments using indentation pre-cracked samples rather than abraded ones (Fig. 8). One set of samples were aged in distilled water, and silicone oil, and a second set underwent an annealing treatment of 24 h at 52OC, a treatment known from previous studies to be adequate to remove residual

1128

D. R.

CLARKE and K. T. FABER

Annealed

Ashdented

decreased, the fully aged crack grows to c&, which now determines the plateau stress described by the hatched region to the right of Fig. 9. In the annealed state, the crack still grows to c, but the residual stress component is no longer of appreciable magnitude. Hence, the upper curve is followed, and a higher strength obtains. If the cracks were being blunted, then annealed cracks would show similar increases in strength as aged samples. Furthermore, if blunting did occur, an increase in the crack length would not be expected, and based upon the Griffith criterion would actually serve to decrease the strength. Therefore, these most recent studies provide little credence to the blunting model, but suggest instead that sharp cracks remain sharp. V. ENVIRONMENT ASSISTED CRACK GROWTH

100

ld

I 104

I 10s

1Oa

Aging Time, s

Fig. 8. Aging of indented soda-lime glass in water. The strength and crack lengths as a function of aging time are represented for as-indented and post-indentation annealed

specimens. The hatched regions denote the average values immediately after indentation and, on the right, at the limit of measurement. (Courtesy of B. R. Lawn.)

stresses from indentations in soda lime glass. The measured strengths after these treatments are shown in Fig. 9, indicating that regardless of the treatment, the strength increased with time. In inert atmospheres (oil), the magnitude of the strength increase was identical to that in moist atmospheres, but the time taken for the strength to reach a maximum was longer. Most importantly, the annealed samples demonstrated higher strengths than either the aged or untreated samples. One of the advantages of the indentation experiment over that of Moulds was that Lawn and colleagues were able to measure the crack length in situ as well as the strength. They found that, contrary to the blunting hypothesis, the cracks extended during the aging treatment. These seemingly contradictory observations of strength increases accompanied by crack extensions are inconsistent with the blunting hypothesis but are to be expected if aging reduces the residual stress field of the crack. This behavior can be understood in terms of the schematic diagram of Fig. 2. For strength tests carried out immediately following the indentation, the crack has an associated residual stress field and growth follows the lower curve in a stable manner to a length c,. On aging, as the K, term is
t Stress corrosion cracking is known to occur in metallic and polymeric materials [63] too. The same stages of crack growth have been identified, but to the authors knowledge the same detailed chemical arguments described below have not been explored.

One of the properties of glasses that has intrigued investigators for many years is the stable growth of cracks at stress intensities less than I(lc measured in fast fracture or in an inert atmosphere. This phenomenon, variously termed slow crack growth, stress corrosion cracking, static fatigue and sub-critical crack growth, has also been found to occur in a number of crystalline ceramics in appropriate environments.7 Figure 10, taken from the original work of Weiderhorn [64,65] on soda-lime (window) glass is typical of the dependence of crack velocity on the applied stress intensity. Below the critical stress intensity, K,c, the crack velocity may be described by three principal regions as indentified by Wiederhorn: region I, at low stresses, in which the velocity is a

As-Indented

400 ~ _______~~______________--_-__ E : f g s t L 0 300 Radia 200 Eg+p_9~~ _I ,(I Lateral v / ,gnzcc@

100

11
102

I
104

I
106

I
1Oe

Id

Aging Time, s

Fig. 9. Repeat of aging experiment of Fig. 8 but with aging in an inert environment (silicone oil) rather than water. (Courtesy of B. R. Lawn.)

Fracture of ceramics and glasses

1129

higher stresses, crack sharpening occurs as the chemical reaction is enhanced at the stress concentrated crack tip and crack growth occurs. The crack velocity, v, could then be related to the reaction rate:
.

.
0

. -9

10 %

2
r^

10-s

. . . . . .

v = kT/h[A][B]

exp (AG*/RT),

i $
E z

. .. . III : * 1.0% .o *
yo

(27)

. : 0.2%
c: .*

_~

H20(1)

. . / 0 0

where [A] and [B] are the concentrations of reactants and AC* represents the change in free energy of reacion as the reaction goes to the stress activated state. The stress dependence of the reaction rate can be obtained by expanding the term AC* to AC* = AE* - TAS* - aAV* + yu,/p, (28)

* * + _. .. \__~

0.017%

10-7

*.p.-.* .. :. II ,,_f_:

a_* I
I ./ 1
I

0.4
STRESS

0.5
INTENSITY

0.6
FACTOR,

0.7

0.6

K,, MPa-my2

Fig. 10. The principal regions of subcritical crack growth indicated on a u-K plot for soda-lime-silicate glass illustrating the effect of relative humidity. (Courtesy of S. M. Wiederhom.)

strong function of the stress intensity above an initial threshold; region II, a central region in which the

crack velocity is independent of the stress intensity; and a third region, at KI approaching K,, where the velocity exhibits a strong stress intensity dependence. The bulk of research in this area has focussed on region I of the (u-K) curve. Since the majority of the life of a material exhibiting sub-critical crack growth is spent with the crack in region I, an estimate of the life expectancy of a material under a load can be made on the basis of the slope of the curve in region I. This has led to numerous life prediction models, but does not provide any insight into the origin of sub-critical crack growth, the slope of the (u-K) curve, or address why certain atmospheres, such as water vapor, promote crack growth and others do not. V.l Reaction rate theory approach A significant first step in understanding the origin of environment assisted crack growth was the proposal by Charles and Hillig [66] that the stress at the crack tip locally enhanced a corrosion reaction enabling crack growth to occur at applied stress intensities below K,c and that the crack velocity could be calculated from reaction rate theory once the stress was stipulated. The situation envisaged by Hillig and Charles is shown schematically in Fig. 11. At low stresses, the stress corrosion reaction serves to dissolve a few molecular layers along the crack faces and crack tip producing a blunted crack-and hence a sub-threshold behavior. At intermediate stresses, neither a rounding or sharpening occurs. At

where AE* is the stress free activation energy, AS* is the activation entropy, AV* is the activation volume, Q is the stress, y the surface energy of the crack, v, is the molar volume of the solid and p is the crack-tip radius. There was thus a quantitative relationship between the applied stress, the reactivity of the environment and the kinetics of crack propagation. Although couched by Hillig and Charles in terms of the geometry of the crack tip, and in particular, the Inglis-Griffith stress concentration argument, the approach is not restricted to any particular model of the crack-tip geometry. Thus, the simple shape changes of Fig. 11, which are now known to be inconsistent with an atomic description of the crack tip and with the deductions of the aging experiments (Section IV.4), may be accurate at a macroscopic level but the details of the actual crack-tip shapes are not important to the development of the reaction rate expressions for crack growth. Since its introduction, the approach of using reaction rate theory to calculate sub-critical crack growth has been extended and refined by a large number of investigators [67,68]. Of particular note is the multiple barrier

\ 1;
I
I

-0

(a)

(b)

(c)

Fig. 11. Crack tip shape changes envisaged by Hillig and


Charles when comparing the rate of dissolution with that of crack advance by stress-induced corrosion. In (a), with the rate of stress corrosion being the greater than that of dissolution, the tip is considered to sharpen, in (b) the rate of dissolution is equal to the rate of crack advance, and in (c) the dissolution rate exceeds the rate of advance leading to an effective blunting. (Redrawn after Hillig and Charles.)

1130

D. R. CLARKE and K. T. FABER

\I/
H-O

0
O\
H...

si
0

\I/
HIP i.

\I/

Si

SI

I 7
H

I,

I
/i\

/y\
2

Fig. 12. Schematic representation of the chemical reaction proposed by Michalske and Freiman for moisture enhanced Si-O-Si bond rupture at the tip of a crack in silica. The strained crack tip bond (1) absorbs a water molecule, (2) is cleaved by a dissociative reaction, and (3) is converted to surface silanol groups.

formalism of Fuller and Thomson [67]. All such analysis provide a general descriptive framework within which to calculate kinetic crack growth, but fail to address the detailed mechanisms of crack advance or the specificity of certain chemicals in causing crack growth.

V.2 Molecular reactions at the crack tip A different approach has been taken by Michalske and coworkers, in seeking a detailed molecular mechanism of how bonds at the crack tip react chemically with the environment molecules [69-7 11. I I The mechanism Michalske and Freiman [69] propose A WATER for the role of water in enhancing subcritical crack 0 AMMONIA ____ NZ growth in silica is outlined in Fig. 12. At the tip of HYORAZINE a crack, the strained S&-O-Si bonds are free to - - FORMAMIDE lo- react with any environmental species small enough to = METHANOL occupy the region at the crack tip. In the distorted T state, the Si-0 bond distance is increased and the 5 Io-3 > tetrahedral symmetry around the Si atoms is altered t such that the Si becomes strongly acidic and the t lo+:: bridging oxygen strongly basic. In a moist environd ment, it is possible for a water molecule to orient IO+ itself such that a hydrogen bond forms at the z Is bridging oxygen. Simultaneous interaction of the o lo-6 lone pair orbitals with a Si atom occurs. In the next step, the lone proton is lost to the bridging 0 and 10-l electron transfer takes place between the 0 from the AA water molecule to the Si atom. The requisite final A. I 10-e step is rupture of the hydrogen bond between O,,,, 0.6 0.6 0.4 (oxygen from the water molecule) and the transferred STRESS INTENSITY K, (MPa m*) proton. Where the previously S&U-Si crack tip Fig. 13. v-K, diagram for vitreous silica at room temexisted, two B-O-H groups now form the surface perature, showing that water, hydrazine, ammonia, methdirectly behind a new crack tip. Crack propagaanol and formamide all have the effect of increasing the rate tion may proceed in the same fashion, limited only of slow crack growth. (Redrawn after Michalske and Bunker.) by the supply of corrosive species to the crack tip.
l

Region II of the v-K curve demonstrates a plateau precisely because the crack velocity is limited by the transport of the species to the tip of the crack. A the the measure of success of MichalskcFreiman model is the prediction, and the subsequent demonstration, that any species having the same structural and bonding features as water, i.e. lone pair orbital on the opposite sides of a proton donor, should also promote slow crack growth [70]. Species such as ammonia and hydrazine provide nearly identical V-K curves, while N,, CO and acetonitrile inhibit it (Fig. 13). A further limiting factor is the opening at the crack tip. A threshold in the v-K curve arises naturally from the fact that at low K the crack opening is insufficient for a reactive species to reach the crack tip. This is also observed when conducting crack growth studies in environments containing large molecules. Molecules such as CH,OH,, though fitting the same structural and bonding criteria as described above, are too large to be accommodated in the near-crack-tip region, and therefore are not responsible for slow crack growth. This steric effect is responsible for a reduction in crack velocities by more than 5 orders of magnitude. The phenomenon of sub-critical crack growth is also well documented for crystalline oxides. On the basis of sub-critical crack growth in glasses, the observation in many crystalline oxides is attributed to the presence of glassy phases along the grain boundaries. However, the same phenomenon has been observed both in single crystal materials, such as sapphire, and in polycrystalline ceramics not containing an intergranular phase. Recent studies by

Fracture

of ceramics

and glasses

1131

Michalske et al. [71] elucidate some of the important features of stress corrosion in crystalline ceramics. Crack growth in sapphire, which has similar ionic/covalent bonding to SiO, (but with greater acid-base surface activity) was compared to silica along with MgF*, nearly fully ionic with weak acid-base character. In A1203 similar dissociative chemisorption reactions give rise to stress corrosion as in SIO,. However, in highly ionic materials, such as in MgF,, the chemically active sites do not have sufficient acid-base character for dissociative chemisorption to occur. Instead, stress corrosion is suggested to occur by ion solvation in which the corroding species act to neutralize the electrostatic attraction between ions. Hence, solutions having a large dielectric screening effect (i.e. high dielectric constant) such as water and methanol would promote stress corrosion. This is in contrast to environments which are more effective for dissociative chemisorption, such as ammonia. V.3 Continum dissipation model An alternative descrition of the origin of subcritical crack growth has been put forward by Maugis [17]. He proposes that sub-critical crack growth is a more general phenomenon exhibited by all materials, and that the crack velocity, v, results from a balance between the crack extension force (G - 2y), and the rate of energy dissipation at the crack tip.? In viscoelastic materials, the losses are due to viscous relaxations of the polymer structure. In glasses, the internal loss mechanisms have not yet been identified. Though Maugiss proposal is a tentative one, it is consistent with the thermodynamics of crack growth. In this description, an active environment reduces the surface energy of the freshly created surfaces, which has the effect of reducing the stresses needed to break the crack-tip bonds. Thus, the introduction of a more reactive environment is to reduce the surface energy and shift the (u-K) curves along the K-axis. Maugis calculates the magnitude of the shift as a function of adsorption of an active species, using the Gibbs equation, to show that the shift is proportional to the square root of the change in surface energy produced by the adsorption. Unlike the work of Wiederhorn, Michalske and others, who seek to describe the detailed mechanisms by which the crack-tip bonds are broken, Maugiss work emphasizes the energy losses associated with t One of the interesting features of Maugiss analysis is that there can exist a range of crack driving forces over which the crack will propagate in a stick-slip manner. In this regime, the crack velocity oscillates between two extremes, and as Cook (private communication) has noted can be expressed mathematically as an attractor. $ The exception here is failure at extremely high loading rates, where the time for the stress build-up is much faster than the characteristic time for viscous flow, so brittle fracture considerations dominate.

crack propagation. As in the work of Rice [16], a static fatigue threshold arises quite naturally as corresponding to the value at which the crack driving force falls to that given by the surface energy; in other words, to the Griffith condition for crack equilibrium in the chemical environment of the fracture. Here, the emphasis is on an active environment, one that affects the surface energy of the freshly formed surfaces, rather than on a corrosive environment. V.4 Amorphous phase assisted crack growth With the exception of sub-critical crack growth due to diffusive transport under creep conditions (as considered by Stevens and Dutton [72]), much of the attention devoted to slow crack growth or static fatigue has focused on ceramics at low temperatures in the brittle fracture regime. There is recent evidence to suggest that stress corrosion can occur at elevated temperatures in single-phase ceramics. Enhanced crack growth in regions of a polycrystalline alumina contaminated by impurities (in their case a silica impurity) has been reported by Cao et al. [73]. They attribute this to a stress corrosion reaction in which an amorphous phase is formed, wets the crack and enhances the local reaction rate at the crack tip. There is every reason to believe that this is a very general, commonly occurring phenomenon in ceramics, and that Cao et al. are merely the first to clearly identify it. The most important practical consequence of such observations is that crack growth can take place at applied stress intensities below the creep threshold (see Section VI.l) for the uncontaminated material, and thereby lead to premature failure. Their observations also demonstrate the importance of both clean fabrication and preventing exposure to an environment that may introduce a wetting amorphous phase.

VI. HIGH-TEMPERATURE FRACTURE In contrast to fracture at low temperatures where much of the essential physics is understood at least in outline, the high-temperature failure of ceramics is much more complex and, presently, only poorly understood. In part this lack of understanding stems from the rich variety of fracture behavior exhibited by ceramics, and in part from a paucity of comprehensive creep rupture studies with accompanying microstructural observations. The available studies do show that mechanical failure at elevated temperatures is generally accompanied by permanent deformation and exhibits a strong dependence on temperature, strain rate and microstructure.$ The permanent deformation can take a number of forms, in large part dependent on the microstructure of the material. For instance, it can be large-scale viscous deformation in single-phase materials at low stresses or damage in the form of localized cavitation at high stresses in materials containing glass phases.

1132

D. R. CLARKE and K. T. FABER

lDAMAGEl ! CONTROL
(CREEP DAMAGE)
1

LOG (FAILURE

TIME)

Fig. 14. Schematic diagram indicating two regimes of hightemperature failure, one of crack growth at high stresses and the other at lower stresses controlled by the accumulation of creep damage. (Redrawn after Dalgleish et al.)

between highOne important difference temperature failure and brittle fracture at low temperatures, is in crack initiation or nucleation. In the brittle fracture regime, cracks can be readily introduced (by, for instance, handling or machining) and the thrust of understanding has been to establish the conditions for crack propagation and instability. By contrast, at elevated temperatures failure need not originate from cracks that would be strength determining at low temperatures but rather can originate from highly damaged regions. A clearer understanding of what determines high-temperature fracture has really only come from studies in which the microstructure has been observed at various stages up to final failure [74-801. Such observations confirm that failure can result from either the growth of pre-existing flaws or from a local accumulation of damage. Indeed, there is a competition between damage and slow creep crack growth during deformation.

One of the consequences of this competition is that in nominally uniformly loaded materials, such as in a tensile test, pre-existing large flaws or cracks can be rendered benign and failure can result from local concentrations of damage rather than from the crack itself [75]. Thus, there appear to be two quite distinct regimes of behavior, one of stress-controlled creep crack growth and the other of creep damage controlled rupture (Fig. 14). This is borne out by recent data [15] on the flexural creep of a hot-pressed aluminum oxide reproduced in Fig. 15. At stresses below approximately 160 MPa, the materials fail by creep rupture, whereas above failure was seen to originate from large, crack-like flaws. The general conclusion that there are two distinct regimes of fracture appears to hold for both classes of ceramics of interest today. One class, exemplified by the fine-grained (2-4 pm), hot-pressed aluminum oxide, is a predominantly single-phase material of high purity. In these materials deformation at high temperatures is widely believed to take place by diffusional and dislocation flow, as is the case for most simple metals. The other class is a polyphase ceramic containing one phase that deforms much more readily than the other, and generally in a viscous manner at elevated temperatures. Examples of the latter include the vitreous-bonded (sometimes referred to as debased) aluminum oxide ceramics, and the silicon-bonded silicon carbide. This latter class of material is of particular practical import since the majority of ceramics contain intergranular glass phases separating the crystalline grains [29,81]. The films of glass may be only 10 A in thickness, as they are in, for instance, the hot-pressed silicon nitride ceramics, or sheaths of up to 1 pm in thickness in some of the vitreous-bonded refractory ceramics.

300 - Crack prop0 atian cantralle 8 0 failures zso\ B

60-

I
10-2

2 itf

I
10-l

Fig. 15. Experimental data from hot-pressed alumina plotted as failure strain, lf,, as a function of applied stress, u,. The two regimes of failure are evident. (Courtesy of Dalgleish et al.)

Fracture of ceramics and glasses l-2 pm Ai203

1133

Fig. 16. A schematic illustration of the characteristic behavior of pre-existing cracks under creep conditions. (After Evans.)

One of the intriguing findings made to date is that when either class of ceramic fails within the creep damage controlled rupture regime it appears to obey the Monkman-Grant relationship [82]. This relationship between time to fracture, 1/, and the minimum creep rate, i,. log rr+ m log i, = constant (29)

holds for the majority of metals and alloys. Whilst this finding has yet to be substantiated for a large number of ceramic materials under a variety of testing conditions, it is of considerable importance since it would allow the same design codes developed over the years for metallic alloys to be used with some confidence for the design of load-bearing ceramic structures. VI.1 Morphological observations of crack growth Observations of the crack-tip region in ceramics deformed at elevated temperatures have enabled a general picture of the failure process to be established. In a typical experiment a crack is deliberately introduced at room temperature, for instance by indentation, and its growth at high temperatures subsequently observed either directly or through a series of interrupted loadings. For different materials, the relative importance of the different modes of

damage may vary depending on the microstructure and the phase content. For instance, in the fine-grained polycrystalline alumina, cavity growth within shear bands appears to be a dominant damage mode [79], whereas in aluminas containing an amorphous phase, such deformation band damage is not seen in tensile creep [78], but only under compressional loading [74]. The behavior of cracks as a function of applied stress intensity is illustrated in Fig. 16 for a fine-grained polycrystalline material. The features shown in this figure are general, and aspects of the same behavior have been reported in the deformation of other materials. At relatively high applied stresses, there exists a well-developed damage zone consisting of both individual and coalescing cavities around the crack. The formation of such zones is generally attributed to the enhanced cavitation rate produced by the crack-tip stress field. The zone is particularly evident in those materials containing a second, viscously deforming phase, presumably due to the fact that the cavity nucleation stress is lower in this phase than in a purely crystalline material. A cavitational zone forming ahead of an indent crack in a siliconized silicon carbide deformed at 1300C is illustrated in Fig. 17. Few studies have been made on to the extent of the cavitational zone, but in one such study [74] involving a material containing a viscous

1134

D. R. CLARKEand

K. T. FABER

Fig. 17. Development of a cavitational zone about a flexurally loaded indentation crack in siliconized silicon carbide after 75 min at 1300C and 300 MPa. The cavities appear black and the silicon phase appears bright in this micrograph. (Courtesy of S. M. Wiederhom.)

phase, the zone size was found to depend on temperature as might be expected if the cavity nucleation stress depends on the viscosity (Fig. 18). How a crack advances through a cavitated material is not known with any certainty. One picture of the process is that it is a piecewise advance in which the crack remains stationary until the local damage builds up to a critical level, at which juncture the adjacent cavities merge into the crack (Fig. 16). As the process repeats, a quasi-steady-state crack growth results. Modeling of the process invariably leads to a power law dependence between the crack velocity and the applied stress intensity, as is found experimentally (Fig. 19). An alternative picture is that the crack advances continuously in the form of a fingering interface, much as was envisaged by Argon and Salama [83] for the advance of a crack in a crazed polymer. In support of this idea, examination of the crack in the polyphase materials at high magnification often reveals the presence of ligaments of the viscous phase spanning the crack walls. This is illustrated by the scanning electron micrograph of Fig. 20 of a crack in a hot-pressed silicon nitride material fractured at 1300C At lower applied stresses, the cavitational zone is smaller in extent, presumably due to the lower crack-tip stress fields and hence lower cavitational rate, until at some particular value of the applied stress the crack ceases to advance and no zone is evident. This results in an apparent creep crack

threshold, below which pre-existing cracks will not grow. The material then fails by creep damage rupture processes. Slow crack growth above the threshold applied stress intensity suggests a number of parallels to the threshold behavior observed in crack growth in the brittle fracture regime at low temperatures (Section V), not the least of which is the possibility of a sensitivity to the test environment. VI.2 Threshold stress intensity Thresholds for crack growth are not commonly observed in polycrystalline metals, so evidence for a threshold in polycrystalline ceramics is especially noteworthy. It is known that a threshold can occur for crack growth along the boundary of a bicrystal as a result of the constraint developed by the redistribution of matter on the grain boundary ahead of a growing crack. At the threshold, the constraint reduces the stress on the crack to zero, and hence, there is no net crack driving force. No such sustained build-up of contstraint can occur in a polycrystalline material since viscous flow will always act in such a way as to relax the constraint. Thus, the observation of a threshold for crack growth in ceramics must have an alternative origin. The suggestion made by Thouless and Evans [84], is that the critical stress required to nucleate a cavity ahead of the crack provides an apparent threshold stress. When the stress at the tip of the crack falls below the critical nucleation stress, the tip will sinter and retract. This

Fracture of ceramics and glasses

1135

d-3pm

t t 1
850 950
TEMPERATURE (C)

1
1050

50

Fig. 18. Cavitational zone size produced by high-temperature fracture as a function of temperature in a liquid phase sintered (debased) aluminia ceramic. The size, measured from scanning electron micrographs, is normalized by the average size, d, of the aluminum oxide grains.

corresponds to a case where the driving force for healing by sintering is greater than the local driving force for crack extension. By extension of this reasoning, it is expected that ceramics containing an intergranular glass phase will also exhibit a crack growth threshold. This has recently been reported by Jakus et al. [78]. In the simplest case, the value corresponds to that required to propagate the liquid meniscus through the microstructure, i.e. the capillarity stress (Fig. 21). In both cases, the threshold stress will be. strongly temperature dependent, decreasing with increasing temperature, and consequently, may not be detectable.7 One of the interesting findings of a recent study of crack growth in a vitreously bonded aluminium oxide, was that although a creep stress threshold existed for the growth of indentation cracks, cracks of an equivalent size but nucleated by the creep process at the same conditions of temperature and stress exhibited no threshold stress [78]. Furthert The existence of a threshold in a purely viscous material is to be expected, since as shown by Berg [85], a slot oriented
perpendicular to an applied stress field will re-orient by viscous flow until it lies parallel to the stress axis. P.C.S. 4*/r I--L

more, creep cracks grew at a faster rate than the indentation cracks. VI.3 Creep damage controlled rupture Crack initiation in ceramics when loaded in the creep damage controlled rupture regime is essentially the same problem as encountered by the metallurgical community when trying to predict creep fracture. Although, as indicated above, the materials studied to date have been found to satisfy the Monkman-Grant relationship, this affords little insight into the mechanisms leading up to failure. The observations that have been made of materials prior to rupture, however, suggest that the sequence of events leading up to failure is a general one and can be described as follows: isolated cavities form throughout the material; some grow across grain facets to become facet-sized cavities, and then stabilize; others that form on adjacent boundaries coalesce; the process continues until such contiguous coalescence cavities constitute a distinguishable crack capable of propagating to failure by slow crack growth. Although the sequence of these events is useful, for instance, in developing statistical models and ana-

1136

D. R. CLARKE and K. T. FABER

the local stresses by viscous and diffusional flow [87]. The manner in which grain boundary sliding occurs is not normally considered in developing cavity nucleation models, but such transients are generally expected to develop whenever loads readjust. This is clearly seen, for instance, in the loading of granular materials, and can be visualized by photoelastic means as shown by [88&t The next stage, that of cavity growth along grain facets under stress, has been the subject of a number of studies. In these analyses, the rate of propagation of the cavity is found to depend on the dominant mode of mass transport [89,90], since the material displaced by the growing cavity has to be accommodated by plating the material onto the adjoining grain boundaries and redistributing on the cavitating facet by altering the shape of the cavity as it grows. In ceramics containing an intergranular glass phase along their grain boundaries, the sequence of events by which individual cavities form and link up is similar [74,91]. Cavities are found to nucleate within the intergranular phase, principally at three and four grain junctions, and then grow within the intergranular material by viscous flow of the glass into adjoining channels until they deplete the pockets of all but a wetting film of glass. The glass-vapor meniscus then advances into the two grain junction channels until meeting with a similarly growing cavity at an adjacent three grain junction. The cavity meniscus may advance across the two grain channel in either a uniform manner or become unstable and exhibit finger-like perturbations [74,91]. Such features could result from nonuniform thickness of the two grain channel or flow instability as the cavity grows. The latter would be an example of the finger-like instabilities analysed by Fields and Ashby [92], and more recently for the case of cavity advance by Marion et al. 1911. In essence, the principal differences between the various ceramics studied appears to be the mechanisms (and rates) by which damage forms, grows and accumulates. Cavity growth in ceramics with an intergranular phase is less straightforward than in the pure, polycrystalline materials since there are three distinct mechanisms of mass transport, viscous flow, atom migration and solution-reprecipitation, by which the matter can be redistributed. The dominant one depends very much on the material and temperature. Some of the other differences in behavior between the two classes of material probably result from the higher temperatures that can be tolerated by the single-phase materials. At temperatures of 1300 and 14OOC, deformation of aluminum oxide can occur by a number of mechanisms, including bulk and grain boundary diffusion, activated slip and climb, and grain boundary sliding. Thus the stresses in the vicinity of any crack tip can be relieved by any one or combination of these mechanisms, and blunting can occur. In the debased materials, especially at the lower temperatures, the stresses can be accommo-

10-7

L 5
c a s 2 :: 2 10-6 -

\
POWER-LAW GROWTH

0 10-g 0.3 1 0.4 K/Kc 01 , I 0.6 8 I 0.6 L I

Fig. 19. Crack tip velocity as a function of normalized stress intensity factor, K/K,, at 1300 and 1400C for indentationinduced cracks. Hot-pressed aluminum oxide material. (Redrawn after Blumenthal.)

lytical expressions for creep strain and time to failure, as has been presented by Evans and Rana [WI, it does not address the detailed mechanisms underlying each event. These can be very different and depend in detail on the microstructures of the materials. In the single-phase, high-purity materials, such as the polycrystalline aluminas referred to in the previous section, cavities nucleate primarily near triple grain junctions and then grow along grain facets by the migration of atoms along the cavity surface and grain boundaries. Both the nucleation and growth of the cavities is strongly dependent on the applied stress, the constraint afforded by the surrounding creeping material and the properties of the material, namely surface energy, surface diffusivities, and grain size. Nucleation of a cavity requires that a sufficiently high stress is attained that vacancies can coalesce into a size that exceeds the critical nucleation size. The requisite nucleation stresses are so high that it is normally considered that they can only be achieved by local stress concentrators. One proposed mechanism is the generation of a transient stress by grain boundary sliding, which requires that the sliding process has a very short relaxation time compared with the time for the surrounding material to relax
t Such transient rearrangements can be seen in the popular childs game, Boobytrap.

Fracture of ceramics and glasses

1137

Fig. 20. Ligaments of glass spanning the walls of a crack in creep deformed (HS130) silicon nitride material. Double torsion test. 1400% (Photomicrograph courtesy of N. J. Tighe.)

most readily within the viscously deforming intergranular phase by cavitation, solution-reprecipitation and viscous flow. As a result the aluminium oxide grains are less easily deformed. VI.3.1 Damage localization. So far we have discussed how cracks form by the growth and coalescence of cavities. One important feature of this process that is coming to light as a result of tensile tests at high temperatures is the strong co-operative nature of the cavity accumulation resulting in local regions of enhanced damage; a phenomenon of damage localization. This localization phenomenon is illustrated here by work on crept siliconized silicon carbide [93], since the cavities in this material are readily observable in the optical and scanning microscopes. At low stresses and short times, the cavities appear to form at random sites throughout the material. At longer times and/or higher stresses there is a propensity for cavities to form in the vicinity of ones formed earlier, as well as new ones being formed at random. This indicates that the growth of a cavity, and in particular its growth linking it to another, appears to be biased by the presence of other adjacent cavities in such a way that together they form local concentrations of cavities. This localization of cavitation continues and distinct dated

rows or bands of discrete cavities form, as illustrated

in Fig. 22. Such a row of cavities can, of course, act as a critical flaw when loaded at low temperatures, but they are also expected to limit the creep rupture life by localizing the applied stress across the reduced section of uncavitated material. How such bands of cavities form, and why a freshly nucleated cavity forms just ahead of an existing row of cavities is not understood, but clearly involves the elastic interaction stresses between cavities, the shape of the clusters of cavities, and stress effects on cavity nucleation. Another related observation that is expected to be of importance is that the rows of cavities that form appear to lie at regular distances apart in a direction perpendicular to the applied stress axis. The phenomenon has yet to be analyzed, but there may well be a direct analogy with the problem of matrix cracking observed in fiber composites (section VII.3.2) and indeed mudcracking. VI.3.2 Role of microstructural inhomogeneities. Just as in the brittle fracture regime, where local inhomogeneities can act as stress concentrators and lead to low strengths, so large-scale inhomogeneities can lead to premature high-temperature failure either by nucleating cracks or localized damage. A number

1138

D. R. CLARKEand K. T.

FABER

Fig. 21. Scanning electron micrograph of finger-like perturbations in advancing cavity fronts on the fracture surface of a liquid phase sintered aluminum oxide ceramic. Material deformed in creep, frozen under the load and subsequently fractured at room temperature.

of such inhomogeneities have been identified in the work of Dalgleish et al. [76], which serves to show that even the simplest of ceramic microstructures, a hot-pressed single-phase, polycrystalline material, can contain strength degrading inhomogeneities. They showed that regions of unusually large grains, second-phase particles, impurity phases, hot-pressing flaws, and even pockets of isolated amorphous material could all act as nucleation sites for local damage and/or fracture. Very little quantitative analysis of the role of such heterogeneities has been performed to date despite the obvious importance of such defects. The one exception is the analysis by Hsueh and Evans [94] of regions of above-average grain size in enhancing the local stress. The basis of their calculation is that the viscosity has a strong dependence on grain the size (for both NabarreHerring and Coble creep). Thus, the large-

grained regions behave as regions of high viscosity embedded in a less viscous matrix, leading to the generation of mismatch stresses which can attain levels twice that of the applied stress. Remarks The foregoing has, necessarily, been a qualitative description of the high-temperature fracture of ceramics. There is still some way to go before the high-temperature fracture of ceramics is properly explored and systematized, and further still before rigorous mechanics analyses are developed for predicting the response of the various materials to high-temperature crack growth. Some of the concepts involved+onstrained cavitational zones, ligament bridging and microstructural scale-are, however, similar to those invoked in analyzing brittle fracture. There is, thus, an expectation that, on the

Fracture of ceramics and glasses

1139

Fig. 22. A row of cavities (appearing dark) formed perpendicular to the applied tensile stress direction in a silicon&d silicon carbide material. 100 MPa tensile stress at 1330C for 140 h. The tensile stress is in the horizontal direction. (Courtesy of S. M. Wiederhom.)

descriptive mechanics level at least, considerable progress can be made once the detailed deformation mechanisms accompanying crack growth have been identified by microstructural observation.

VII. 1 Crack dejection

VII. TOUGHENING

OF CERAMICS

Can ceramics ever be sufficiently fracture resistant that they can be used as structural engineering materials? The drive to make tougher ceramics has focused attention on the microstructural mechanisms that lead to enhanced toughness, and has led to a better appreciation of the physics of toughening in materials as diverse as ceramics, polymers and steels. As will be described below, the interaction of a propagating crack with inhomogeneities in the microstructure (crack deflection), with transformable particles in a process zone, and with particles that span the crack can lead to increases in the fracture toughness of a brittle material. These contributions amount to microstructural strategies for altering the individual terms in eqn (2). For instance, by causing a crack to deflect from a pure mode I loading condition, the applied crack driving force, d U, /dc, is reduced. Similarly, the crack resistance term, dU,/dc, is increased by energy dissipation or bridging forces.

The general concept of causing a propagating crack to deviate out of the crack plane and thereby increasing the fracture toughness of a brittle material has been suggested by many investigators over the years, but a quantitative analysis of the attainable toughening was not presented until the work of Faber and Evans [95]. Until that analysis, the prevailing view in the literature was that the enhancement in toughness derived from the increased area of the fracture surface, as might be expected from a superficial examination of the Griffith equation. Faber and Evans argued that when a mode I crack is deflected out of plane, the loading of the crack changed to one of mixed mode character. Correspondingly, the energy release rate would not scale simply with the area of the fracture surface. Crack deflection occurs whenever the crack path adopts a locally least resistant path. This may occur as a result of the presence of a less tough grain boundary, the presence of spatially varying residual elastic stresses, or as a result of two or more phases in the microstructure having different elastic moduli. Crack deflection is illustrated in Fig. 23 with scanning electron micrographs of cracks in three different ceramics. In the top figure the crack is hardly

1140

D. R. CLARKEand K. T.

FABER

Fig. 23. Profiles of cracks seen to be deflecting through the microstructures of three materials. Top, silicon carbide sintered with boron. Middle, silicon carbide hot-pressed with aluminum oxide. Bottom, hot-pressed composite of silicon carbide and monoclinic zirconia.

deflec:ted, whereas in the bottom figure, of a two: material, the crack is far from planar.

As is shown may propagate

schematically in Fig. 24, the crack so that different segments of the

crack are deflected with respect to one another, i.e. the crack segments may become non-coplanar. The overall crack driving force is then the sum of the local contributions from the Mode I, Mode II (p lane

Fracture of ceramics and glasses

1141

a) CRACK TILTED ATANGLE

8 ABOUT Z AXIS

y
t/,

x
Z

->

---

b)

CRACK TWISTED

AT ANGLE

4 ABOUT

X AXIS

Fig. 24. Schematic diagrams of typical crack deflections: (a) tilt and (b) twist of the crack front around spherical particles.

shear) and Mode III (anti-plane shear) loading acting on each segment. When the crack is caused to tilt by some angle, 8, with respect to the normal to the applied Mode I load, the crack driving force is

the crack paths and the greater the fraction of highly deflected crack segments the larger the toughening. As is intuitively expected, grain shapes having higher aspect ratios are more effective at reducing the crack driving force than those with equiaxed morphologies. Similarly, the higher the concentration of deflecting particles the larger the magnitude of attainable toughening. If the crack is large enough to sample the average microstructure, and in turn, an adequate distribution of deflection angles, the measured fracture toughness is expected to be scale invariant. Hence, the full extent of crack deflection toughening is realized when crack propagation has reached the steady-state regime. The measured fracture toughness of a number of systems, including glass-ceramics, hot-pressed silicon nitride, silicon carbide and zirconia, have now been analyzed [95-971 and the contribution due to crack deflection related to the volume fraction and aspect ratios of the deflecting phase. In these materials significant toughness increases (SO-80%) have been attributed to the crack deflection process (Fig. 26). One of the features of the crack deflection model is that the extent of the toughening is expected to be strongly dependent on the stress state. In the Faber-Evans analysis the undeflected crack propagates perpendicular to a tensile applied stress, so the effective driving force is reduced wherever the crack is deflected away from the tensile axis. In a biaxial stress field, however, deflection of the crack will not

reduced to EG/( 1 - v) = [cos3(0/2)K,]* + [sin(B/2)cos2(0/2)K,]*. (30)

A similar equation pertains for a segment under going a twist. As this is a mixed mode problem, the individual stress intensities for the three modes cannot simply be summed, but the strain energy release rates for the modes can be. The calculation of the toughening due to crack deflection becomes a statistical problem involving the computation of the driving force averaged over all the possible angles of deflection for any particular microstructure consisting of the volume fraction of deflecting particles and their morphology. The resulting toughening, GT, may be written in terms of the calculated average driving force, <G > : G:=G/<G >G,, (31)

where G and G, represent the strain energy release rate and toughness of the deflection-free material respectively. The principal findings of the analysis are summarized in the curves of Fig. 25. The more tortuous

--h--e+
VOLUME

FRACTION,

V,

Fig. 25. Calculated increase in strain energy release rates as a function of the volume fraction of crack deflecting particles for three different shapes of the particle.

1142

D. R. CLARKE and K. T. FAISFX that the stress on the crack tip now follows the solid line, so that outside the process zone, the stress field is determined by the applied stress whereas inside it is lessened by the extent of the shielding. To date, two examples of process zone toughening have been identified in ceramics. The most thoroughly analyzed is transformation toughening, which results when the martensitic phase transformation from tetragonal-tomonoclinic zirconia is triggered by a propagating crack. The other example is of stress-induced microcracking, cracking at grain boundaries in the vicinity of a main crack. The shielding action of the process zone can be understood through the use of a series of imaginary cutting and welding operations originally introduced by Eshelby in his 1957 analysis of the inclusion problem [99]. This particular thought experiment forms the basis of the McMeeking-Evans model for transformation toughening [loo] and has subsequently been extended to treat the case of stressinduced microcracking (1011. From the Eshelby calculation of the stresses in the zone, the crack closure force is calculated by integrating the effect of the constraint forces on the crack-tip stresses. The thought experiment is reproduced here for the case of a matrix material containing a concentration of particles that can undergo a stress-induced dilatational transformation, cr. The sequence of steps is represented schematically in Fig. 28. In the first step, a zone of material, large enough to contain a representative concentration, V,, of particles, around the crack tip is cut from the material so that it is free of any stress and constraint. The particles within the zone are now imagined to transform, so that the size of the zone expands by a strain of V+r. The zone is now too large to fit back within the hole in the matrix it came from. In order to replace it in the hole (step II) a pressure must be applied to the surface of the zone, where the pressure is simply given by
p=V,cTEz 3(1 - 2v)

100 t

_L

t 0

I 2

I 4

I 6

ASPECT RATIO. R

Fig. 26. Measured fracture toughness of a series of hotpressed silicon nitride ceramics having different aspect ratio grains of fi silicon nitride.

necessarily result in a decrease in driving force, as can be seen from examination of eqn (30) for the case of a simple tilt. Thus, the degree of toughening resulting solely from crack deflection processes is predicted to be less pronounced as the stress state becomes more complex. This is substantiated by recent fracture toughness measurements (by indentation strength in bending) on glass and a fine-grained A&O3 [98]. Under biaxial stressing, the strength of the A&O, was consistently lower than the corresponding uniaxial tests for a wide range of crack sizes. The glass, which has no microstructural features for causing crack deflection, exhibited the same strength under both modes of loading [98]. By extrapolation to three dimensions, a triaxial stress would be expected to increase the effective crack driving force further. VII.2 Process zone phenomena The most general description of process zone toughening is of microstructural changes, induced by the stress field of the propagating crack, altering the net stress at the crack tip, and reducing the effective crack driving force. In essence, changes in a localized region around the crack shield it from the externally applied stress field, thereby necessitating a higher applied stress to further propagate the crack. The effect of the process zone shielding on the crack-tip stress field is shown schematically in Fig. 27. In the absence of the process zone, the crack plane stress is proportional to the applied stress intensity, K,, and falls off as l/Jr with distance, r, from the tip. The action of the shield is to decrease the stress around the crack tip thereby shifting the stress function to the lower dashed line on Fig. 27. The net result is

Crack-Plane Stress

II I
i 1:

\ 1 \ .
--__ K a-J -----___ I +I Process Zone b-w-m_

u$&
l _I Crack

Fig. 27. Schematic diagram of crack-tip stress field ahead of


the crack in the absence of a process zone (upper dashed curve) and in the presence of a shielding zone (lower curve).

Fracture of ceramics and glasses


ESHELBY THOUGHT PROCESS APPLIED MLATANT TRANSFORMATION TO

1143

CRACK

----------.
1 I I :I: ,-. ._t c_i ,-. \. ::.

.\ )

c>

REGION TO TRANSFORM Vt .ET.EZ

3 (I-2l/)

CONSTRAINT FORCES

-17 p ; ;
IMPORTANT MATERIALS

! c

2.
\
RELAXATION FORCES

PARAMETERS

ST,

Vt,

E,.E,

Fig. 28. Schematic diagram illustrating the sequence of steps in the Eshelby thought experiment for calculating the constraint stresses acting as a result of a dilatational transformation.

where E, is the bulk modulus of the zone material.

This pressure corresponds to the constraint stresses that would be required to prevent the size of the zone increasing when the particles increased their volume on transformation. In the third step of the Eshelby experiment, the pressurized zone is placed back into the hole in the matrix from which it originally came, and the pressure forces nullified by the application of equal but opposite surface tractions. The zone now relaxes until the net stress on the surface of the zone becomes zero, which leaves the interior of the zone stressed. (If the zone had been chosen to be spherical in shape, then the stress inside the zone would have been a pure hydrostatic compression.) In general, the stress in the zone is a complicated function of the shape of the zone the concentration of particles, the transformation strain and the elastic properties of the zone and surrounding material. The more rigid the material surrounding the transformed zone the less the relaxation of the zone from its pressurized state PCS. 48,1,--M

in step III, and correspondingly the larger the magnitude of the stresses in the zone. Having established the stresses within the zone, the final step of the McMeeking-Evans calculation requires assessment of the closure forces produced on a crack by this contraint stress field. They use the weightfunction method of Buecker [102], in which the stress intensity factor is the integral of the zone tractions over the surface of the zone but weighted by their distance from the crack tip. For the case of steady-state crack growth where the zone length is large and there is no change in elastic modulus, the McMeeking-Evans analysis leads to an expression for process zone toughening due to a purely dilatational transformation given by: AK = -0.22V,eE,/(h)/(l -v), (33)

where h is the size of the zone. The negative sign in this expression indicates that a decrease in stress intensity results from a positive transformation vol-

1144

D. R. CLARKEand

K. T. FABER

tune change. The measured fracture toughness, K,,-, is then equal to the fracture toughness of the material in the absence of the process zone, KE, plus that due to the transformation:
K,c = Kg + 0.22 V,ET E,/(h)/( 1 - v).

(34)

In the process zone description, the size and shape of the zone is given by the condition that the stress at the zone boundary is equal to that required to nucleate the crack shielding process. In the case considered by McMeeking and Evans this corresponds to the critical hydrostatic stress, u,, for nucleating the transformation. For the particular assumptions that all the particles within the zone transform irreversibly, the zone size, h, can be written h = J(3)(1+ v) K, * 12x a,.

[3

(35)

The Eshelby approach is especially attractive because many of the essential ideas of process zone toughening are clearly brought out: the central idea of the surrounding material constraining the transformation and building up stresses that act as a closure force on the crack; the notion that the zone is defined by the critical value of the crack-tip stress field required to nucleate the shielding process; the dependence on the size of the zone; and the materials parameters. Although not specifically described above, the approach also identifies the important role of the wake region behind the crack in developing R-curve behavior. The increase in fracture toughness can also be calculated in other, equivalent, ways. One, presented for the toughening due to a purely dilatational transformation, is to explicitly calculate the energy change accompanying the transformation [103]. A second approach is to calculate the change in stress intensity at the crack given the constitutive equation for a volume element of material subject to the stress field of the crack [104]. Both are very powerful analytical techniques for calculating the increase in fracture toughness. They are based on the hysteresis in the stress-strain curve of the process zone material as an element of material is loaded up as the crack stress field approaches it and subsequently unloads as the crack passes by. VII.2.1 Transformation toughening. A paper [105] published in 1975 with the intriguing title Ceramic Steel? proposed that the toughness of a zirconia ceramic was the result of the incorporation and retention of the (metastable) tetragonal polymorph of ZrO,. The authors argued that as in iron-based steels, the martensitic transformation (from the tetragonal to the monoclinic form) of zirconia in the stress field of a crack conferred increased toughness to the material. Based on this idea, an entirely new family of ceramic materials has been developed in the succeeding years which exhibits unprecedented

(for brittle solids) mechanical properties by incorporating the tetragonal zirconia into matrices of cubic zirconia, other oxides, carbides or nitrides. Three features of the tetragonal-to-monoclinic phase transformation in zirconia are unusual: (1) it is martensitic and hence, diffusionless, (2) in going from the high-temperature tetragonal form to the lowtemperature monoclinic form the unit cell increases in volume by 45%; (3) the transformation temperature and, to a lesser extent, the volume change is altered by alloying with a suitable oxide such as Y,03, CaO, MgO and CeOr. This last feature has been the basis of much alloy design and means that by appropriate additions of the stabilizing oxide, the transformation and the toughness can be determined by processing. The models of transformation toughening [106] make specific predictions as to how the contribution, AK,,-, depends on the stress criterion for the transformation, and hence the shape of the transformation zone, and on the relationship to the square root of the transformation zone size, h. The earliest studies of the transformation zone were directed towards proving whether a zone of transformed material existed around a propagated crack thereby establishing that stress-induced transformation had occurred. In pursuing such confirmation, three distinct techniques were adapted for measuring the transformation zone size: transmission electron microscopy; X-ray diffraction; and Raman microprobe spectroscopy. The transmission electron microscopy observations, such as that reproduced in Fig. 29, provided direct evidence for the existence of a transformation zone, especially when the material could be strained in the microscope and the advance of the crack followed [107]. The non-destructive techniques of X-ray diffraction [108, 1091and Raman microprobe spectroscopy [l lo] were used by other investigators to measure the zone size. Not only were these more rapid but because they could be used on bulk samples they circumvented many of the potential thin-foil effects associated with transmission microscopy. By using the bulk analysis methods, the fracture toughness due to transformation could be compared with the measured zone size, and with the predictions of the models. One such comparison is shown in Fig. 30 for the results of two PSZ materials. In addition to the experimental data, the figure presents the straight-line predictions corresponding to two assumptions for the criterion for the transformation to occur. The line marked AK, is for the assumption that only the dilatation accompanying the transformation contributes to the crack shielding and the zone profile is dictated by the hydrostatic crack tip field. The second line, marked AK,, or AK,, denotes the toughening predicted for a uniaxial transformation strain or from a pure dilation with a zone profile bounded by n/3 shear bands, AK,. According to the comparison provided by Fig. 30, dilatation within a zone boundary prescribed by a

Fracture of ceramics and glasses

1145

Fig. 29. A zone of transformed tetragonal to monoclinic zirconia particles on either side of a crack in a partially transformed zirconia (PSZ) material. The monoclinic particles in the zone appear bright in this

dark-field transmission electron micrograph. The crack is rather indistinct here but passes between the points marked F in the figure. (Courtesy of A. H. Heuer.)

critical hydrostatic stress cannot account for more than one-third to one-half of the observed toughening. Separate transmission electron microscopy observations of the alignment of twins in the monoclinic particles suggest that the transformation criterion of a critical uniaxial transformation strain is also not an appropriate one. This leads one to conclude that most, but not all, of the measured toughness can be attributed to a purely dilatational transformation within a zone whose shape is determined by a critical value of the crack-tip shear stress field. The mechanistic origin of the remaining toughening is variously attributed to crack deflection processes, stress-induced microcracking and ligament effects. Up until this point, steady-state transformation toughening has been assumed in which the length of the zone behind the crack tip (the wake) is sufficiently large that is length does not enter into the analysis.

t This conclusion holds for any arbitrary zone shape when the transformation is a dilatational one. It does not hold, however, when the transformation is biased in some way, for instance by autocatalytic processes, rather than by the crack-tip stress field alone.

However, one of the interesting features of all process zone toughening mechanisms is that the magnitude of the attainable toughening increases as the zone and its wake grows (with crack propagation) leading to potential R-curve effects. Analysis of the development of a dilatational transformation zone as a crack advances has brought fresh insight into how the process zone toughening develops [loo, 1061. Initially a crack in an untransformed material (produced by, say, heating the materials back above the reverse monoclinic-tetragonal temperature after inserting a crack) is considered. As the crack is loaded, a zone of transformed material ahead of the tip (the frontal zone) forms and spreads out until the stress at the crack tip reaches the local fracture condition. As the crack advances, transformed particles, originally in front of the crack, are now left behind thereby building up a wake region of transformed material behind the crack tip. This partial zone grows in length until a steady state is attained and the toughness reaches an asymptotic value. Before the crack advances, the net toughening provided by the frontal zone is calculated to be zero despite the formation and growth of the zone with increasing stress.7 This surprising result is now known to be a consequence

1146

D. R. CLARKEand K. T. FABER

02

0.5 t,K

I.0 s pm fi

I .5

Fig. 30. Comparison of toughening enhancement for partially stabilized zirconia (PSZ) with measured transformation zone size and concentration. (Courtesy of R. M. Cannon.)

of the path independence of the J-integral around the frontal zone [104]. As the crack advances, the wake region of transformed particles is created behind the tip and a net toughening develops with the extent of the wake. The toughening can be. considered to develop from the constraint provided by the untransformed material around the zone as outlined above. Equivalently, it can be obtained from the appropriate strain energy release rate equations [104, 1061.This indicates that the toughening is determined by the transformation within the wake alone. For a dilatational transformation, the toughening reaches the asymptotic value given by eqn (34) when the crack has advanced a distance of 5-6 times the zone width, h. Recent measurements have shown that transformation toughened zirconia ceramics can exhibit such R-curve behavior, but it remains to be seen whether the extent is as predicted on the basis of the relative width of the zone. Whilst it is conceptually convenient to analyse transformation toughening in terms of the mechanics of constraint and the crack-tip stress fields, the increase in fracture toughness derives from the difference in total free energy of the ceramic between when the zirconia is in its tetragonal and monoclinic phases. This difference may be termed, AG. From thermodynamic considerations,
AG- = (T - M,)AS,

materials is the control of the magnitude of the chemical free energy, AG, and this is most conveniently achieved by the concentration of stabilizing oxide. This affects the value of the martensite start temperature, M, in the above equation. As indicated by eqn (36), transformation toughening can only occur for T > M,. As T is decreased towards M,, the magnitude of the toughness increases until a maximum is attained at T N M,. Below the M, temthe transformation proceeds autoperature, catalytically in cooling of the sample before it can be fractured. This predicted behavior with test temperature has been reported by a number of workers [l 11, 1121 for Ce02, Y,O,, and MgO solute stabilizers. One of the interesting characteristics of the transformation toughened ceramics is that the Griffith relationship between strength and fracture toughness does not hold [113]. This is shown in Fig. 31 for two partially stabilized zirconia ceramics and a composite ceramic of alumina with zirconia v(Al, O9)-PSZ]. In sharp contrast to the normal behavior of ceramics, the toughest materials exhibit the lowest fracture strengths, while the strongest zirconia materials have fracture toughnesses comparable with only Al,O, or S&N., [ 1131.This behavior permits two possible explanations [114]. The first is that the strength of the transformation toughened materials may be R-curve limited. The second is that the strength is determined by the stress required to nucleate the transformation. In this latter case, the strength may be thought of as being yield limited in much the same way as in many metallic alloys.

Mq-PSZ

A Y-PSZ 0 Y (AI,O,)-PSZ

LImIted 0 Stress 5 lntenslty IO Factor, K,,

Strength 15

, MPo

v-iii

(36)

where T is the test temperature, M, is the temperature at which the tetragonal particle will spontaneously transform (AG = 0) and AS is the associated entropy change. The basis of microstructural design in these

Fig. 31. Strength against measured stress intensity factor plotted for three zirconia materials. The curve cc is the critical stress to initiate the transformation as a function of stress intensity factor. The thin lines radiating from the origin are the anticipated strengths for the indicated critical flaw sizes, C,. The vertical dashed lines indicate the transition from flaw controlled to transformation limited strength. (Figure courtesy of M. V. Swain.)

Fracture of ceramics and glasses

1147

VII.2.2 Stress-induced microcrack toughening. The second example of toughening by the formation of a process zone is where the material microcracks in the stress field of the propagating crack. Such a phenomenon has been invoked on numerous occasions in the ceramics literature over the last 30 yr, but only rarely has the existence of stress-induced microcracking actually been substantiated experimental1y.t However, in the last few years a number of studies have confirmed that crack-tip stress-induced microcracking occurs, as shown for instance in Fig. 33. There have also been studies indicating that such stress-induced microcracking leads to a consequential increase in fracture toughness. Recently, Ruhle et al. [120], were able to measure both the microcracking zone size and the toughening in an Al,O,-ZrO, material, thereby confirming for the first time a causal relation between stress-induced microcracking and toughening. STRESS INTENSITY KI .(MPo m? The mechanics of toughening due to stress-induced microcracking are similar to that of transformation Fig. 32. Slow crack growth behavior in Al,O,-2Ov/ o(ZrOz.xY,O,), showing the effect of increasing the frac- toughening although the detailed physics of the two ture toughness on the resistance to slow crack growth. Data processes are quite different. In both cases there is a for two yttria concentrations (x = 1 and 3 mole %) is volume change and it occurs once a critical stress is shown. (Redrawn after Becher.) reached in the material. For this reason the methods of calculating the toughening are extensions to those developed for analyzing transformation toughening, and the general form of the toughening can be written The stress-induced transformation affects not only the steady-state fracture toughness, but also, as as in eqn (34). Microcracking typically occurs to shown by Becher [115], increases the materials re- relieve the residual stresses that are created as a result of thermal expansion anisotropy in single-phase masistance to slow crack growth. Becher showed that terials or thermal expansion mismatch in two-phase adding partially stabilized zirconia to alumina shifted the v-K curve to higher values of applied stress or multiphase systems. The opening of the microintensity without appreciably altering the value of the cracks provides the dilatation that characterizes the slow crack growth exponent (Fig. 32). Whether it also process, analogous to the volume expansion in transformation toughening. Thus, for residual stresses resulted in an increase in the threshold was not created as a result of thermal expansion anisotropy, measured. An important implication of these observations is that the stress-induced transformation is the microcracking strain, 0, is related to the temperature differential, AT by 0 = Aa AT and replaces activated at applied stress intensities below K,c, and 6 r in eqn (34). The term Act is the thermal expansion the compressive tractions which result from the transmismatch and AT is the temperature difference beformation are also sufficient to retard stress corrosion tween the strain point and the temperature of interest. cracking. This may be examined in the light of surface As in transformation toughening, the size and shape forces on the v-K curve. The transformation provides of the microcracking zone depends on the stress an auxiliary force to alter the shape of the crack and prevent crack growth at stress intensities where the criterion for the nucleation of the microcracks. In the case of microcracking to relieve residual thermal crack growth would otherwise occur in materials not transformation toughened. This was documented in stresses, the criterion might be when the local driving force, for instance the sum of the thermal expansion Al,O,-ZrOz containing sufficient stabilizer to prevent stresses and the crack-tip stress, attains a critical the transformation from taking place. value. One significant difference between stress-induced microcrack toughening and transformation tought The term microcracks generally refers to the for- ening is that the former is accompanied by a reducmation of grain facet, or sub-facet, length cracks along the tion in the elastic modulus in the process zone. The grain boundaries in polycrystalline ceramics. Microcracking effect of this modulus change can, as before, be throughout a material as a result of thermal shock or on cooling is well established; it can be measured, for instance, considered either in terms of the constraint from the by its effect on elastic modulus or thermal conductivity [ 1161 surrounding material [loll or from the constitutive and there exist well-defined critical grain sizes below which behavior of the zone material [121]. In terms of the a material will not microcrack [117-1191. The issue has constraint picture, the reduction in modulus of the always been whether materials microcrack in a crack stress zone effectively makes the surrounding material apfield, and whether this can affect fracture toughness.

1148

D. R. CLARKEand K. T. FABER

Fig. 33. Microcracks in a yttrium oxide stabilized zirconia ceramic observed by defocus [( (b) over-focus] imaging in the transmission electron microscope.

and

pear more rigid, the constraint pressure is larger, and consequently the apparent toughening increases. In terms of the constitutive stress-strain behavior, microcracking has the effect of altering the stress--strain curve for the near-tip region. Since the toughening may be thought of in terms of the area encompassed by the stress-strain hysteresis, the

microcracked solid unloads along a line of reduced modulus. One result of these models is the prediction that for small concentrations of microcracks and small microcracking strains, the modulus reduction on microcracking determines the degree of toughening rather than the dilatation perse. Establishing whether crack-tip stress induced

Fracture of ceramics and glasses

microcracking has occurred has proven to be extremely difficult, both because the microcrack zone sizes are believed to be small (a few pm in height) and because the crack openings (approximately l-20 nm) are rarely visible without the aid of the transmission electron microscope. (In fact new microscope imaging methods have had to be developed to observe them [ 1221.)To date, stress-induced microcracks have only been observed in zirconia materials, which raises the question as to whether they are only visible because they are wedged open by the permanent dilatation of the zirconia particles. Indirect evidence for stress-induced microcracking has been established in a glass-alumina system using sensitive density measurements [123]. It might be remarked that there is a counterpart to microcracking in certain brittle polymers. In rubber toughened blends of poly (2,6-dimethyl-1,6phenylene) oxide and polystyrene, microcrazing occurs in the vicinity of propagating cracks and has been found to be responsible for an increase in the materials fatigue crack resistance [124]. In these materials, which are transparent when undamaged, the craze zones become opaque. Since the zone sizes are several hundred pm in size they can be easily measured. The authors found that, as predicted by eqn (33) the size of the zone is linearly proportional to the square of the stress intensity. VII.3 Crack bridging phenomena One general approach to increasing the fracture resistance of a brittle solid is to retard or prevent the crack opening as the crack is driven through the material by bridging the crack walls. As the bridges are loaded they provide a restraining force against further opening of the crack and thereby act as a crack closure force. Such crack bridging, sometimes referred to as ligament or fiber toughening, is a widely occurring phenomenon that can act over a wide range of scales. The smallest-scale manifestation of the phenomenon is probably in the role of surface and intermolecular forces. Even though these may be as small as van der Waals forces and act over only a few angstroms, they have appreciable effect at stress intensities close to the threshold [29]. It includes the effect of unbroken grains in singlephase polycrystalline ceramics, fiber reinforcements in composites and ductile metal particles in otherwise brittle matrices [125]. At the largest scale, the same phenomenon is operative in the mechanical bracing of sub-critical cracks in structures [126]. A general description of the contribution of bridges to the crack driving force can be expressed in terms of the change in strain energy release rate determined from energy balance (J-integral) evaluated for a path around the bridging zone [125]: AC =f %e&)du, s0 (37)

MATRIX
CRb2K

7-l-l-r
A t-+---

Fig. 34. Schematic diagram of crack bridging zone due to intact fibers of radius R acting over a length D behind the crack tip.

where u is the crack opening, ub is the normal stress in the ligament acting on the crack walls, u, is the crack opening at the end of the bridging zone and fis the fraction of the crack plane containing bridges (Fig. 34). For the special case in which the bridging zone length, D, is small compared with the overall crack length, the corresponding contribution to the stress intensity, AK, is given by
AK = -,/(2/Q DQbodx. s 0 J(x) (38)

Solution of these equations requires a knowledge of how the bridging force depends on the crack wall displacement, which in turn necessitates a knowledge of how the shape of the crack depends on the net crack driving force. This requires the mathematical solution of the nonlinear integral equations [127J. For an embedded straight crack these are a(x)=4 (1 -vZ) zE x 1 __ s o J(s)

OS s

b, - adt)l dt Jofs

$)?

(39)

where s and t are dummy co-ordinates. net stress intensity is given by

In turn, the

w
One of the characteristic features of crack bridging phenomena is that R-curve behavior is always predicted. For small crack sizes, few bridges, if any, are intersected so the contribution from them is small. As the crack extends, it encounters more bridges and so their effect becomes larger. Finally, a plateau toughness is attained under steady-state conditions when the number of bridges being broken behind the crack equals the number being created as the crack

1150

D. R. CLARKEand K. T. FABER

Fig. 35. A region behind an advancing crack tip before (left) and after (right) the crack has been advanced further. Grains (ligaments) spanning the crack are seen in the right-hand micrograph. Optical micrographs. (Courtesy of R. Steinbreeh.)

tip advances. The existence of R-curve behavior recommends this class of toughening mechanisms for those technical applications where flaw tolerance is required. VII.3.1 Bridging in polycrystalline ceramics. When a polycrystalline ceramic is fractured and then examined under the microscope the fracture suface is commonly seen as a mixture of transgranular and intergranular fracture. From the fracture markings it can generally be concluded that the crack does not advance through the material as a straight front, and therefore it is deduced that the crack opening can be locally pinned behind the crack tip. Other evidence for unbroken grains acting to restrain the opening of

the crack comes from direct visual observation of crack growth, from the jerkiness of the loaddeflection curves when measured with fast resolution, and from the remnant strength of the samples after failure [34]. An example is shown in Fig. 35, in which the same region behind a propagating crack in a coarse-grained A&O3 is photographed at two succesive positions of the crack extension. The most convincing evidence for the efficacy of unbroken grains acting as ligaments enhancing fracture toughness and for causing R-curve behavior comes from the work of Steinbrech et al. [128, 1291. Their work on polycrystalline alumina takes on particular significance since it clearly demonstrates

Fracture. of ceramics and glasses

1151

al

0.2

I 0.4 a/w

I 0.6

I 0.6

I I

01 0.3

I 0.4

0.5

I I 0.6 0.7 a/w

00

I 0.9

1
I

Fig. 36. Crack resistance (R-) curves for a polycrystalline aluminum oxide material. Top, as a function of notch depth. Bottom, the R-curves (individual symbols) for specimens where the cracked region behind the tip was machined away. The hatched area-represent the R-curve behavior for non-machined specimens. (Redrawn after Knehans and Steinbrech.)

the role of processes operating behind the crack tip. Their observations, reproduced in Fig. 36, demonstrate that the R-curve in a polycrystalline alumina was due to events behind the crack tip as distinct from the formation of a microcrack process zone ahead of the crack.? They showed this by machining out the crack to just behind the crack tip and then reloading. As shown in Fig. 36, the crack began to extend at much lower values of the driving force than prior to the machining out of the crack bridging region. The picture that emerges of the bridging role of unbroken grains may be summarized as follows. At small crack sizes the probability of intersecting a grain that will subsequently act as a ligament is small and the fracture toughness will be given by the

t At the time it was suggested that their experiments ruled out the possibility of microcracking in a process zone as being responsible for the increase in toughness. However, as it is now known that microcracking in the wake of a crack can result in toughening, the removal of the wake region does not allow a separation of the effects of microcracking in the wake from the effect of ligaments.

intrinsic fracture toughness of the grain boundary. As it extends, the microstructure is sampled and the crack passes by a grain leaving it intact but spanning the crack. As the crack advances further, the unbroken grain constrains the opening of the crack, and tensile stresses are built up in the grain. When the stress exceeds the fracture stress of the bridging grain, it breaks (in a transgranular manner) and the crack can jump forward until it encounters another bridging grain. This model has recently been formulated in a quantitative manner by Cook er al. [39]. It has also been used by Cook to experimentally determine, for the first time, both the fracture toughness of grain boundaries (as a function of calcium segregation) and the strength of the grain bridges in a series of calcium-doped polycrystalline aluminas [34]. Significantly, the fracture strength he calculated for the bridges corresponds to that of single crystal alumina (sapphire). VII.3.2 Fiber-reinforced ceramics. The original emphasis of much of the development of composite materials and their analysis was to produce stronger and stiffer materials by the incorporation of highmodulus and/or high-strength fibers. In this respect the emphasis was in the reinforcement obtainable by the use of fibers. The work gave rise to a body of constitutive equations, the simplest being the socalled Rule of Mixtures, describing the modulus and strength of a composite in terms of the physical properties of the individual components. (see for instance Refs [130] and [ 1311). In the late sixties, attention shifted to what toughening might be expected by the incorporation of a fibrous phase, particularly in an othewise brittle matrix such as a ceramic. Little progress was made in developing as comprehensive a picture of toughness as had been made for strength. Applied fracture mechanics was not as advanced as today, it was not known how to measure the toughness of these materials, and the mechanisms of imparting toughness were not well established. Also, in fiber composites the toughness is not an average, macroscopic property of a material, as is elastic modulus or thermal conductivity, and as a result, bounds of self-consistent methods that had been applied so successfully to such properties could not be used. In the 1980s there has been a resurgence of interest in such fiber-reinforced ceramics, in large part as a result of the demonstration by Prewo et al. [7, 132-1341 that fiber-reinforced ceramics have potential for high-temperature applications. With the possible exception of carbon-carbon composites, these materials have not yet reached commerical application. However, the heightened interest has spurred an exploration of the mechanics underlying fiber toughening of brittle materials. The fracture of uniaxial fiber composites is surprisingly complex, because a variety of behaviors can occur depending on the relative values of the strength and fracture toughness of the matrix, the fibers and the interface between them. In addition, the volume

1152

D. R. CLARKE and K. T. FABER

/
EPOXY MATRIX COMPOSITE

LAS-II

MATRIX (CERAMED)

COMPOSITE 0 0 0.2

I
0.4

I 0.6 STRAIN, (%)

1 06

I 1 .o

1.2

Fig. 37. Comparison of the tensile stress-strain curves for two O-SC fiber-reinforced matrices. The glass-ceramic matrix material suffers multiple matrix cracking, with an associated break in the curve, whereas the epoxy matrix which has a higher elongation to failure does not and the stress-strain curve is unbroken. (Courtesy of K. M. Prewo.)

fraction of fibers, their size and spacing all enter into

the problem. One possible failure sequence associated with the extension of a crack in the matrix is as follows: as the crack front advances the load borne locally by the matrix is transferred to the fibers which are acting to restrain further opening of the crack. As the crack passes individual fibers, the interface between the fiber and matrix begins to fracture (debond) and propagate up the interface. The debond crack propagates in a largely Mode II manner, although, depending on the Poissons ratios of the fiber and matrix, there will also be a Mode I component. The debonded lengths of the fibers are now free to slide in the matrix (against coulombic friction), so the further the fibers are behind the crack tip the greater is the bridging force. At some distance behind the crack tip, the stress in the fibers either reaches the fracture strength of the fiber, in which case it will break, or for discontinuous fiber-toughened ceramics, reaches a sufficiently high stress to overcome the frictional stresses all along the fiber and the fiber pulls out. An added complication is that the fibers generally exhibit a statistical variation in strength (this is certainly true of fibers [135]). So, they may break anywhere along their length and not necessarily between the crack faces as has hitherto been assumed (largely for ease of analysis). In general, ceramic composites are made out of combinations of materials for which the failure strain of the fibers is greater than that of the matrix. Thus, under tensile loading in the direction of the fibers, the matrix can exhibit extensive cracking perpendicular to the fibers. The matrix cracking stress may well be substantially greater than the fracture stress of the

unreinforced ceramic. Furthermore, if the fibers remain intact, the composite can continue to sustain additional load up to the fracture stress of the fibers. This behavior is illustrated in Fig. 37 with the stress-strain curve of a Sic fiber-reinforced lithium aluminosilicate glass-ceramic composite. (The second curve is for the same fibers but embedded in an epoxy resin matrix. Since the strain to failure of the epoxy matrix is greater than that of the fibers, matrix cracking does not occur and no break appears on the stress-strain curve.) The slope of the initial portion of the curve is closely approximated by the rule of mixtures based on the fiber and matrix material moduli. When matrix cracking occurs, the matrix becomes crossed by many, more or less equally spaced, cracks traversing the cross-section of the material.7 In these materials, because failure does not result from the extension of a single crack, conventional fracture mechanics cannot be used to establish a failure criterion. The onset of matrix cracking is of great practical importance since it marks the load above which the matrix can no longer protect the fibers from environmental attack (oxidation and corrosion). It is hence often considered as a design limit. Because of its importance, and because it marks the stress at which fatigue damage initiates, the phenomenon of multiple matrix cracking has attracted considerable attention. The first analysis was that of Aveston et al. [ 1371,who used an energy argument to determine the condition for multiple cracking to occur. Recently, Marshall et al. [138] have treated a more general steady-state cracking model, one limit of which reduces to the Aveston et al. result. Marshall et ds treatment examines the effect of fibers on (a) the closure forces operative on the crack tip to provide notch insensitivity and (b) conditions under which the matrix demonstrates multiple cracking. It is based on a crack bridging argument in which debonded fibers restrain the crack opening. Behind the crack tip. the crack surface is subject to a net pressure from the applied loading, Q,, and the resultant pressure from intact fibers (displaced relative to the matrix). The resultant stress intensity is of the form of eqn (38). The closure pressure can be related to the crack opening, u, by the following: P = G@E/(l +

rl)lRl

(41)

t The spacing between the multiple matrix cracks has attracted some attention. By drawing an analogy to the one-dimensional parking problem in order to solve the spacing, Kimber and Keer [ 1361argue that the average crack spacing should be given by 1.337x, where x is the transfer length for transferring the load to the matrix.

where t is the inter-facial friction stress, q = fE//(l -f)E,,, and R is the fiber radius. In the stress intensity approach of Marshall et al., the criterion for matrix cracking is assumed to be when the applied stress is equal to the stress for steady-state crack growth, and that the stress intensity for the composite, KL, scales as
K:= KTEJE,,,. (42)

Solving for the crack opening

displacement

as a

Fracture of ceramics and glasses

1153

Fig. 38. Sequence of optical micrographs showing matrix crack on tensile surface of four-point flexure

beam during an unload/reload cycle. (AHC) decreasing load; (C)-(E) increasing load. The crack is not visible in (C) and the crack opening is larger in (B) than in (D) despite the load being the same. (Courtesy
of D. B. Marshall.)

function of position from the crack tip results, for steady-state conditions, in crO = V[(l - v2)KfzE/f2(1 -f)(l +

q)2/REml, (43)

where V is a dimensionless constant. An important finding of the Marshall analysis, and one shared by the Aveston-Coope-Kelly analysis and more recent one of Budiansky et al. [ 1391, is that when the matrix crack is bridged by unbroken fibers, the applied stress necessary to extend the crack is no longer dependent on the crack size. This is in sharp contrast to the behavior of monolithic ceramics, where the Griffith relation holds and the strength falls with crack size. The consequence of this finding is that the stress for matrix cracking is not expected to be adversely affected by the presence of large processing flaws; an important result since the processing of ceramic composites is more difficult than that of other ceramics and the formation of large flaws proves particularly difficult to prevent. Equation (43) provides microstructural guidelines

for the design of ceramic composites with enhanced resistance to matrix cracking. Ideally stiff fibers of small diameter should be used in a tough, yet compliant matrix. Though eqn (43) suggests a high interfacial friction stress is attractive for a large matrix cracking stress, high values of T enhance the tensile stress in the fibers and fiber failure may occur coincident with matrix cracking. Optimization of the interfacial frictional stress is then required to prevent a change in failure mode. Unfortunately, the interfacial friction stress is the least well understood and most difficult to measure of all the parameters in eqn (43). One particularly simple and direct way of measuring T is by pushing on the end of a fiber embedded in a matrix with a hardness indentor [140]. It has also been measured from the hysteresis in matrix crack opening displacement when loaded and unloaded at small tensile loads. Such a hysteresis is evidence of the existence of sliding friction, and is illustrated in Fig. 38 taken from such a measurement of a Sic fiber-reinforced glass ceramic.

1154

D. R. CLARKEand K. T. FABER

notch (SEN) fracture toughness tests on a model system consisting of annealed nickel wires aligned in an epoxy matrix and compared their results to a simple crack bridging theory. Their work and the accompanying photomicrographs provide a striking illustration of the role of crack bridging and interface debonding in fracture (Fig. 39). They first studied the pull-out of individual wires embedded in the matrix so as to establish the interface debonding strength and the frictional stress, and then measured the crack resistance as a function of crack length and number of wires spanning the crack faces. From this they were able to evaluate the crack bridging force. They found that the debonding stress for the interface was just greater than the yield stress of the nickel wires, and remained constant as the debond propagated up the fiber interface. This constancy meant that the bridging force applied by the fibers to the matrix crack was independent of the distance back from the crack tip, enabling the authors use the simplification of a Dugdale zone in calculating the effect that the fibers had on the fracture toughness. (This corresponds to the value of Q~(u) in eqn (38) being a constant independent of the variable x so that the toughness is proportional to the square root of the bridging length.) VII.4 Toughening at high temperatures Of the toughening mechanisms discussed above, crack bridging seems to be the most promising for high-temperature use. Transformation toughening is ineffective at temperatures above the monoclinictetragonal phase transformation since there is no chemical driving force for the transformation to occur. For pure zirconia this places an upper bound of somewhere above about 800C depending on the particle size and stabilizing solute content. Higher temperatures might be achievable by using hafnia instead of zirconia, since it has a significantly higher transformation temperature to tetragonal (1620C) or with some, as yet undiscovered phase transformation having similar attributes. However, it is likely that the upper use temperature would not be limited by the transformation temperature, but more seriously by the temperature at which the constraint of the surrounding material relaxes away by viscous flow. Likewise, microcracking toughening would be similarly restricted. The conclusion that crack bridging is the most likely approach to increasing the fracture toughness of ceramics at high temperatures has led to the present intense interest in fabricating ceramics by incorporating fibers, such as the Nicalon Sic-based fibers, and with whiskers of Sic, S&N,,, and A&O, into the starting powders and subsequently sintering or hot-pressing. One material, sintered silicon nitride with ceria and alumina, is of particular note since exaggerated grain growth of the silicon nitride is encouraged during processing so the fibers are grown in situ.

Fig. 39. The progress of a crack in a single edge notch beam specimen of epoxy resin reinforced with nickel wires (3.2u/o) viewed through crossed polarizers. The crack runs horizontally and the wires are vertical. The point on a wire where

the photoelastic fringes are most closely spaced marks the position of the debonding front. (Courtesy of G. W. Groves.)

In the last 2 or 3 yr, development of the mechanics of fracture has out-distanced experiment, so many critical tests of the theories remain to be performed. For instance, although an essential feature of all fiber models is that the fibers exert a restraining or closure force on the crack walls, no direct or in situ measurements of the stresses in the fibers as a function of their position back from the crack tip have been reported in all-brittle composites. One experimental verification of the contribution to fracture resistance from the fiber crack bridging force is available from the work of Bowling and Groves [141]. They carried out both double cantilever beam (DCB) and single egde

Fracture of ceramics and glasses Two additional mechanisms appear to give rise to some increase in toughening, albeit over a narrow range of temperature [142]. One is the restraint of a propagating crack at high temperatures by the viscous extension of glassy ligaments behind the crack tip, as seen in Fig. 20. The second contribution is the constraint of a cavitating region around the propagating crack which is formally similar to stress-induced microcracking. VII.5

1155

12. Hutchinson J., (this volume). 13. Griffith A. A.. Phil. Trans. R. Sot. A. 221. 163 (1920). 14. Engel U. and Hubner H., J. Mater. &i. 1s 2od3
(1978). 15. Irwin G. R., Handbuch der Phvsik. Vol. VI (Edited bv

Summary

remarks

In closing, it should be emphasized that whilst distinct toughening mechanisms have been discussed in the preceding sections, it is possible for the observed toughness to be the result of several mechanisms occurring concurrently. Likewise, it is possible to envisage designing a microstructure that can exhibit two or more toughening mechanisms simultaneously, or in which different mechanisms operate in different temperature regimes. Some progress in developing such materials is presently under way. Some of the mechanisms discussed above when combined are expected to lead to increases in fracture resistance that are not simply additive, but rather are multiplicative. For instance, combining crack bridging with transformation toughening would lead to a form of symbiotic toughening.

_ Flugge S.) Springer-Verlag. Berlin (1958). . 16. Rice J. R.. J. Mech. Phvs. Solids 26. 61 11978). 17. Maugis D:, J. Mater. Sii. 20, 3041 (1985). 18. Chudnovsky A. and Moet A., J. Mater. Sci. 20, 630 (1985). 19. Johnson K. L., Kendall K. and Roberts A. D., Proc. R. Sot. Land. A, 324, 301 (1971). 20. Tabor D. J., Colloid Interface. Sci. 58, 2 (1977). 21. Derjaguin B. V., Muller V.M. and Topordv Yu. P.,
J. Colloid Interface Sci. 53, 314 (1975).

22. Derjaguin B. V:, Muller V. M. and Toporov Y. P., J.


Colloid Interface Sci. 73, 292 (1980). 23. Greenwood J. A. and Johnson K. L., Phil. Mag. 43, 697 (1981). 24. Lawn B. R., Evans A. G. and Marshall D. B. J. Am.

ceram. Sot. 63, 574 (1980). 25 Sih G. C., Handbook of Stress Intensity Factors. Lehigh University Press, Bethlehem (1974). 26. Marshall D. B., Lawn B. R. and dhanukul P., J.
Mater Sci. 14. 2225 (1979).

27. Cook R. F., j. Mater. Res 1, 852 (1986). 28. Thomson R. M. (this volume). 29. Clarke D. R., Lawn B. R. and Roach D. H., Fract.
Mech. Ceram. 8, 341 (1986). Clarke D. R., J. Am. ceram. Sot. 70, 15 (1987). Thomson R. M., J. Mater. Sci. 15, 1014 (1980). de Gennes P. G., Rev. Mod. Phys. 57, 827 (1985). Cook R. F., Lawn B. R. and Fairbanks C. J., J. Am. ceram Sot. 68, 604 (1985). 34. Cook R. F., Acta Metall. (to be published). 35. Swain M. V., J. Mater. Sci. Lerr. 5, 1313 (1986).

30. 31. 32. 33.

Acknowledgemenfs-It is a pleasure to acknowledge the discussions we have had over the years with many of the authors cited in this work. We are also grateful to the following for generously supplying the micrographs and figures used to illustrate this review: R. F. Cook, A. G. Evans, G. W, Groves, A. H. Heuer, B. R. Lawn, D. B. Marshall, K. M. Prewo, R. Steinbrech and S. M. Wiederhorn. Funding for KTF has been provided by the National Science Foundation (DMR-8351476) in conjunction with Champion Spark Plug, IBM Corporation, Eastman Kodak, Owens Corning Fiberglas and 3M Co.

36. Kendall K., Alford N. McN., Tan S. R. and Birchall J. D.. J. Mater. Res. 1. 120 (1986). 37. Cook R. F. and Clarke D. R., A& Metall. (in press). 38. Cook R. F., Ph.D. thesis, University New South Wales (1985). 39. Cook R. F., Fairbanks C. J., Lawn B. R. and Mai Y-W., J. Am. Ceram. Sot., (in press). 40. Mai Y-W. and Lawn B. R., A. Rev. Mater. Sci. 16,
415 (1986).

41. 42. 43. 44.

Willis J. R., J. Mech. Phys. Solids 15, 151 (1967). Rice J. R., J. appl. Mech. 35, 379 (1968). Wolf D., Acta Metall. 32, 245 (1984). Kelly A., Tyson W. and Cottrell A. H., Phil. Mag. 15,
567 (1967).

45. Rice J. R. and Thomson R. M., Phil. Mag. 29, 73 REFERENCES 1. Heuer A. H. and Hobbs L. W. (Editors), Science and Technology of Zirconia. Advances in Ceramics, Vol. 3. American Ceramic Society, Columbus, OH (1981). 2. Riley F. L. Progress in Nitrogen Ceramics. Martinus Nijhoff, Boston(1983). 3. McGeehin P. and Hooper A., J. Mater. Sci. 12, 1 (1977). 4. Clarke D. R., A. Rev. Mater. Sci. 13, 191 (1983). 5. Schwartz B., J. Phys. Chem. Solids., 45, 1051 (1984). 6. McMillan P. W. ~Glass-Ceramics. Academic Press, New York (1979). 7. Prewo K. M. and Brennan J. J., J. Mater, Sci. 15,463 (1980). 8. Bednorz J. G. and Mueller K. A., Z. Phys. BCondensed Matter 64, 189 (1986). 9. Lawn B. R., Fract. Mech. Ceram. 5, 1 (1983). 10. Blau P. J. and Lawn B. R. Microindentation Techniques in Materials Science and Engineering. ASTM Publication 889, Philadelphia (1985). 11. Weibull W., J. appl. Mech. 18, 293 (1951).
(1974).

46. Elliott H. Proc. phys. Sot. Land. 59, 208 (1947). 47. Barenblatt G. I. Ado. appl. Mech. 7, 55 (1962). 48. Bilby B. and Eshelbv J. D.. Fracfure (Edited bv H. Liebowitz), Vol. 1; p. 99. Academic Press, New York (1968). 49. Chan D. Y. C., Hughes B. D. and White L. R.. J. Colloid Interface SC; 115 (l), 240 (1987). 50. Gehlen P. C. and Kanninen M. F.. Inelastic Behavior of Solidr (Edited by Kanninen, Adler, Rosenfield and Jaffee), p. 587. McGraw-Hill, New York (1970). 51. Gehlen P. C., Hahn G. T. and Kanninen M. F.,
Scripta Metall 6, 1087 (1972). 52. Hockey B. J. and Lawn B. R., J. Mafer. Sci. 10, 1275

(1975). 53. Hockey B. J., Fract. Mech. Ceram. 6, 637 (1983). 54. Ohr M., Mater. Sci. Enana 72. 1 (1985). 55. Sinclair J. E. and Lawn B. R., iroc. R. ioc. Land. A,
329, 83 (1972). 56. Sinclair J. E., Phil. Mag. 31, 647 (1975). 57. Haneman D., Roots W. D. and Grant J. T. P., J. appl. Phys. 38, 2203 (1968).

1156

D. R. CLARKE and K. T. FABER 97. Faber K. T. and Evans A, G., J. Am. ceram. Sot. 66, C94 (1983). 98. Gisi P. and Faber K. T., (to be published). 99. Eshelby J. D., Proc. R. Sot. Lond. A, 241,376 (1957). 100. McMeeking R. M. and Evans A. G., J. Am. ceram.
Sot. 65, 242 (1982).

58. Horn R. G., Israelachvili J. N. and Pribac F., J. Colloid Interface Sci. (in press). 59. Israelachvili J. N. and Tabor D., Proc. R. Sot. Lund. A, 331, 19 (1972). 60. Israelachvili J. N. and Adams G. E. J. them. Sot.
Faraday Trans. 174, 975 (1978).

61. Mould R. E., J. Am. ceram. Sot. 43, 160 (1960). 62. Lawn B. R., Jakus K. and Gonzalez A. C.. J. Am.
ceram. Sot. 68, 25 (1985).

63. Kramer, E. J., Developments in Polymer Fracture (Edited by Andrews E. H.), p. 55. Applied Science Publishers, London (1979). 64 Wiederhom S. M.. J. Am. ceram. Sot. 50. 407 (1967). 65. Wiederhom S. h., Int. J. Fract. Mech. i, Ii? (1968). 66. Hillig W. B. and Charles R. J., High Strength Materials (Edited by Zackay V. F.), p. 682. John Wiley, New York (1964). 67. Fuller E. R. and Thomson R. M., Fract. Mech.
Ceram. 4, 507 (1978).

101. Clarke D. R., J. Am. ceram. Sot. 67, Cl5 (1984). 102. Buecker H. F., 2. Angew. Math. Mech. 50,529 (1970). 103. Marshall D. B., Drory M. and Evans A. G., Fract. Mech. Ceram. 6, 289 il983). 104. Budianskv B.. Hutchinson J. W. and Lambronoulos _ J. C., Inr: J. $olids Strucr. 19, 337 (1983). 105. Garvie R. C., Hannink R. J. H. and Pascoe R. T.,
Nature, (1986). Lond. 258, 703 (1975).

106. Evans A. G. and Cannon R. M., Acta Metall. 34,761 107. Ruhle M., Strecker A., Waidelich D. and Kraus B.,
Ado. Ceram. ceram. Soc.C, 12, 256 (1984). 72 (1981).

108. Kosmac T.. Wagner R. and Claussen N., J. Am. 109. Garvie R. C.. Hannink R. H. J. and Swain M. V.. J. Mater. Sci. htt. 1, 437 (1982). 110. Clarke D. R. and Adar F., J. Am. ceram. Sot. 65, 284
(1982).

68. Brown S. D., Fract. Mech. Ceram. 4, 597 (1978). 69. Michalske T. A. and Freiman S. W., J. Am. ceram.
Sot. 66, 284 (1983).

70. Michalske T. A. and Bunker B. C., J. appl. Phys. 56,


2686 (1984).

71. Michalske T. A., Bunker B. C. and Freiman S., J. Am. ceram. Sot. 69, 721 (1986). 72. Stevens R. N. and Dutton R., Mater. Sci. Engng 8,
220 (1971).

73. Cao H., Dalgleish B. J. and Evans A. G., J. Am. ceram. Sot. 70, 257 (1987). 74. Clarke D. R., J. Mater. Sci. 20, 1321 (1985). 75. Dalgleish B. J., Slamovich E. B. and Evans A. G., J.
Am. ceram. Sot. 68, 575 (1985).

Lange F. F., J. Mater. Sci. 17, 247 (1982). Becher P. F., Acta Metall. 34, 1885 (1986). Swain M. V., Acra Metall. 33, 2083 (1985). Swain M. V. and Rose L. R. F., J. Am. ceram. Sot. 69, 511 (1986). 115. Becher P. F., J. Am. ceram. Sot. 66, 485 (1983). 116. Hasselman D. P. H. and Haller R. A. (Editors), Thermal Stresses in Severe Environments. Plenum, New York (1980). 117. Davidge R. W. and Green T. J., J. Mater. Sci. 3, 629
(1968).

111. 112. 113. 114.

76. Dalgleish B. J., Johnson S. M. and Evans A. G., J.


Am. ceram. Sot. 67, 741 (1984).

77. Johnson S. M., Dalgleish B. J. and Evans A. G., J.


Am. ceram. Sot. 67, 759 (1984). 78. Jakus K., Wiederhom S. M. and Hockey B. J., J. Am. ceram. Sot. 69, 725 (1986).

118. Clarke D. R., Acta Metall. 28, 913 (1980). 119. Swain M. V., J. Mater. Sci. 16, 151 (1981). 120. Ruhle M., Claussen N. and Heuer A. H., J. Am.
ceram. Sot. 69, 195 (1986).

121. Faber K. T. and Evans A. G., J. Am. ceram. Sot. 67,


255 (1984).

79. Dalgleish B. J. and Evans A. G., J. Am. ceram. Sot. 68, 44 (1985). 80. Lange F. F., Davis B. I. and Clarke D. R., J. Mater. Sci. 15, 601 (1980). 81. Clarke D. R., J. Phvs. C4, 51 (1985). 82. Monkman F. C. and-Grant.N. J.: Pro;. Am. Sot. Test
Mater. 56, 593 11956). 83. Argon A. US.and Salama M. M., Phil. Mag. 36, 1217 (1977). 84. Thouless M. D. and Evans A. G., Scripta Metall. 18,

122. Clarke D. R. and Green D. J., Advances In Materials Characterization (Edited by Rossington D.). Plenum, New York (1983). 123. Faber K. T., Iwagoshi T. and Ghosh A., J. Am. ceram. Sot. (In Press). 124. Morelli T. A. and Takemori M. T., J. Mater. Sci. 18, 1836 (1983). 125. Evans A. G. and McMeeking R. M., Acra Metall. 34,
2435 (1986).

1175 (1984). 85. Berg C. A., A.S.M.E. 2, 885 (1962). 86. Evans A. G. and Rana A., Acta Metall.
(1980).

28,

129

87. Evans A. G., Rice J. R. and Hirth J. P., J. Am. ceram. Sot. 63, 368 (1980). 88. Drescher A. and de Josselin de Jong G., J. Mech. Phys. Soli& u), 337 (1972). 89. Chuang T-J. and Rice J. R., Acta Metall. 21, 1625
(1973).

126. Rose L. R. F. Int. J. Fract. 18, 135 (1982). 127. Sneddon I. N. and Lowengrub M., Crack Problems in the Classical Theory of Elasticity. John Wiley, New York (1969). 128. Steinbrech R. and Knehans R., J. Mater. Sci. L&l. 1,
327 (1982).

129. Steinbrech R., Knehans R. and Schaarwachter W., J.


Mater. Sci. 18, 265 (1983).

90. Chaung T.-J., Kagawa K. I., Rice J. R. and Sills L. B., Acta Metall. 27, 265 (1979). 91. Marion J. E., Evans A. G., Drory M. D. and Clarke D. R., Acta Metall. 31, 1445 (1983). 92. Fields R. J. and Ashby M. F., Phil. Mag. 33, 33
(1976).

93. Wiederhom S. M. and Chuck L., (to be published). 94. Hsueh C. H. and Evans A. G., Acta Metall. 29, 1907
(1981).

95. Faber K. T. and Evans A. G., Acra Metall. 31, 565


(1983).

96. Faber K. T. and Evans A. G., Acta Metall. 31, 577


(1983).

130. Kelly A., Strong Solids. Oxford University Press, Oxford (1966). 131. Hashin Z., J. appl. Mech. SO, 481 (1983). 132. Prewo K. M., J. Mater. Sci. 17, 3549 (1982). 133. Brennan J. J. and Prewo K. M., J. Mater. Sci. 17, 2371 (1983). 134. Prewo K. M.. J. Mater. Sci. 21, 3590 (1986). 135. R. D. Maurer, Applied Physics Letters 27, 220 (1975). 136. Kimber A. C. aid Keer J. C., J. Mater. Sci. I&t. 1, 321 (1982). 137. Aveston J., Cooper G. A. and Kelly A., Properties of Fiber Composites p. 15. IPC Science and Technology Press (1971). 138. Marshall D. B., Cox B. N. and Evans A. G., Acta
Metall. 33, 2013 (1985).

Fracture of ceramics and glasses 139. Budiansky B., Hutchinson J. W. and Evans A. G., .I. Mech. Phys. Solids 34 (2), 167 (1986). 140. Marshall D. B., .I. Am. cerum. Sot. 67, 259 (1984).

1157

141. Bowling J. and Groves G. W., J. Muter. Sci. 14, 443 (1979). 142. Clarke D. R. and Wiederhorn S. M., (to be published).

Вам также может понравиться