4.1. Brief Introduction on Physically Based Failure Theories
Faced with numerous efforts in this area, several researchers, including Li et al. [
24] and Christensen [
25,
26], have taken a firm stance, asserting that failure theories lacking a strong foundation in physics and mathematics can never truly be considered robust. While this opinion may be somewhat radical, particularly from the standpoint of engineering applications, it underscores the critical role of physical soundness in the pursuit of a reliable failure description. In this context, several failure theories (often referred to as physically based) have outperformed other competitors and have gained recommendations from researchers and the design community. Notable among them are contributions by Hashin [
15,
16], Puck [
18], Pinho [
19,
20], and various extensions [
27,
28,
29].
In principle, the construction of a failure theory with physical considerations entails the dissection of different failure modes and the development of failure criteria based on the analysis of underlying mechanisms. The distinction between longitudinal and transverse failure in unidirectional FRPs has become a consensus, which can be traced back to Hashin, who laid the groundwork for failure criteria based on failure mechanisms. In 1973 [
15], he instructively categorized unidirectional FRP failure into fiber-dominated and matrix-dominated, each of which was further subdivided into tensile and compressive modes in 1980 [
16].
Traditionally, matrix failure has been formulated as a polynomial in the stress components (usually excluding
). Many of these criteria are transplanted from the yield criteria of von Mises or Hill, which were originally developed for ductile materials. However, the matrix fails primarily in a plane parallel to the fiber direction and behaves in a brittle manner. A more convincing criterion for matrix failure can be developed by introducing the so-called ’action plane’, i.e., a plane in which fracture is likely to take place. Although the concept was initially mentioned in Hashin’s 1980 article, it was Puck [
18] who pioneered putting it into practice and vividly described this matrix failure phenomenon as “inter-fiber fracture (IFF)”.
Rooted in Mohr and Coulomb’s hypothesis, which was later adapted by Paul for transversely isotropic materials, the inter-fiber fracture condition is exclusively formulated with one normal and two shear traction components exerted on the action plane:
where
,
, and
can be easily calculated using the transformation rules for a second-order tensor, as depicted schematically in
Figure 11. For a generic stress state, the fracture plane angle
is unknown in advance, which calls for a search for the maximum value of
. Some efficient algorithms for carrying this out have been suggested [
30,
31].
Fiber kinking is a prevalent phenomenon in unidirectional fiber-reinforced composites, particularly those with a high fiber volume fraction, when exposed to intense longitudinal compression. There has long been an ongoing debate concerning the motivation behind the formation of kink bands. Microscopically, fibers within kink bands undergo a sudden change in direction, accompanied by localized deformation and damage. This abrupt loss of microstructural stability naturally evokes associations with fiber buckling under compression. However, mounting evidence [
32,
33] increasingly supports the contention that kink bands arise from a separate mechanism, distinct from the global micro-buckling mode proposed by Rosen [
34].
Argon [
35] was the first to establish the correlation between the longitudinal compressive strength
and the in-plane shear strength
, by assuming that fiber kinking arises from local microstructural imperfections when matrix cracking takes place in the vicinity of the misaligned fibers. Afterwards, both analytical and numerical models based on the fiber-kinking theory (FKT) were developed to gain a deeper understanding of fiber compressive failure. Davila et al. [
36] developed an FKT-based criterion in terms of homogenized in-plane stress components, which was subsequently extended to three-dimensional stress states by Pinho et al. [
19].
Figure 12 illustrates the transformation procedure and relevant coordinate systems in the three-dimensional kinking model. To establish a fiber compression criterion in accordance with the FKT, the stress in a misaligned frame needs to be given by two successive transformations of coordinate systems, i.e.,
where
and
are, respectively, the rotation tensor from 1-2-3 to
and from
to
. Then, the onset of fiber kinking can be identified using a matrix failure criterion.
For a plane stress state under off-axis compression, the angle of the kinking plane,
, is assumed to be zero. The misalignment,
, is the sum of an initial misalignment angle, assumed to be a material constant, and a rotation,
, induced by the applied shear stress in terms of the shear constitutive law expressed as
. So,
can be determined by solving the following equation:
and the angle
becomes:
To solve Equation (
9) exactly, numerical methods such as the Newton–Raphson algorithm can be used (for details, see [
20]). Assuming small angle approximations and a linear material response, the previous equation results in:
which can be solved in close form as:
Four representative failure theories are briefly described here: Hashin [
16], Camanho [
17], Puck [
18], and Pinho [
20]. The last one is more frequently referred to as the LaRC05 criteria, which amalgamate the wisdom of several predecessors, encompassing the concept of the action plane and fiber-kinking theory, as well as the in situ effect of strength. Distinct formulas for matrix failure and fiber failure are presented in
Table 3 and
Table 4, respectively. While some of them differ between tension and compression, the current discourse specifically centers on the compression scenario. To keep consistency, parameter symbols with analogous meanings across different theories have been unified.
The 1980 version of Hashin’s matrix criterion is adopted, which includes a linear term that enhances its flexibility in characterizing the failure envelope. Nevertheless, to circumvent the need for additional experimental measurements during parameter determination, it still incorporates certain simplifications, notably the controversial assumption of infinite equi-biaxial transverse compressive strength. Camanho provided an elegant and concise formulation for matrix failure by taking advantage of the material’s symmetry. In this formulation, , , and represent a set of irreducible stress invariants, while , , , and are coefficients awaiting calibration through experimental results. Hashin’s criterion also falls under the category of invariant-based approaches but, owing to its widespread familiarity, it is usually expressed in terms of stress components and basic strength properties.
Both the Puck and LaRC05 theories employ the action-plane-related approach, defining matrix failure based on the IFF condition. In this context, traditional basic strengths are no longer applicable because they are not necessarily relevant to the fracture plane. Instead, fracture resistances, which represent the maximum sustainable normal or shear traction on the action plane when applied alone, should be incorporated into the fracture conditions. When compression is applied on a potential fracture plane, two resistances,
and
, play a role against fractures caused by
and
, respectively. This is because a negative
only impedes fractures by increasing shear fracture resistances (or, equivalently, by reducing the effective shear traction [
20]). To account for the increase in shear fracture resistances, two additional parameters
and
are involved, which can be physically interpreted as the friction coefficients of the fracture plane in the sense of the Mohr–Coulomb law, when a linear effect of
is assumed.
For better consistency with experimental data, Puck and his colleagues introduced several refinements to the IFF condition. The version utilized in this section is the most widely accepted, distinguished by its notable feature that the fracture body displays analogous parabolic contours on any longitudinal sections rotated about the -axis. The matrix-dominated criterion in LaRC05 draws inspiration from Puck’s theory while broadening its capabilities to include the influence of adjacent plies or external boundaries. However, for sufficiently thick unidirectional laminates, the so-called in situ effect is absent, indicating that the matrix compression criterion in LaRC05 essentially mirrors the original Puck criterion.
With regard to fiber compression failure, the maximum stress criterion continues to be extensively employed, even though Hashin and Puck have independently elucidated the underlying mechanisms in their theories. In contrast, both Camanho and Pinho embrace the FKT-based criterion, with the only distinction being their individual methodologies for detecting matrix failure in the misaligned coordinate system.
According to Pinho’s explanation [
37], localized damage to the supporting matrix only develops into a kink band when the compression in the longitudinal direction is extremely high; otherwise, it leads to splitting, which is essentially a matrix-dominated failure (or inter-fiber fracture). The transition between kinking and splitting is hard to identify, as they share the same expression and there is no specific change in the trend of the failure envelope. Experimental research by Jelf and Fleck [
6] found that specimens only fail in the form of splitting when the longitudinal compression is low and in-plane shear is high. Accordingly, the transition between kinking and splitting can be roughly predicted based on the compressive stress in the fiber direction at the point of failure. In detail, fiber kinking occurs when
, while fiber–matrix splitting occurs when
. This issue was not explicitly mentioned by Camanho, but it can be guessed that the same treatment was adopted.
4.2. Input Parameters Related to Failure Criteria
During the application and evaluation of different failure theories, inaccurately determined or misinterpreted parameters not only introduce errors but, more critically, lead to illogical conclusions. The conventional methods for determining strength parameters in the four failure theories are described below.
The parameters required in Hashin’s matrix failure criterion have been measured and listed in
Table 1, with the only exception being for the transverse shear strength
. The determination of
poses some difficulties but, fortunately, there seems to be an inherent relationship between
and
. According to [
27], as long as
, it is reasonable to approximate that:
For an invariant-based formulation of matrix failure in Camanho theory, it is easy to see that:
However, relying on uniaxial compression results alone is insufficient for determining the remaining two unknown parameters. It is recommended to introduce hydrostatic or equi-biaxial transverse compressive strength, denoted as
, so that:
A total of four parameters are required for IFF conditions in either Puck or LaRC05 (an additional three parameters when the tensile mode is considered). It is acknowledged that a pure longitudinal shear loading (e.g., or ) always causes fracture in its action plane; thus, the corresponding fracture resistance is equal to . The situation is quite different for , as pure transverse shear stressing (e.g., ) typically leads to tensile fracture in the plane of maximum principal stress. In view of this, is measured indirectly through a transverse compression test. The critical condition is reached when the Mohr circle corresponding to transverse compressive loading touches the fracture curve. can be determined simultaneously since two curves are tangent at this point. Therefore, the values of and can be calculated from the compressive strength and the measured fracture angle .
In addition, a simplification of the coupling between the inclination parameters
and
is introduced as
With regard to fiber compression failure, Hashin and Puck utilize the maximum stress criterion, relying directly on the uniaxial strength property
. In the FKT-based criterion embraced by Camanho and Pinho, the initial fiber misalignment angle
plays a pivotal role and is usually calculated by considering the critical condition that the material fails under pure longitudinal compression, i.e.,
. Assuming that
is the total misalignment angle at the moment of failure, the stresses in the misalignment coordinate system are:
By substituting them into the matrix failure criteria of Camanho and LaRC05, respectively, the values of
can be determined. The initial misalignment angle is then calculated by reversing the deformed rotation. For a material that exhibits linear response in shear,
can be calculated as follows:
Curiously, there is a scarcity of experimental studies concerning the validation and measurement of such fiber imperfections.
All the parameters in these failure criteria are determined a priori, summarized in
Table 5.
4.3. Comparison with Experimental Results
The predicted ultimate strength as a function of the off-axis angle according to various failure theories is displayed in
Figure 13, alongside the experimental results. All predicted curves show a similar trend to the experimental results, but there are some differences in the range of small off-axis angles. The Hashin and Puck theories overestimate the compressive strength in the range of 0–30
, while the LaRC05 theory shows an underestimation in the same range. When the loading direction starts to deviate from 0
, the predicted curve of the Camanho theory initially coincides with that of the LaRC05 theory, but it agrees slightly better with the experimental results at 15
and 20
.
Figure 13 also presents the variation in fracture angles predicted by two theories, Puck and LaRC05, respectively. For small off-axis angles, the inclination of the fracture plane remains unnoticeable. However, as the off-axis angle increases, the transverse compression begins to dominate, resulting in a rapid increase in the fracture angle, eventually reaching approximately 51
degrees under pure transverse compression. It can be concluded that the action-plane-related matrix criterion can accurately identify the transition between the two matrix failure modes of transverse compression and in-plane shear.
The predictive capabilities of different theories can be compared more intuitively in two-dimensional diagrams depicting and . The predicted curves generated by each theory encompass scenarios of both off-axis and ideal biaxial compressive loading, delineated by solid and dashed lines, respectively. The latter actually represents the theoretical failure envelope in the two-dimensional stress space, and is presented for comparative analysis of the off-axis test with the biaxial test from a theoretical point of view.
Figure 14a illustrates the different predictions for the stress combination
. In the high longitudinal compression region, the predictions exhibit two distinct patterns, reflecting different considerations of fiber failure. For the Hashin and Puck theories, the longitudinal compressive strength is insensitive to shear stress, so their predicted curves take on a rectangular or nearly rectangular shape. In contrast, the Camanho and LaRC05 theories predict an approximately linear relationship with the assumption that the presence of in-plane shear stress promotes the formation of kink bands. Comparison with experimental results indicates that neither the maximum stress nor the FKT-based criteria agree well with the experimental data. Specifically, the failure loci measured by the off-axis compression tests fall between the linear and rectangular predictions.
It has been stated that the off-axis testing results depicted do not fully represent the failure envelope in this two-dimensional stress plane, as evidenced by the noticeable disparity between the predicted curves for off-axis and biaxial loading in the region of low longitudinal compression. Since all four failure theories invariably assume that matrix failure is independent of , there is a common upper limit at among the predicted envelopes, which represents an in-plane shear failure resulting from alone. When dealing with off-axis loading, the effect of the third stress component beyond this diagram is addressed through different matrix failure criteria.
A more comprehensive illustration and comparison of the predictions for matrix-dominated failure is attainable in the
diagram, as shown in
Figure 14b. Under relatively high transverse compression, the predictions of Comanho, Puck, and LaRC05 closely match the experimental results, whereas the prediction of Hashin shows a slight underestimation. In the region of high in-plane shear, while all of the predicted curves capture the shear strengthening effect to some extent, they show greater differences from each other. Both the Hashin and Puck criteria overestimate
at failure when superimposed on a relatively low compressive
, but Puck’s prediction shows a better fit to the variation trend. At first glance, the curves predicted by Comanho and LaRC05 appears to agree well with the experimental results, especially with the latter effectively reproducing the near-linear enhancement of
by compressive
. However, it should be noted that their predictions in this region are not determined by the matrix failure criterion but rather controlled by the FKT-based criterion for fiber failure, which appears to be inconsistent with the observed failure modes. According to the theoretical introduction, this problem can be eliminated by choosing a suitable
. Nevertheless, the predicted results of Camanho and LaRC05 are still questionable as they are at odds with the previous discussion of the experimental results, i.e., the failure loci projected onto this stress plane roughly coincides with the two-dimensional failure envelope. There may be some misunderstanding about the initial misalignment angle in the FKT-based criterion.
The predictions for ideal biaxial loading indirectly support this suspicion, as all four predicted envelopes are non-conservative relative to the experimental results. Graphically, these envelopes converge at the same point on the -axis that corresponds to failure under pure in-plane shear. Consequently, there is further speculation that the unsatisfactory predictions are not solely attributed to the criteria themselves; an accurate determination of the in-plane shear strength carries a parallel significance.
4.4. Recalibration with Off-Axis Test Results
In case
is not accessible, Camanho et al. [
17] alternatively suggest resorting to off-axis testing. However, they do not explicitly require whether the off-axis results should be from tensile or compressive tests, nor do they specify off-axis angles. It is only emphasized that the critical stress state must be taken from a matrix-dominated failure. Based on a combination of stress components (
,
), the parameters are determined as follows:
Predicted curves calibrated with different off-axis test data are presented in
Figure 15, together with those generated for a variety of assumed
values. It can be observed that only the test results for large off-axis angles can be relied upon. In the low compression region, the elliptical curves dictated by the formulation of the Camanho criterion do not truthfully fit the experimental failure envelope. Thus, if the curve is forced to pass through a particular failure locus within this region, the overall prediction will be severely biased. It should also be noted that calibration results derived from off-axis tests are more sensitive to experimental errors, as evidenced by the predicted envelopes in the
stress plane. Since having a closed failure envelope in this stress plane is a major advantage of the Camanho theory, it is essential to verify that the estimates of
are within acceptable limits after being calibrated by off-axis testing results.
As mentioned earlier, there is a need for an accurate method to determine the in-plane shear strength
. Tsai et al. [
13] previously proposed that the pure in-plane shear strength could be extrapolated from off-axis test results. However, at that time, their focus was solely on the shear strength values, without clarifying the exact form and meaning of the extrapolation function. In addition, it is noticed that the derivation process for the parameters involved in action-plane-related IFF conditions relies on the measurement of the fracture angle, which tends to be quite scattered. As depicted in
Figure 16,
and
are sensitive even to marginal changes in the fracture angle, and this sensitivity is also transferred to
via Equation (
18). Consequently, it is sometimes more convenient to directly select a typical value for
[
38]. It must be pointed out that the suggested range for
and the assumptions about parameter coupling are derived empirically in the absence of experimental data. Utilizing them without further validation carries a certain risk; thus, more dependable methods are required.
Following Puck’s suggestion, the parameter
can be derived from the experimentally established
envelope since the fracture plane angle is zero in this segment of the failure envelope, where
and
. Combining these pieces of information, the values of
(or
) and
in the matrix failure criteria of Puck and LaRC05 can be determined by fitting the outcomes of off-axis compression tests, as shown in
Figure 17. It can be seen that whether or not fiber rotation is taken into account has a certain but very limited impact on the parameter determination. For the sake of theoretical self-consistency, experimental results presented in co-rotational coordinates are preferred. Accordingly, fiber rotation must also be considered during failure prediction, which is roughly carried out in this paper by solving Equation (
10) with
.
In order to avoid uncertainties in the measurement of the fracture angle, the off-axis test results were also used to calibrate the predictions of the Puck and LarC05 codes in the high transverse compression range within the stress plane . Whether there exists an analytical solution of and by explicitly eliminating depends on the specific form of the IFF condition. However, in most cases, it can be solved numerically.
The validity of matrix failures predicted by FKT-based criteria has not been further explored until this point in the paper. As a result, only the matrix criteria from each failure theory are employed here to predict the failure envelope in the
stress plane. The results are shown in
Figure 18.
Because of the limitations of their mathematical forms, the matrix failure predictions by Hashin and Camanho struggle in simultaneously conforming to the failure envelope trends for both high and low transverse compression. When the in-plane shear strength is tuned down, although there is a slight improvement in the predictions near the -axis, the weakening effect of high compressive on is irrationally amplified. Conversely, the action-plane-related IFF criteria, represented by Puck and LaRC05, show superior performance. Not only do they better approximate the experimental results, but they also identify the shift in failure modes.
4.5. Further Discussion on Fiber Compressive Failure
In describing fiber failure under the combined longitudinal compression and in-plane shear, neither the maximum stress criterion nor the FKT criterion effectively takes into account the influence of shear stress. The former, as one of the most traditional and straightforward failure criteria, bluntly disregards the influence of . On the contrary, the latter seems to exaggerate the weakening effect of on the longitudinal compressive strength, probably stemming from an overestimation of microstructural imperfections.
In the current FKT model, fiber-dominated failure is initiated exclusively by matrix failure in the region of misaligned fibers. Therefore, there must be an initial misalignment angle to ensure the formation of a kink band under pure longitudinal compression conditions. Taking the material studied in this paper as an example, its initial misalignment angle, as determined from the longitudinal compressive strength, is approximately 4
. This value is slightly above the suggested range [
39], and well above the standard deviation of fiber misalignment reported in [
9], i.e., the standard deviation in the fiber direction is only about 0.5
for both unidirectional and quasi-isotropic laminates.
Figure 17 also validates a question raised above as to whether the initial fiber misalignment angle in the FKT model is real and continues to have an effect even though longitudinal compression is not a major factor. Assuming initial misalignment angles of 1
, 2
, and 4
, and considering their role in specimen failure at off-axis angles ranging from 10
to 30
, it is not difficult to invert the critical stresses for matrix failure in the deflection coordinates of the FKT model. These outcomes and their variation patterns contradict most of the existing experimental and theoretical findings. Therefore, it can be concluded that the determination and interpretation of the initial fiber deflection angle in the FKT-based criterion needs to be re-examined.
From a microscopic perspective, fiber kinking can be viewed as a chain reaction resulting from a microstructural disruption that is multifactorial. Recent experimental evidence suggests that kink bands originate not only in regions of severe fiber misalignment but also in regions where fiber rotation has been induced by other failure mechanisms. Gutkin et al. [
40] conducted an experimental study on the compressive failure of single-edge notched carbon/epoxy specimens under longitudinal compression, with particular insight into the sequence of events leading to failure at the microscale. Failure initiates in the 0
-ply as a 45
crack near the notch and subsequently evolves into a kink band due to fiber rotation at the crack front. In situ and post mortem fractographic analysis of the failure process, coupled with numerical discussions presented in a separate paper [
41], confirms the existence of shear-driven fiber compression failure in addition to the well-known kinking/splitting.
Informed by these facts, it is rational to speculate that the
failure envelope should be composed of two segments, as schematically drawn in
Figure 19: the first runs nearly parallel to the
-axis in the high compression region and the second approximates a straight line with a positive slope. The location of the segment for shear-driven failure along the
-axis is determined based on the longitudinal strength
. As a rule of thumb, a modest influence of the shear stress should also be considered. In practice, however, the experimental data needed to calibrate the parameters of interest are often lacking, so the maximum stress criterion is still adopted as an approximation. On the other hand, the slope of the segment associated with kinking/splitting depends on the initial fiber misalignment angle, which is now an independent parameter denoted as
for distinction. When
is very high or
is very large, the first segment vanishes completely, thus degenerating into the classic FKT-based criterion.
Figure 20 illustrates some simple modifications to the FKT-based criterion.
is determined by iteratively computing the
failure envelope to optimize its alignment with experimental results in the range of
(Method 1). Notably, the value of
is significantly smaller than the
of the pure FKT model. This implies that the microstructural imperfections do not appear to be as severe as originally anticipated and a prediction with
is perfectly acceptable (Method 2). Henceforth, the criteria for kinking/splitting and matrix cracking can be harmonized, provided that the matrix failure criterion is checked in a co-rotational coordinate system Another possible modification is to assume that the influence of initial fiber imperfections gradually diminishes as the longitudinal compression decreases. Thus,
is defined as a function of
, which decreases from
at
to 0
at
(Method 3). It is seen that, with these improvements, the prediction of pure longitudinal compression remains unchanged, but there is a noticeable improvement in the case of small off-axis angles or combined in-plane shear.
Overall, there are still challenges in accurately predicting fiber compressive failure under combined or off-axis loads. Some of the underlying failure mechanisms may not yet be fully comprehended. The proposed modifications to FKT-based criterion demonstrate improved agreement with experimental results, but further validation regarding the reasonableness and robustness of these predictions is needed.