Introduction

Given the current trends of globalization and active competition on the world market, especially for automotive, the reduction of the lead time can be decisive. Virtual manufacturing using finite element analyses contribute to this reduction. The finite element analysis has been applied extensively to compare design options, understand the influence of process conditions on both formability and structural performance and to reduce the trial and error of tools for optimum performance. For realistic simulations, the use of improved constitutive models for the description of the mechanical behaviour and accurate prediction of formability are essential.

In particular, the use of yield conditions that capture the key features of the plastic behaviour during forming operations are essential. Intrinsic anisotropy associated to a given crystal structure as well as the anisotropy induced by processing of polycrystalline metallic materials place severe restrictions on the form of yield conditions. In Section Anisotropic yield criteria, after a brief review of the classical isotropic yield functions, we discuss the rigorous approaches that enable the extension of isotropic yield criteria such as to account for specific material symmetries. Examples of very versatile anisotropic two-dimensional (2D) and three-dimensional yield criteria (3D) are provided for metallic materials with various crystal structures. For the case of 3-D yield orthotropic yield criteria based on Hershey-Hosford yield criterion involving six anisotropy coefficients, equivalent expressions in terms of stresses and analytical identification procedures which enable to directly correlate the anisotropy parameters to mechanical properties, and consequently standardize their use in F.E. codes are discussed. The capabilities of anisotropic yield functions to describe the plastic behaviour for loading conditions other than those used for parameter identifications are also illustrated for FCC aluminium, HCP magnesium and titanium materials. Specifically, it is shown that currently it is possible to describe with great accuracy the earing profile of certain highly textured aluminium alloys. The importance of consideration of the combined effects of anisotropy and tension-compression asymmetry in modelling yielding of HCP materials is put into evidence for torsional loadings and bulge forming. The influence of the particularities of yielding on damage evolution and ultimately failure and recent yield functions accounting for plasticity-damage couplings in both FCC and HCP materials are discussed in Section Formability of metallic materials.

The anisotropic yield conditions employed in sheet metal forming simulations usually contain one state variable that account for the strength of the material or the associated plastic work-based effective strain. This means that hardening is isotropic and that the yield surface keeps the same shape during plastic deformation. However, isotropic hardening in not always sufficient to accurately predict plasticity during an entire forming process [1], for instance, for the prediction of sprinbgback [2]. In fact, beside process parameters, it is known that the prediction of springback is very sensitive to material properties including the variations of the elastic modulus as discussed in [3, 4]. Nevertheless, regarding plastic anisotropy, springback strongly depends on the Bauschinger and permanent softening effects that can be mostly observed after one or several load reversals. In these deformation-induced plasticity cases, it is necessary to capture the anisotropic hardening effects. This require the introduction of additional state variables in the formulations allowing drastic changes of the yield condition after severe loading changes. Non-linear kinematic hardening [5] that corresponds to a translation of the yield condition in stress space and / or distortional plasticity have been useful for this purpose. In fact, kinematic models with two or several surfaces [6, 7] have been successfully employed for the modeling of anisotropic hardening effects in forming simulations.

Since the approaches such as those briefly reviewed in Section Anisotropic yield criteria require proper validation to ensure that they are appropriate for given materials and processes, careful uniaxial and multiaxial experiments are required, which is the topic of Section Experimental validation of the anisotropic models. Moreover, it is important to note that the prediction of formability, as discussed in Section Formability of metallic materials, strongly depends on the constitutive model employed in the failure analyses considered [8]. This explains why these three subjects are treated simultaneously in a single review. The plastic deformation behavior of sheet metals is commonly characterized using uniaxial tensile test data, i.e. uniaxial yield stresses and r-values with respect to several tensile directions. In addition, a uniaxial compressive yield stress in the thickness direction can be used as the equi-biaxial yield stress for pressure-independent materials [9] [10] (it should be noted that the in-plane stress at the apex of a hydraulic bulge specimen is not necessarily equi-biaxial if the material is anisotropic [11]). The deformation characteristics for other stress states are automatically determined by the material model (yield function) assumed in the analysis. However, there is no guarantee that the assumed material model can accurately reproduce the deformation characteristics of the material under the stress states other than those used for the material characterization. In actual sheet metal forming processes the material is subjected to complex multiaxial stress paths. In addition, unloading makes the material trace multiaxial unloading stress paths (nonlinear unloading behavior of the material significantly affects the magnitude of springback). Therefore, to improve the prediction accuracy of sheet metal forming simulations, the material characterization should be based on the data obtained using multiaxial stress tests.

The shape of the yield surface changes depending on the strain history. It is therefore impossible to formulate the evolution of the yield surface associated with arbitrary strain paths as there are an unlimited number of strain paths occurring during actual metal forming. From this reason a practical material characterization using multiaxial stress tests is usually based on a linear stress path experiment; stress-strain curves are measured for several different linear stress paths and a contour of plastic work and the directions of the plastic strain rates, Dp, associated with a specific value of reference plastic strain are determined. The work contour is assumed to represent a yield surface of the material. The yield function that accurately reproduces the work contour and the direction of the Dp is identified as a proper material model for the material, e.g. [12,13,14]. Moreover, it is possible to take into account the evolution of the work contour by changing the material parameters of the yield function as a function of the plastic work per unit volume, or equivalently, of the reference plastic strain, e.g. [13, 15, 16].

The hydraulic bulge test with optical measuring systems and the biaxial tensile test using a cruciform test piece have been standardized as ISO 16808 [17] and ISO 16842 [18], respectively, in 2014. That is an important step in evolving the knowledge about the deformation characteristics of materials under multiaxial stress and in improving the accuracy of material characterization. Shear [19,20,21] and plane strain tension [22] tests are also effective in material characterization. In particular shear tests are useful to measure the work hardening characteristics in a strain range far exceeding that of uniform elongation in a uniaxial tensile test. The use of full-field measurements, e.g. digital image correlation (DIC), makes it possible to choose complex geometries for the test specimens [23, 24], introducing heterogeneous strain fields. This enables the plastic deformation of the test specimen to be probed at many different stress states at once; a proper material model can be identified by minimizing the difference between the experimental and simulated strain distributions. The virtual fields method (VFM) integrated into a DIC platform is another efficient technique for extracting material parameters from full-field measurements [25,26,27,28]. It is computationally time efficient and there is no need to use FEA; therefore, it emerges as a user-friendly tool for identifying proper material models in industrial forming simulations. We present these experimental methods for material characterization in more detail in the Section Experimental validation of the anisotropic models.

In the last decade, researchers in the field of sheet metal formability analysis have focused their attention on the following directions: the development of new concepts for FLC; refining the method for experimental determination of FLC; improving their prediction models. New concepts such as “Generalized Forming Limit Concept (GFLC)” ([29, 30]) and “polar diagram of the effective plastic strain- PEPS diagram” ([31, 32]) have emerged, promising an increasing applicability in industrial practice, which is essential for the analysis of processes that have a pronounced non-proportional strain-path. The standardization of the FLC determination method in 2008 [33] was an important step in increasing the robustness and accuracy of determining the limit deformations. Subsequently, the German IDDRG Group, based on the experience of companies that developed the Digital Image Correlation (DIC) method, proposed a new method based on a time-dependent technique that was standardized by ISO in 2012 [34]. In recent years, the time-dependent method has been improved by various groups in Germany [35] [36], Spain [37], US [38], Canada [39], Norway [40], etc. Scientists proposed new procedures for FLC determination, such as hydraulic bulging of a double specimen [41] or cruciform specimens test [35, 42] while also analyzing the influences of different parameters on the shape and position of FLC, such as: temperature [43,44,45]; strain rate [43, 44]; strain path [29] etc. Crystal-plasticity-based FLC prediction is another research area that has recently seen significant results [46,47,48]. Implementing ductile damage models (Gurson-Tvergaard-Needleman -GTN- model) in FE codes has improved the FLC prediction [49]. Several research groups, especially Iranian and Chinese, have studied the effect of the normal pressure on the formability [50,51,52,53,54] etc. Finally, for topics such as the formability of multi-layer (sandwich) sheets ([55,56,57,58,59,60,61]), the extension of the Modified Maximum Force Criterion-MMFC ([62, 63]) or the perturbation approach ([64, 65]) have also seen significant progress in the last decade. We present these approaches in more detail in the following sections of the paper. Due to the volume limitation, this paper does not cover all areas of research in the field. Specifically, topic such as sheared edge fracture [66, 67] is not covered in this paper.

Anisotropic yield criteria

For a fully-dense metallic material, the plastic response is classically considered as independent of the hydrostatic pressure (e.g. for single-crystals, see [68, 69]; for polycrystalline materials, see [70]). Thus, the yield function is a function of s, the deviator of the Cauchy stress tensor, σ. If the material is isotropic, the yield function should have the same form irrespective of the coordinate frame. This requirement dictates that the yield function depends on s through its invariants, J2 and J3. It follows, that an isotropic yield function should be a symmetric function of the eigenvalues sj, j = 1…3 of s.

The most used isotropic yield function that satisfies this mathematical requirement was proposed by von Mises in 1913. It states that the material enters the plastic regime when the second-invariant, J2, reaches a critical limit. Results of tests under combined tension-torsion on various isotropic materials (e.g. [71]) showed the influence of the third-invariant, J3 on yielding. Examples of classic yield functions depending on both invariants include Tresca, Drucker [72] and Hershey-Hosford (see [73, 74]). Drucker yield function is defined as:

$$ {J}_2^3-c{J}_3^2={\tau}_Y^6, $$
(1)

where τY is the yield stress in pure shear, and c is a constant. This constant is expressible as the ratio between the yield stress in uniaxial tension, σT, and τY (e.g. see [75]). For c = 0, the Drucker [72] yield surface reduces to the von Mises’s one while for c > 0, it lies between those of von Mises and Tresca’s. The Hershey-Hosford yield function is expressed in terms of the eigenvalues sj as

$$ \phi ={\left|{s}_1-{s}_2\right|}^a+{\left|{s}_2-{s}_3\right|}^a+{\left|{s}_3-{s}_1\right|}^a=2{\overline{\sigma}}^a $$
(2)

where a is a constant larger than 1 while \( \overline{\sigma} \) defines the effective stress associated with this criterion (i.e. for simple tension it reduces to the uniaxial yield stress, denoted σT). For an exponent a = 2 or a = 4, Eq. (2) reduces to the von Mises criterion, whereas for a = 1 and the limiting case a → ∞, it leads to Tresca yield condition. For BCC and FCC isotropic materials the recommended values are a = 6, and a = 8, respectively (e.g. see [76]).Very recently, it has been demonstrated that for any even and positive integer a, the Hershey-Hosford yield function is a homogeneous polynomial of J2 and \( {J}_3^2 \). In particular, Hershey-Hosford yield function for BCC materials (a = 6) and FCC materials (a = 8) are is given by:

$$ {\displaystyle \begin{array}{c}{\left({s}_1-{s}_2\right)}^6+{\left({s}_2-{s}_3\right)}^6+{\left({s}_3-{s}_1\right)}^6=66\kern0.1em {J}_2^3-81{J}_3^2,\\ {}{\left({s}_1-{s}_2\right)}^8+{\left({s}_2-{s}_3\right)}^8+{\left({s}_3-{s}_1\right)}^8=258\kern0.1em {J}_2^4-648{J}_2{J}_3^2\end{array}} $$
(3)

It is clearly seen that Hershey-Hosford with (a = 6) is a particular case of Drucker [77] corresponding to c = 81/66 (see Eq. (1)). For more details and the new expressions of Hershey-Hosford criterion in terms of invariants for other values of the exponent a, the reader is referred to Cazacu [78] and the monograph of Cazacu, Revil, Chandola [79].

Generally, in polycrystalline metallic sheets the constituent crystals are not randomly oriented, but are distributed along preferred orientations that result from rotations that occur during processing. For a given fabrication process, the textures that develop contain one or several ideal components (e.g. [80, 81]). In the framework of crystal plasticity, the most widely used approach for describing plastic anisotropy and for forming of single crystal or textured polycrystalline sheets is to use Schmid law for modeling slip at the single crystal level, and Taylor’s assumption of homogeneous deformation of all crystals ([82, 83], generally known as Taylor-Bishop-Hill model, see also later discussion in the next sections). While increasingly complex homogenization schemes have been proposed, use of such models for large-scale metal forming is still limited, mainly due to the prohibitive computational cost (e.g. see [84]).

Using the new single crystal yield criterion [85], Cazacu et al. [75] showed that for ideal texture components, the yield stress and plastic strain ratios can be obtained analytically. For the case of strongly textured sheets containing a spread about the ideal texture components, these authors showed that the polycrystalline response can be obtained numerically on the basis of the same single-crystal criterion with appropriate homogenization schemes (see [86]). Moreover, it was shown that for textures with misorientation scatter width up to 25°, the numerical predictions are very close to those obtained analytically for an ideal texture and irrespective of the number of grains in the sample Lankford coefficients have finite values for all loading orientations. Illustrative examples for sheets with textures containing a combination of few ideal texture components were also presented in [86, 87]. For examples of applications of this polycrystalline model to industrial textured steel and aluminum alloys, the reader is referred to [79].

Orthotropic yield criteria for textured polycrystalline metals with cubic crystal structure

In the framework of the mathematical theory of plasticity, the plastic anisotropy polycrystalline textured metallic materials is modeled using analytical yield criteria in conjunction with appropriate hardening descriptions (see Section Anisotropic hardening). An extension of the isotropic Mises criterion such as to account for orthotropy was proposed by Hill [88]. For general loadings, this criterion involves six anisotropy coefficients. The use of this yield function in conjunction with associated flow rule, and isotropic or kinematic hardening laws has led to significant advances in metal technology, in particular in forming of certain textured steels ([79, 84]). With the development of new aluminum alloys, it has become evident the need for yield criteria that could describe the observed plastic anisotropy of these materials. To this end, several attempts have been made to extend to orthotropy Hershey-Hosford’s yield function given by Eq. (2) (e.g. Hill [89]). The first formulations involving shear stress components that are well-posed and fulfill the mathematical restrictions imposed by orthotropy were developed by Barlat and collaborators, namely Yld89 yield function for plane stress loadings (see [90]), and Yld91 yield function for three-dimensional (3-D) loadings (Barlat et al. [76]). As demonstrated in Karafillis and Boyce [91], Yld91 can be obtained by substituting in Eq. (2) the stress \( \overline{\sigma} \) with a transformed stress tensor \( \overset{\sim }{\mathbf{S}} \)defined as:

$$ \overset{\sim }{\mathbf{S}}=\mathbf{C}\boldsymbol{\upsigma } $$
(4)

where C is a symmetric fourth-order tensor, orthotropic, and deviatoric. Similarly, it can be shown that in fact according to Hill [88] criterion the onset of plastic deformation occurs the second-invariant of \( \overset{\sim }{\mathbf{S}} \) reaches a critical value. Although Yld91 represents a clear improvement over Hill [88] criterion, its use is more limited than Hill’s or Yld89 or more recent 2-D orthotropic yield criteria (for examples of 2-D yield criteria, see the monograph of Banabic [92]).

This is mainly due to the fact that in its original formulation this criterion is expressed in terms of the eigenvalues of the transformed stress tensor \( \overset{\sim }{\mathbf{S}} \) and as such the quality of the parametrization depends on the experience of the analyst. Recently, Cazacu [78, 93] derived equivalent explicit expressions of Yld91 [76] and the Karafillis and Boyce [91] orthotropic yield criterion in terms of the Cauchy stress components and established direct correlations between the anisotropy coefficients and mechanical properties. Moreover, it was shown that as in the case of Hill [88] criterion, the anisotropy coefficients involved in these yield criteria can be determined analytically based only on four yield points or only using the three Lankford coefficients and the uniaxial tension yield point along RD. These analytical identification procedures enable to directly correlate the anisotropy parameters to mechanical properties, and consequently standardize their use in finite element (F.E.) codes. Another 3-D orthotropic yield function involving a unique linear transformation was proposed by Cazacu and Barlat [94]. This orthotropic yield function, denoted by these authors YLDLIN, was obtained by substituting in the expression of Drucker’s criterion given by Eq. (1), the stress tensor by \( \overset{\sim }{\mathbf{S}} \), i.e.

$$ \varPhi =g\left(\overset{\sim }{\mathbf{S}}\right) $$
(5)

It is also important to note that for plane-stress loadings, Hill’s criterion or any other orthotropic yield criterion obtained using one linear transformation of type (3) involves four anisotropy coefficients. As a consequence, these criteria cannot predict with accuracy both the anisotropy in r-values and yield stresses. The implications in terms of accuracy of the predictions for various forming processes have been discussed in several review papers (e.g. [84]) and benchmarks.

An important question when modeling anisotropic materials is related to the number of anisotropy coefficients involved in the formulation. This is because material symmetries place severe restrictions on the form of the yield functions. The number of independent anisotropy coefficients is not arbitrary, the general form of a well-posed yield function and the number of anisotropy coefficients being dictated by invariance properties associated with the material’s symmetries. For example, orthotropy dictates that for general 3-D loadings the yield function cannot contain more than 18 anisotropy coefficients (for the full mathematical proof and the orthotropic form of the invariants J2 and J3, the reader is referred to Cazacu and Barlat [94]).

A rigorous approach that allows to extend any isotropic criterion such as to account for any material symmetry consists in substituting in the expression of the isotropic yield function, J2and J3 with their anisotropic counterparts. Using general theorems of representation of tensor functions, Cazacu and Barlat [94, 95] developed orthotropic invariants of the stress tensor, denoted as \( {J}_2^o \) and \( {J}_3^o \), respectively. \( {J}_2^o \)and \( {J}_3^o \)are homogeneous polynomials of degree two, and three in stresses, respectively; invariant to any transformation belonging to the symmetry group of the material; pressure-insensitive, and for isotropic conditions reduce to J2 andJ3, respectively. Irrespective of the mathematical form of the isotropic yield function, the anisotropic yield criterion obtained using the linear transformation approach is a particular form of the extension of the same isotropic criterion obtained using orthotropic invariants (see [79, 84]).

An orthotropic extension of Drucker’s yield function expressed in terms of the orthotropic invariants \( {J}_2^o \) and \( {J}_3^o \) was proposed by Cazacu and Barlat [94]. This 3-D yield function involves the maximum admissible number of anisotropy coefficients was derived by Cazacu and Barlat [94]. It has been applied to various aluminum alloys, its predictive capabilities being demonstrated in benchmarks (e.g. Numisheet 2016 benchmark 1, see [77, 96, 97]. Recently, Cazacu [98] proposed the following orthotropic yield function:

$$ \phi ={\left({J}_2^o\right)}^4-\alpha \left({J}_2^o\right){\left({J}_3^o\right)}^2 $$
(6)

with α being a parameter. This criterion was mainly used for description of the plastic anisotropy of FCC materials (e.g. [99].

In the last two decades, very versatile orthotropic yield functions were developed using two linear transformations. The corresponding transformed stress tensors are defined as:

$$ \overset{\sim }{\mathbf{S}}^{\prime }=\mathbf{C}^{\prime}\mathbf{s},\overset{\sim }{\mathbf{S}}^{\prime \prime }=\mathbf{C}^{\prime \prime}\mathbf{s}, $$
(7)

with the fourth-order tensors Cand C being orthotropic. For plane stress, the anisotropic Yld2000-2d yield function was defined as (see [100]):

$$ F={\left|{\overset{\sim }{S}}_1^{\prime }-{\overset{\sim }{S}}_2^{\prime}\right|}^a+{\left|2{\overset{\sim }{S}}_2^{{\prime\prime} }+{\overset{\sim }{S}}_1^{{\prime\prime}}\right|}^a+{\left|2{\overset{\sim }{S}}_1^{{\prime\prime} }+{\overset{\sim }{S}}_2^{{\prime\prime}}\right|}^a=2{\overline{\sigma}}^a $$
(8)

\( {\overset{\sim }{\mathbf{S}}}_1^{\prime },{\overset{\sim }{\mathbf{S}}}_2^{\prime } \) and \( {\overset{\sim }{\mathbf{S}}}_1^{{\prime\prime} },{\overset{\sim }{\mathbf{S}}}_2^{{\prime\prime} } \) are the principal values of \( {\overset{\sim }{\mathrm{s}}}^{\prime } \)and \( {\overset{\sim }{\mathrm{s}}}^{{\prime\prime} } \), respectively. It involves eight independent coefficients, namely three \( {C}_{ij}^{\prime } \) and five \( {C}_{kl}^{{\prime\prime} } \). The capabilities of this yield function to describe the plastic anisotropy of textured aluminum and steel sheets have been demonstrated for numerous forming applications (see [84, 100]).

Based on the Barlat and Richmond [101] formulation, independently, Banabic and collaborators, developed since 2000 (see for example [102]) the BBC family criteria. The BBC2005 criterion (see [103,104,105]).) is defined as:

$$ \alpha {\left|\varGamma +\varPsi \right|}^a+\alpha {\left|\varGamma -\varPsi \right|}^a+\left(1-\alpha \right){\left|2\varLambda \right|}^a={\sigma}^a $$
(9)

where Γ, Ψ and Λ involve the three plane stress components and anisotropy coefficients. It was shown that this yield function contains eight independent coefficients and that, in fact, it is the same as Yld2000-2d only written in a different form (see [106]).

For highly anisotropic sheets, a plane-stress yield function, called BBC2008, expressed in the form of a finite series that can be expanded to retain more or fewer terms, depending on the available experimental data was proposed by Comsa and Banabic [107]. The capabilities of the BBC2008 yield criterion to predict the earing profiles of the aluminum alloys AA5042-H2 and AA2090-T3 were illustrated in Vrh et al. [108].

For certain anisotropic materials for which the microstructure evolution occurs even for the simplest loading conditions (e.g. monotonic simple tension or compression) or for applications where texture evolution during plastic deformation cannot be neglected, the anisotropy coefficients involved in orthotropic yield criteria cannot be considered as being constants. Since it is not possible to perform all the mechanical tests necessary to calibrate evolution laws for all the anisotropy coefficients, instead virtual tests using crystal plasticity codes may be conducted. Such an approach was proposed by Plunkett et al. [109] who used the self-consistent crystal plasticity code VPSC ([110]) to calibrate Cazacu et al. [111] yield function and thus model the evolving anisotropy of hcp zirconium during monotonic loadings. To account for texture evolution during cup drawing of aluminum alloy AA6016, recently, Gawad et al. [112] used the Alamel crystal plasticity model (e.g. [113]) to provide adaptive updates of the local anisotropy in the integration points of a macroscopic F.E. model with yielding described by the BBC 2008 criterion [107]. Moreover, an enhanced yielding description and computationally efficient identification strategy for the anisotropy coefficients involved in the formulation was proposed. For general loadings, Barlat et al. [114] proposed the yield function Yld2004-18p, which extends Eq. (2)

$$ \varphi \left({\overset{\sim }{S}}_i^{\prime },{\overset{\sim }{S}}_j^{{\prime\prime}}\right)=\sum \limits_{i,j=\mathrm{1...3}}{\left|{\overset{\sim }{S}}_i^{\prime }-{\overset{\sim }{S}}_j^{{\prime\prime}}\right|}^a=4{\bar{\sigma}}^a $$
(10)

with the fourth-order tensors Cand C (see Eq. (7)). Specifically, relative to the orthotropy axes, the transformed stress tensors are written in matrix form as

$$ \tilde{S}\equiv \left[\begin{array}{c}{\tilde{s}}_{xx}\\ {}{\tilde{s}}_{yy}\\ {}{\tilde{s}}_{zz}\\ {}{\tilde{s}}_{yz}\\ {}{\tilde{s}}_{zx}\\ {}{\tilde{s}}_{xy}\end{array}\right]=\left[\begin{array}{cccccc}0& -{c}_{12}& -{c}_{13}& 0& 0& 0\\ {}-{c}_{21}& 0& -{c}_{23}& 0& 0& 0\\ {}-{c}_{31}& -{c}_{32}& 0& 0& 0& 0\\ {}0& 0& 0& {c}_{44}& 0& 0\\ {}0& 0& 0& 0& {c}_{55}& 0\\ {}0& 0& 0& 0& 0& {c}_{66}\end{array}\right]\left[\begin{array}{c}{s}_{xx}\\ {}{s}_{yy}\\ {}{s}_{zz}\\ {}{s}_{yz}\\ {}{s}_{zx}\\ {}{s}_{xy}\end{array}\right] $$
(11)

with the appropriate symbols (prime and double prime) for each transformation. Of the 18 parameters (nine per linear transformation) involved in Yld2004-18p, two can always be set equal to unity (for more details, see van den Boogaard et al. [115]). It was shown that this yield criterion predicts with great accuracy the plastic response of highly anisotropic materials [116, 117] (e.g. see Fig. 1).

Fig. 1
figure 1

Finite-element simulations of earing cup profiles for 5019A-H48 aluminum alloy using the Yld2004-18p stress potential in comparison with experimental data (simulation data provided by Prof. Jeong-Whan Yoon) [117]

Yoshida and collaborators [11] proposed a yield criterion defined as the sum of “n” YLDLIN yield functions. Although the formulation involves 6xn parameters, not all these parameters are independent. For an in-depth discussion of other anisotropic formulations the reader is referred to the recent monographs [79, 92] [118] and reviews presented in (e.g. [84, 117, 119,120,121,122,123,124,125,126].

Anisotropic yield criteria for hexagonal-close packed metals

Improvements in energy efficiency and reduction of greenhouse gas emissions have been one of the central concerns in the last two decades, with light metallic alloys based on magnesium and titanium receiving increased attention. For example, there has been a renewed interest in magnesium alloys in view of potential applications in the automotive, computer, communication and consumer electronic products industries. The global magnesium production each year was about 250,000 tons before 2000, but over the past 5 years an average of more than 650,000 tons of magnesium metal was produced each year (see [127]. However, existing applications are mainly based on cast products. The reasons for the limited use of sheets are related to their comparably poor formability, especially at room temperature (e.g. see [128]. This is because most cold-rolled hexagonal close packed (hcp) alloy sheets have basal or nearly basal textures. As a result, the yield surfaces are not symmetric with respect to the stress free condition (see [129, 130]). Since HCP metal sheets exhibit strong textures (e.g. for AZ31 Mg the c-axis of most grains is oriented predominantly perpendicular to the thickness direction), a pronounced anisotropy in yielding is observed. As mentioned in the previous section, the rigorous methods to extend any isotropic yield criterion such as to account for the initial plastic anisotropy (initial texture) or to describe an average material response over a certain deformation are general, and as such applicable to any material, including hcp metals. The major difficulty encountered in formulating analytic expressions for the yield functions of hcp metals is related to the description of their tension-compression asymmetry. Yield functions in the three-dimensional stress space that describe both the tension-compression asymmetry and the anisotropic behavior of fully-dense hcp metals have been developed. To describe yielding asymmetry in isotropic pressure-insensitive materials that results either from twinning or from non-Schmid slip effects at single crystal level, Cazacu and Barlat [131] proposed an isotropic criterion expressed in terms of all invariants of the stress deviator:

$$ f\equiv {\left({J}_2\right)}^{\frac{3}{2}}-c{J}_3={\tau}_Y^3, $$
(12)

where τY is the yield stress in pure shear and c is a material constant. For c = 0, this criterion reduces to the von Mises yield criterion. It was shown that this isotropic yield criterion describes with great accuracy the crystal plasticity simulation results of Hosford and Allen [130] for randomly oriented polycrystals deforming solely by twinning or for isotropic polycrystals with constituent grains deforming by slip governed by a modification of Schmid law involving normal stress components that was proposed by Vitek et al. [132] (for more details, see Cazacu and Barlat [133]). The isotropic criterion (10) was further extended such as to incorporate orthotropy using the generalized invariants approach. The orthotropic criterion is defined as:

$$ f\equiv {\left({J}_2^o\right)}^{\frac{3}{2}}-c{J}_3^o $$
(13)

where \( {J}_2^o \) and \( {J}_3^o \) denote the orthotropic generalizations of the isotropic invariants J2, J3 respectively. For general stress states, the criterion involves 17 anisotropy coefficients. Comparison between this orthotropic criterion and data on magnesium and its alloys (see Graff et al. [134]) and titanium alloys (see Cazacu and Barlat [131]) show that this orthotropic criterion accurately describes both the anisotropy and tension-compression asymmetry in yielding of these materials. If for a given material mechanical data are limited, it is recommended to use the orthotropic criterion of Nixon et al. [135]. This latter criterion is the extension of the isotropic yield function given by Eq. (10). It was obtained using one linear transformation, the fourth-order tensor C in Eq. (3) being taken symmetric, orthotropic, and deviatoric.

Cazacu et al. [111] developed another isotropic pressure-insensitive yield criterion that accounts for yielding asymmetry between tension and compression. This isotropic criterion involves all principal values of the stress deviator and is defined as

$$ \varphi (s)={\left(\left|{s}_1\right|-k{s}_1\right)}^a+{\left(\left|{s}_2\right|-k{s}_2\right)}^a+{\left(\left|{s}_3\right|-k{s}_3\right)}^a $$
(14)

where k is a material constant while a is an integer. For the yield function to be convex: −1 ≤ k ≤ 1. It is important to note that for k ≠ 0, the yield function is not an even function. Therefore, according to the criterion the yielding response in tension and compression are different.

To capture simultaneously anisotropy and tension/compression asymmetry in yielding, the isotropic yield criterion given by Eq. (14) was extended to orthotropy using one linear transformation. The effective stress associated with the orthotropic criterion is:

$$ {\overset{\sim }{\upsigma}}_{\mathrm{e}}=\mathrm{B}{\left[{\left(\left|{\Sigma}_1\right|-\mathrm{k}{\Sigma}_1\right)}^{\mathrm{a}}+{\left(\left|{\Sigma}_2\right|-\mathrm{k}{\Sigma}_2\right)}^{\mathrm{a}}+{\left(\left|{\Sigma}_3\right|-\mathrm{k}{\Sigma}_3\right)}^{\mathrm{a}}\right]}^{1/\mathrm{a}} $$
(15)

In Eq. (15), Σ1, Σ2, Σ3 are the principal values of the transformed stress tensor Σ = Cs, where the fourth-order tensor C is not symmetric; B is a constant defined such that the equivalent stress, \( {\overset{\sim }{\upsigma}}_{\mathrm{e}} \), reduces to the tensile flow stress along RD. Thus, for full 3-D loadings, 8 anisotropy coefficients are involved in the criterion (15).

The quadratic form of this orthotropic yield criterion (i.e. a = 2 in Eq. (15)) was shown to exhibit accuracy in describing the yield loci of a variety of HCP materials. As an example, in Fig. 2a are presented in the (RD-TD) plane the predicted yield loci for a high-purity HCP-Ti material corresponding to several levels of the equivalent plastic strain \( {\bar{\varepsilon}}^p \). To capture the difference in strain hardening rates between tension and compression loadings observed experimentally (see Nixon et al. [135]), all the material parameters involved in the expression of the yield function, namely the anisotropy coefficients as well as the parameter k were considered to evolve with \( {\bar{\varepsilon}}^p \). The equivalent plastic strain \( {\bar{\varepsilon}}^p \) was calculated using the expression of \( {\overset{\sim }{\upsigma}}_{\mathrm{e}} \) given by Eq. (15) and the work-equivalence principle. The calibration was done using as input the flow stresses in uniaxial tension and uniaxial compression as well as the Lankford coefficients. Additional constraints were imposed such as to ensure that for any given stress state, the yield surface at a given level of equivalent plastic strain is exterior to that corresponding to lower levels of plastic strain (i.e. \( \overline{\sigma}\left({L}_{ij}^u,{k}^u,\sigma {\left({\theta}_l\right)}^u\right) \)>\( \overline{\sigma}\left({L}_{ij}^u,{k}^u,\sigma {\left({\theta}_l\right)}^{u-1}\right) \), with σ(θl)u and σ(θl)u − 1 denoting the stress state belonging to the yield surface “u” and “u − 1”, respectively corresponding to same θl arbitrary stress state in the biaxial planes (σxx, σyy) and(σxx, σzz), with θl = 0 corresponding to uniaxial loadings along the x-axis (RD) (for more details, see for example [136]. Because at initial yielding and for strains under 10%, the tension-compression asymmetry of the Ti material is small, according to the criterion the yield surfaces of this material have an elliptical shape. At 20% strain and beyond when the observed difference in response between tension and compression is pronounced, it is predicted that the yield surfaces have a triangular shape. In Fig. 2b are shown the yield surfaces corresponding to different levels of plastic strain for Mg AZ31 that were calculated using the same yield criterion. Note that for Mg AZ31, the criterion predicts that the shape of the surface evolves from a triangular shape for low strain levels to an elliptical shape for large strain levels. The evolution in tension-compression asymmetry with accumulated plastic deformation for AZ31 Mg is completely different from the yield surface evolution for HCP-Ti material. Nevertheless, with the quadratic form of the yield function (15), it is possible to account for these strikingly different yielding evolutions.

Fig. 2
figure 2

Theoretical yield surfaces in the (RD, TD) plane according to the quadratic form of the Cazacu et al. [111] orthotropic criterion corresponding to fixed levels of accumulated plastic strain for: a high-purity alpha Ti; b Mg AZ31 (after [79])

Moreover, on the basis of Cazacu et al. [111] criterion a new interpretation and explanation of the development of plastic axial strains under free-end torsion, the so-called Swift phenomenon, was provided in [131]. Moreover, Revil-Baudard et al. [137] showed that axial strains should develop in free-end torsion of AZ31B Mg, the nature of these axial strains (i.e. elongation or contraction) depending on the tension-compression asymmetry ratio in the direction about which the material is twisted. Fig. 3 show comparisons between F.E. predictions obtained with this criterion and Hill’s criterion and the data of Guo et al. [138]. Note that if the twist axis is along RD both models predict shortening of the specimen, with the Hill criterion largely underestimating the level of strains. On the other hand, for ND torsion, only Cazacu et al. [111] criterion correctly predicts that the specimen elongates, while the Hill criterion predicts zero axial plastic strains (see Fig. 3).

Fig. 3
figure 3

Variation of the axial strain with the shear strain during free-end torsion along RD and ND directions for AZ31 Mg alloy: Comparison between experimental data by Guo et al. [138] and the F.E. predictions obtained with the Cazacu et al. [111] yield criterion and Hill [88], respectively. Note that depending on the twist direction, the axial strain is either positive (for torsion along ND direction) or negative (for torsion along RD direction) (after [136])

Considerable efforts have been devoted to the mechanical characterization of the response of titanium and its alloys for quasi-static strain rates under uniaxial loadings (e.g. [135, 139,140,141,142], etc.). Despite their outstanding mechanical properties because of high processing and manufacturing costs the use of titanium materials is still limited to applications requiring high performance. It is to be noted that the great majority of formed titanium parts are made by hot forming, with the greatest improvement in formability being for temperatures above 540 C. With these challenges in mind, efforts in the past decade have been done towards developing models that would enable a better description of the mechanical behavior of titanium materials during cold forming operations. As an example, Fig. 4 shows the predicted evolution of the thickness at the pole of the bulge as a function of the fluid pressure according to the Cazacu et al. [111] and Hill [88] criterion along with the experimental points (represented by symbols) obtained from post-test DIC in two hemispherical bulge tests on a commercially pure hcp titanium T40. Note that Cazacu et al. [111] yield criterion predicts with accuracy the thickness reduction that occurs in the bulge. Most importantly, this criterion captures the almost “vertical” drop in thickness that occurs at a pressure p = 25.5 MPa. On the other hand, Hill’s criterion greatly underestimates the thickness reduction. According to this model, a thickness of 1.2 mm at the pole would be reached for a fluid pressure p = 29 MPa against 24.4 MPa experimentally, and 24.9 MPa according to the Cazacu et al. [111] yield criterion. For a detailed discussion on the importance of consideration of the tension-compression asymmetry displayed by titanium materials on the predictions of the level of plastic strains at which instabilities occur under hydrostatic bulging for various die inserts geometries, and for the accuracy of the predictions of other cold forming operations, the reader is referred to the monograph [136].

Fig. 4
figure 4

Thickness at the pole of the bulge as a function of the fluid pressure for CP hcp titanium T40: comparison between the F.E. predictions according to the Hill [88] yield criterion (interrupted line), Cazacu et al. [111] orthotropic yield criterion (solid line), and data obtained in two hemispherical bulge tests (symbols) [after 136]

Anisotropic hardening

The anisotropic yield criteria presented above are very useful because they can be implemented relatively easily in FE codes and employed readily for complex applications involving plasticity, including metal forming. Currently, these yield conditions are mostly associated with isotropic hardening or, possibly, with anisotropy coefficients that evolve with the effective strain or any function of the plastic work. However, they cannot handle the anisotropic hardening effects that often occur during strain path changes. In particular, they cannot describe the Bauschinger effect upon load reversal or other transient hardening phenomena that occur during other complex strain path changes. Since it has been shown in many studies that the hardening behavior after reverse loading has a strong effect on springback, anisotropic hardening is important to account for in forming simulations. Typically, kinematic hardening with a back-stress as a tensorial state variable has been used for this purpose, in particular non-linear kinematic hardening as reviewed by Chaboche [143]. Such an approach was developed specifically to account for strain path changes with mutliple tensorial variables by Teodosiu and Hu [144] (see also [145,146,147]). Often, kinematic hardening has been supplemented by yield surface distortion [148,149,150,151]. Other forms of hardening have also been proposed by Kurtyka and Życzkowski [152].

Among the numerous kinematic hardening models available in the literature, the formulation proposed by Yoshida and Uemori (YU) [6] has received much attention in the forming community. This model is based on two surfaces such as in Krieg [153] and Dafalias and Popov [154] with the inner surface called the yield surface and the outer the bounding surface (Fig. 5a). The yield surface only translates in stress space while the bounding surface translates and expands. Therefore, two back-stresses with distinct evolution equations are necessary to describe the translation of the yield and bounding surface centers. The strain hardening rate is also influenced by the contact between the yield and the bounding surfaces and by the expansion rate of the latter. In fact, this framework includes a third surface, the so-called stagnation surface, that controls the amount of permanent softening occurring mostly after reversal, with a center controlled by its own evolution rule. Each of these surfaces can be defined with any isotropic or anisotropic yield function, for instance, one of those described earlier in this Section. The YU model has been used successfully to capture the behavior of materials in forward-reverse loading conditions, in particular, the Bauchinger effect. Its application in finite element simulations of sheet metal forming clearly demonstrated that the prediction of springback was more accurate when the Bauschinger effect was accounted for.

Fig. 5
figure 5

Deviatoric plane representation of multi-surface kinematic hardening and distortional plasticity models. a YU schematic model with yield surface (inner) and bounding (outer) surfaces, reprinted from [6] with permission from Elsevier; b HAH distortional model during uniaxial tension with solid line for isotropic (dash line) and distortional (solid line)

More recently, it has recently been shown that distortional plasticity only i.e., without back-stress, is a viable alternative to account for the Bauschinger effect [114, 150, 155]). The so-called homogeneous anisotropic hardening (HAH) approach proposed by Barlat et al. [114, 156, 157] has been developed in order to employ any isotropic or anisotropic yield conditions as discussed in previous sections and to distort the corresponding shape (see also [155]). One important feature of the HAH approach is that the plastic response under proportional loading is identical to that of the considered yield function under isotropic hardening.

The HAH yield condition can take the form

$$ \bar{\sigma}\left(\mathbf{s},{f}_{-},{f}_{+},\hat{\mathbf{h}}\right)={\left\{\overline{\xi}{\left(\mathbf{s}\right)}^q+{\phi}_h\left(\mathbf{s},{f}_{-},{f}_{+},\hat{\mathbf{h}}\right)\right\}}^{\frac{1}{q}}={\sigma}_R\left(\overline{\varepsilon}\right) $$
(16)

In Eq. (16) \( \overline{\sigma} \)is the effective stress, s the stress deviator, f and f+ two scalar state variables, and q a constant coefficient. \( {\sigma}_R\left(\overline{\varepsilon}\right) \)is a reference stress-strain curve with \( \overline{\varepsilon} \) the plastic work-based effective strain. \( \overline{\xi} \)and ϕh are two functions defining the isotropic hardening yield condition and the distortion, respectively. The function \( \overline{\xi}\left(\mathbf{s}\right) \) is defined as

$$ \overline{\xi}\left(\mathbf{s}\right)={\left[\overline{\phi}{\left({\mathbf{s}}_L\right)}^2+\overline{\phi}{\left({\mathbf{s}}_X\right)}^2\right]}^{1/2} $$
(17)

where sL and sX are two stress deviators obtained by two simple linear transformations of the stress deviator s to account for latent hardening and cross-loading contraction. When these two effects are inactive, \( \overline{\xi}\left(\mathbf{s}\right) \)reduces to \( \overline{\phi}\left(\mathbf{s}\right) \), the effective stress (yield function) associated with isotropic hardening. ϕh determines the amount of distortion that occurs with load reversals

$$ {\phi}_{\mathbf{h}}\left(\mathbf{s},{f}_{-},{f}_{+},\hat{\mathbf{h}}\right)={f}_{-}^q{\left|\hat{\mathbf{h}}:\mathbf{s}-\left|\hat{\mathbf{h}}:\mathbf{s}\right|\right|}^q+{f}_{+}^q{\left|\hat{\mathbf{h}}:\mathbf{s}+\left|\hat{\mathbf{h}}:\mathbf{s}\right|\right|}^q $$
(18)

while \( \hat{\mathbf{h}} \), the so-called microstructure deviator, sets the distortion direction. Other state variables are introduced within the expressions of \( {f}_{-}^q \) and \( {f}_{+}^q \) to define permanent softening. All the variables and coefficients of the HAH model are detailed in references [114, 156, 157]. As a convenient feature, the HAH approach was designed in a way that four sets of constitutive coefficients can be determined independently. The stress-strain curve for monotonic loading is approximated with a mathematical model and the yield function coefficients are determined in a classical manner assuming isotropic hardening. Then, the coefficients corresponding to load reversal, including permanent softening, are optimized and, finally, those that control cross-loading.

A recent version of this model [158], called HAH20, was proposed with improved state variable evolution equations. Moreover, it was shown that the substitution of ϕh in Eq. (16) by an alternative form taking the yield surface normal into account in the distortion was possible and likely more accurate for anisotropic materials. Finally, HAH20 was developed as a pressure-sensitive model in order to account for the higher flow stress in compression (C) than in tension (T) as observed by Spitzig et al. [159]. Note that this influence cannot be clearly observed with only one reversal, e.g. T-C, because permanent softening tends to produce the opposite of a pressure-strengthening effect. However, with two reversals, e.g. T-C-T, the influence of the pressure and permanent softening are more clearly partitioned, particularly for high strength materials.

As an illustration, Fig. 5.b display the normalized yield locus of an EDDQ low carbon steel in the deviatoric plane during pre-strain in uniaxial tension in a direction defined by a unit vector e2. For isotropic hardening, the normalized yield locus keeps the same shape throughout loading. However, for distortional hardening, the yield locus contract near a stress state opposite to loading, allowing the description of the Bauschinger effect and, possibly, in directions orthogonal to loading, corresponding to cross-loading contraction. If load is suddenly changed to another direction, say uniaxial tension in a direction defined by a unit vector e1, the strain hardening is subjected to transient effect until it recover the isotropic hardening curve. Possibly, the new stress state might overshoot the isotropic hardening curve because of latent hardening effect, particularly more so in the cross-loading direction. However, in many advanced materials such as high strength steels and aluminum alloys, the latent hardening effect is not observed. Figure 5.b also suggests that when reloading in uniaxial tension occurs, the normal direction of the yield surface fluctuate until the flow stress recovers fully and permanently the isotropic hardening curve. Assuming the associated flow rule, this explains why the Lankford coefficient (r-value) tends to vary drastically just after reloading.

Both YU and HAH models are useful for the prediction of springback. Choi et al. [160, 161] compared FE simulations results of springback of a U-draw channel for a number of advanced high strength steel sheets of strength close to 1 GPa. For this process in which most of the load changes are reversals, these authors observed that, although not identical, both YU and HAH led to springback profiles in better agreement with the experiments than those predicted with the isotropic hardening model.

Experimental validation of the anisotropic models

In sheet or tube forming processes, materials are subjected to multiaxial loads. Such multiaxial loading experiments are highly desirable for validating the plasticity models to be used in accurate numerical simulations. This section reviews the experimental studies, published after 2010, that investigated the anisotropic plastic deformation behavior of metal sheets and tubes under multiaxial loading to check the validity of anisotropic plasticity constitutive models. See [121, 162] for the related papers published before 2010.

Hydraulic bulge test

The hydraulic bulge test is effective in measuring the work hardening behavior of sheet metals at significantly larger strain levels than those reached in the uniaxial tension test. Readers can see excellent reviews on the previous studies on hydraulic bulge test in [163, 164].

The stress and strain are not uniform in the specimen for the hydraulic bulge test. Therefore, the accuracy of these measurements depends on the uniformity of deformation within the gauge lengths for strain and curvature measurement. They should be small enough relative to the diameter of the die cavity. Yoshida [164] performed a FEA of the hydraulic bulge test using a higher order yield function proposed by Barlat and coworkers [76, 106]. He concluded that for an isotropic specimen accuracy of the stress measurement is within 1% if (i) the meridional strain is evaluated at the middle layer (the bending strain is necessary to be subtracted from the outer surface strain), (ii) the meridional stress is estimated considering the internal pressure acting on the inner surface, and (iii) the geometry of the experimental setup satisfies Rρ/D≤0.1, Rε/D≤0.05, and t0/D≤0.01, where D and t0 are the specimen’s initial diameter and thickness, respectively, 2Rρ is the distance of the two edges of the spherometer for measuring the radius of curvature of the bulging specimen, and 2Rε is the initial gauge length for measuring the meridional strain on the outer surface of the specimen. Moreover, he demonstrated that for typical orthotropic sheet metals, the stress state at the apex deviates from the equibiaxial stress state; 1.01 < σyy/σxx<1.05, depending on the degree of anisotropy of the material, where σxx and σyy are the stress components at the apex in the RD and TD, respectively [164].

An international standard on the determination of biaxial stress–strain curves by means of the hydraulic bulge test with optical measuring systems has been published as ISO 16808 [17]. Mulder et al. [163] performed a detailed analysis of the hydraulic bulge test with optical measuring systems, as suggested by ISO 16808 [17], to evaluate the validity of all assumptions and simplifications applied to the test. They concluded that a highly accurate stress-strain curve can be obtained by fitting the surface coordinates to an ellipsoid shape function and by considering the local strain data to approximate the curvature for the midplane. Moreover, they have demonstrated that the fitting procedure is robust even when the bulge surface geometry includes a realistic revel of measurement noise. Min et al. [165] developed a new method to accurately calculate stresses and strains for both isotropic and orthotropic sheet materials. The new method takes the elastic volume change into account, the bending effect, and the non-balanced biaxial curvatures in the principal directions when calculating the effective stress at the specimen pole. The accuracy is improved in comparison with previous methods including the ISO 16808 [17], which underestimate the thickness at the specimen pole and lead to an overestimation of stresses by up to several percent.

Some authors used the hydraulic bulge test to verify the validity of anisotropic yield functions. Yanaga et al. [166] measured the thickness strain along the meridian directions of the bulged specimens of 6000 series aluminum alloy sheets with high and low cube texture densities. They found that the difference in the thickness strain distributions between the two samples were in good agreement with the FEA results obtained using the Yld2000-2d yield functions identified for the samples using cruciform specimens. Chen et al. [167] developed a methodology for incorporating anisotropy in the extraction of the material stress–strain response from a bulge test. Williams and Boyle [168] applied the elliptical bulge test to the calibration of anisotropic yield functions coefficients for commercially pure titanium and an exhaust grade titanium alloy sheet.

Biaxial stress test using cruciform specimen

Cruciform specimen design

Biaxial tensile testing method using a cruciform specimen and a proper geometry of the cruciform specimen has been standardized by ISO 16842 [18] in 2014 (Fig. 6). Hanabusa et al. [169, 170] performed FEAs of the ISO cruciform specimen to quantitatively evaluate the error of stress measurement, assuming that the material follows the Yld2000-2d yield function [169] or the von Mises yield function [170]. They concluded that the stress measurement error with the ISO specimen is less than 2% when the following conditions are satisfied:

  1. a)

    a ≤ 0.08B, where a is the thickness of the test sample and B is the arm width.

  2. b)

    N ≥ 7, where N is the number of slits per arm.

  3. c)

    L ≥ B, where L is the length of slits made in the arm.

  4. d)

    wS ≤ 0.01B, where wS is the width of slits.

  5. e)

    The positions where strain components are measured are on the centerline of a specimen, approximately 0.35B away from the center of the specimen, parallel to the maximum stress direction.

Fig. 6
figure 6

a Geometry of the cruciform specimen, b and c optimum positions for measuring the normal strain components when b Fx≥ Fy and c Fx<  Fy, where Fx and Fy are the tensile forces in the x- and y- directions, respectively. (ISO 16842 [18])

The maximum plastic strain, \( {\varepsilon}_{\mathrm{max}}^{\mathrm{p}} \), applicable to the ISO cruciform specimen depends primarily on the work hardening rate (n-value) of the test material; \( {\varepsilon}_{\mathrm{max}}^{\mathrm{p}}\le \)0.1 in most cases for materials with 0 ≤ n≤0.3 [18, 170]. A specimen with a reduced-thickness gauge area have been developed to increase \( {\varepsilon}_{\mathrm{max}}^{\mathrm{p}} \) (see [171, 172]).

A methodology is proposed to optimize a cruciform specimen shape for the identification of constitutive laws based on full-field measurements [173]. The effective cross-sectional area of a cruciform specimen for an accurate determination of the stress components from the tensile forces has been discussed [174].

Measurement of forming limit curve using cruciform specimen

Some authors applied cruciform specimens with a reduced-thickness gauge area to the measurement of the forming limit curves (FLC) of an aluminum alloy 5086 with linear and nonlinear strain paths [42, 175, 176] and DP600 with quasi-static and dynamic biaxial tensile tests [177, 178]. The biaxial tensile testing of a titanium alloy was performed to measure the FLC at elevated temperatures [179]. A method for measuring a FLC using a cruciform specimen and a blank holder with adjustable draw beads has been developed [180]. Equibiaxial tensile testing of commercially pure (CP) titanium up to fracture was performed using an optimized cruciform specimen to investigate the ductility, strain hardening exponent, and texture evolution of the test sample [181].

Material modeling based on the biaxial tensile test using cruciform specimen

Biaxial tensile testing using cruciform specimens were applied to measure and model the biaxial deformation behavior of sheet metals: an ultra-low carbon interstitial free (IF) high strength steel [182], cold rolled DDQ steels [183,184,185], a hot rolled steel [186], a precipitation hardened 590 MPa steel [187, 188], a DP 590 steel under continuous loading and unloading [189], DP 780 [190], DP 980 [191, 192] and DP 980 MPa multiphase steel sheet [193]), annealed stainless steel SUS304 foils [194], aluminum alloy sheets (5000 series [192, 195,196,197], 6000-series with different cube textures densities [166] and 6016-O and -T4 [16]), an AZ31 magnesium alloy [198], a pure titanium (JIS Grade 1) [13, 199], a high strength titanium alloy Ti-6Al-4 V [200], and a GH738 nickel-based super alloy at elevated temperatures [201].

The validity of the material models identified using the cruciform tests has been assessed by comparing the experimental results with the numerical analysis results for hole expansion forming [13, 16, 184, 186, 188, 190, 202], Marciniak-Kuczyński-type punch stretching test [183, 187], hemispherical punch stretching test (Nakajima test) [14], the strain distribution on a cruciform specimen [177, 196, 197], and hydraulic bulge forming [166, 194].

Biaxial loading-unloading test

Andar et al. [203] investigated the deformation behavior of BH340 and DP590 steel sheets subjected to biaxial tension with constant stress ratios followed by biaxial unloading with the same stress ratios. An exponential decay model was proposed that provides good reproduction of the unloading stress–strain relations. Sumikawa et al. [204] experimentally obtained the unloading stress-strain curves of high strength steels under four stress states: uniaxial tension, plane strain tension, biaxial tension and shear (in-plane torsion test [205]). Kulawinski et al. [206] investigated a metastable austenitic stainless steel under different biaxial-planar load paths by using a cruciform specimen geometry. The stress state within the cruciform specimens was evaluated by an elastic unloading procedure with subsequent calculation of the stress components. Isotropic initial yielding and non-isotropic hardening were found.

Multiaxial stress test using tubular specimen

In order to increase the magnitude of \( {\varepsilon}_{\mathrm{max}}^{\mathrm{p}} \) that can be attained experimentally the multiaxial tube expansion testing method (MTET) has been developed [183, 199]. A tubular specimen is fabricated by bending and welding, and then combined internal pressure P and axial force T are applied to the tubular specimen using a servo-controlled testing machine. The axial and circumferential true stress components applied to the tubular specimen can be determined and controlled using the measured P, T, the biaxial strain components, and the radius of curvature in the axial direction. For example, the MTET has been applied to a pure titanium sheet (JIS grade #1) [199], a cold rolled interstitial-free steel sheet [183, 184], a DP590 steel sheet [187], a 304 L stainless steel microtube [207] an A5182-O aluminum alloy sheet [195], and A6016-O and -T4 aluminum alloy sheet [16].

Figure 7a shows the results of biaxial stress tests of an aluminum alloy sheet 6016-O using the ISO cruciform specimens for 0.002\( \le {\varepsilon}_0^{\mathrm{p}}\le \)0.04 and MTET for 0.04\( <{\varepsilon}_0^{\mathrm{p}}\le \)0.08, where \( {\varepsilon}_0^{\mathrm{p}} \) is the logarithmic uniaxial tensile plastic strain in RD (see [16]). The isocontours of plastic work measured using biaxial tensile tests are reported for increments of 0.01. The solid lines are the yield loci determined using the Yld2000-2d [100, 208]. The material exhibits differential hardening (DH) as the shapes of the work contours change with work hardening. Figure 7b shows the thickness strain along the radial line parallel to the RD of a hole expansion test specimen (A6016-O) at a punch stroke of 15 mm, compared with those calculated using selected material models. It is clearly seen that the choice of the model has a significant effect on the FEA prediction of the thickness strain distribution. The Yld2000-2d (DH), which reproduces the DH behavior of the material, leads to the closest agreement with the experimental results.

Fig. 7
figure 7

a Stress points forming contours of plastic work of aluminum alloy sheet 6016-O compared with the yield loci determined using the Yld2000-2d yield function [100, 208]. M is an exponent of the Yld2000-2d yield function. b Thickness strain distribution along the RD of a hole expanded test specimen at a punch stroke of 15 mm. c Experimental apparatus for the hole expansion forming (after [16])

Dick and Korkolis [209] proposed an experimental method for generating combined tension and shear stress states in thin-walled tubes, with the purpose of calibrating anisotropic yield functions. Two notches that form the test-section between them are machined. By controlling the orientation of the notches infinite combinations of tension and shear on the test-section can be created.

Yoshida and Tsuchimoto [210] measured the elastoplastic deformation behaviors of thin-walled tubes made of pure aluminum and steel under various combined tension–torsion loadings. They found that the stress state and the strain rate direction are essential parameters characterizing the plastic flow and proposed a pseudo-corner model capable of reproducing experimentally observed plastic deformation behavior. Khalfallah et al. [211] identified the constitutive parameters of several yield criteria, von Mises, Hill [88], Cazacu and Barlat [94] orthotropic criterion calibrated using reduced experimental data obtained from uniaxial tensile tests and a free bulge test and compared the FEA results for height vs. internal pressure curve, thickness distribution and axial bulge profiles to experimental results on tubular materials (mild steel S235 seamed tube and aluminium alloy AA6063 extruded tube).

Shear test and plane strain tension test

Simple shear tests are widely used for material characterization of sheet metals to achieve large deformation without plastic instability. Yin et al. [212] showed that shear stress vs. shear strain curves obtained from different test setups (Miyauchi specimen [162], ASTM [213], and in-plane torsion test [205] are consistent with each other, provided that strains are measured using a DIC system. Fu et al. [214] applied the VFM to a forward-reverse simple shear test to identify the parameters of the homogeneous anisotropic hardening (HAH) model [114] for advanced high strength steel (AHSS) sheets.

Peirs et al. [215] developed a novel shear specimen geometry for sheet metals that can be used over a wide range of strain rates using traditional tensile test device. The specimen has notches with eccentric position that leads to an almost pure-shear stress state up to large strains. Rahmaan et al. [216] applied this specimen geometry to the investigations of fracture behavior of DP600 and 5182-O aluminum alloy sheets with a range of strain rates from 0.01 s−1 to 600 s−1 using in situ digital image correlation (DIC) strain measurement techniques. Abedini et al. [217] investigated the failure behavior of a highly anisotropic magnesium alloy (ZEK100) and a dual phase steel (DP780) at room temperature using two different shear specimen geometries: the butterfly type specimen [218] and that developed by Peirs et al. [215]. They found that the fracture strains obtained using the butterfly specimen were lower for both alloys. Abedini et al. [219] performed simple shear tests for three orientations of 0° (or 90°), 45°, and 135° with respect to the RD of a magnesium alloy (ZEK100-O) to show the asymmetry of flow stress not only in tension-compression regions represented by the 1st and 3rd quadrants of yield locus but also in shear regions represented by the 2nd and 4th quadrants. Moreover, the CPB06 yield criterion with two linear transformations was calibrated with the experimental data.

Rahmaan et al. [220] characterize the high strain rate constitutive behavior and the anisotropic plasticity of three high strength aluminum sheet alloys, 6013-T6, 7075-T6 and a developmental 7xxx-T76, using uniaxial tension, simple shear and through-thickness compression at various strain rates ranging from 10−3 to 103 s −1. The experimental force-displacement curves of a hemispherical punch stretching test and a simple shear test were consistent with the FEA results based on higher order yield functions.

Butcher and Abedini [221] showed that plane strain conditions of pressure-independent metals occur when the third deviatoric stress invariant is zero which leads to a maxima (for plane strain tension) or minima (for shear) in the second deviatoric invariant, and that the data from stress-controlled cruciform tests [16, 183, 190, 191] and simple shear tests [216, 217, 220, 222], support the existence of the generalized plane strain constraints for shear and plane strain tension for FCC and BCC materials. Moreover, they suggested that the generalized plane strain constraints must be enforced upon the plastic potential function during the calibration procedures in order for generalized plane strain loading conditions to be physically-consistent with the assumptions of pressure-independent plasticity.

Flores et al. [223] quantified the influence of the geometry of the plane strain tension specimen proposed by An et al. [22] and material anisotropy on the stress computation error. Aretz et al. [224] proposed a calibration method for identifying yield functions using three directional uniaxial tensile tests and two plane strain tensile tests. They demonstrated that calculated forming limit curves for three materials (IF steel, low carbon steel and 5000-series aluminum alloy) depends strongly on the accuracy of the experimental plane strain tensile test data used for the calibration. Therefore, they suggest that the specimen geometry must be optimized so that the desired plane strain state can be realized.

Combined use of simple mechanical tests

Tian et al. [225] determined a proper material model for an aluminum alloy Al-6022-T4, using uniaxial and plane-strain tension, as well as disk compression experiments. The model is applied to the FEA of a cup drawing process of the material. The thickness and earing profile predictions are in good agreement with the experiments. Abspoel et al. [226] determined the yield loci (contours of equal plastic work) of a wide range of steel sheets and 5xxx, 6xxx and 7xxx series aluminium alloy sheets by the combined use of uniaxial tension, plane strain tension, equibiaxial tension, uniaxial compression in the thickness direction, and simple shear tests. Baral et al. [227] determined the yield loci (contours of equal plastic work) of a 12.7 mm thick CP titanium sheet by the combined use of in-plane and through-thickness uniaxial tension and compression, and plane-strain tension tests.

Zillmann et al. [228] performed biaxial, in-plane compressive testing of deep drawing steel sheets with and without skin-pass treatment, together with biaxial tension and simple shear tests, to determine full yield loci in the principal stress space and to evaluate the tension–compression asymmetry of the test samples.

Mohr et al. [229] evaluated the accuracy of quadratic plane stress plasticity models for a dual phase and a TRIP-assisted steel using both associated and non-associated quadratic formulations. The response of the materials subjected to combined normal and tangential loads were measured. They found that both the associated and non-associated quadratic formulations provide good estimates of the stress–strain response under multi-axial loading.

Application of digital image correlation (DIC) to material modeling

The use of full-field measurements, e.g. digital image correlation (DIC), makes it possible to choose complex geometries for the test specimens, introducing heterogeneous strain fields. That enables the plastic deformation of the test specimen to be probed at many different stress states at once. Coppieters et al. [230] developed a new method to identify the post-necking hardening behavior of sheet metals without using a finite element model. The strain fields in the central zone of a tensile specimen are measured using DIC to calculate the stress fields by assuming a certain yield criterion and hardening behavior. The stress fields are used to compute the internal work in the necking zone, which should be equal to the work exerted by the external forces if the assumed hardening behavior and yield criterion is correct. Based on this hypothesis, the post-necking hardening behavior of a mild steel sheet (DC05) is determined by minimizing the discrepancy between the internal and external work in the necking zone. Coppieters and Kuwabara [27] determined the post-necking stress-strain curve of a cold rolled mild steel sheet (SPCD), assuming the Yld2000-2d yield function. The results were experimentally validated beyond the point of maximum uniform strain using the MTET [183] [199] up to an equivalent plastic strain of 0.35.

The virtual fields method (VFM) [25] is a very efficient technique for extraction of material parameters from full-field measurements. It is fast in the computational time, no need to use FEA, and can be integrated directly into a DIC platform making it more accessible to engineers.

Grédiac and Pierron [231] applied the VFM to the identification of elasto-plastic constitutive coefficients using simulated full-field kinematic data obtained from a thin notched tensile specimen. Experimental validations of the VFM in elasto-plasticity were performed at small strains by Pannier et al. [232], and Pierron et al. [233]. Rossi and Pierron [26] provided a general procedure to extract the constitutive parameters of a plasticity model starting from displacement measurements and using the VFM, which applies to a general three-dimensional displacement field leading to large plastic deformations. The method was validated using a simulated post-necking tensile experiment. Kim et al. [234] characterize the post-necking strain hardening behavior of for three types of sheet metals, DP780, TRIP780 and EDDQ, assuming von Mises yield criterion to retrieve the relationship between stress and strain. Knysh and Korkolis [235] developed a method for identifying the material hardening curves past the limit of necking in uniaxial tension and across a range of strain-rates and temperatures in a fully-coupled way, using full-field measurements of the strain and temperature during testing.

Marth et al. [236] developed an effective and computationally fast method to determine the relationship between true stress and true plastic strain, including post-necking behavior, by using an optical full-field measurement of the localized deformation field. Marek et al. [28] extend the sensitivity-based virtual fields (SBVF) developed by Marek et al. [237] to large deformation anisotropic plasticity. The main advantage of the SBVFs comes from the automation of virtual fields generation.

Teaca et al. [238] measured heterogeneous biaxial strain fields on a non-standard cruciform specimen using an image correlation method and compared them with those calculated from the FEA based on the 8-parameter anisotropic yield function proposed by Ferron et al. [239]. The parameters of the yield function were determined to minimize the difference between the measured and calculated stain fields. Güner et al. [240] applied an optical strain measurement to sheet specimens with varying notch radii. An inverse parameter identification scheme which minimizes the differences between the numerical simulation results and the measurements for principal strains, forces, and biaxial flow stress σb was developed to identify the parameters of the Yld2000-2d yield function. Tardif and Kyriakides [241] developed a systematic methodology to extract the material stress-strain behavior at much larger strains. This was achieved by accurately following the deformation in the necked region of a custom made tensile test specimen. The test was simulated numerically using a 3D FE model and the material response was iteratively extrapolated until the calculated and measured force-elongation matched. The results were used to calibrate the 18-parameter non-quadratic Yld2004-3D yield function [242]. Tsutamori et al. [243] performed biaxial tensile tests using a cruciform specimen with a hole at the center to evaluate the predictive accuracy of the spline yield function they developed. The thickness strain distribution around the hole measured using a DIC system was consistent with the prediction.

Validation of polycrystal plasticity models

Multiaxial stress tests are effective in evaluating the validity of polycrystal plasticity models. An et al. [244] measured yield loci of several forming steel grades using plane strain tension, balanced biaxial tension and simple shear tests. The measured yield loci were consistent with those calculated using a polycrystal plasticity model combining the Taylor full constraint model and the Taylor relaxed pancake model. Yoshida et al. [245] investigated the work hardening behavior of a thin-walled tubular specimen of A3003-O using an axial load-internal pressure-torsion type test machine. They found that the amount of work hardening of the specimen depends on the plastic work and the applied stress path. They concluded that the source of such work-hardening behavior is attributed to the path dependency of the evolution rate of the dislocation density, using a crystal plasticity analysis with hardening models based on accumulated slip and dislocation density. Yamanaka et al. [246] performed numerical biaxial tensile tests based on the crystal plasticity finite element (CPFE) method and the mathematical homogenization approach. The biaxial tensile deformation behavior of a 5182 aluminum alloy predicted by the numerical biaxial tensile tests were consistent with that measured using the ISO-type cruciform specimens [18]. Coppieters et al. [247] performed the biaxial tension tests of a low carbon steel sheet using cruciform [18] and tubular specimens [183]. The measured biaxial deformation behavior of the sample subjected to linear stress paths in the first quadrant of stress space was consistent with that predicted by the ALAMEL multiscale plasticity model [113]. However, the differential hardening behavior of the sample at a small strain range (\( {\varepsilon}_0^{\mathrm{p}}< \)0.05) could not be reproduced by the ALAMEL model. Upadhyay et al. [248] investigated the deformation behavior of 316 L stainless steel cruciform samples subjected to different biaxial load path change to interpret the material response (biaxial Bauschinger effect) using the anisotropic viscoplastic self-consistent polycrystalline model. Hama et al. [249] reproduced the differential hardening under biaxial loadings and tension-compression asymmetry of a pure titanium sheet (JIS Grade 1) using crystal-plasticity FEA. Steglich et al. [250] simulated the plastic deformation of Mg AZ31 under monotonic biaxial loadings using VPSC and successfully reproduced the differential strain hardening character of the sample for a small strain range (less than 0.008). Kim et al. [251] performed nonlinear strain path experiments for a cold-rolled high-strength steel sheet and DP780. The strain paths consist of uniaxial tension (UT) in TD followed by UT in RD and DD, and biaxial tension. The measured data were compared with those calculated using VPSC and the HAH model [114].

In-plane compression and reverse loading tests for sheet metals

Optimum specimen geometry for the in-plane compression and reverse loading test for sheet metals has been proposed [252]. The tension-compression asymmetry of flow stress of steel sheets and its effects on forming simulations was investigated for an ultra-low carbon IF steel sheet [182, 253], cold rolled interstitial-free steel sheet and dual phase high strength steel sheet [254], DP590 steel sheet [255], and DP980 steel sheets [9, 191]. Gröge and Vitek [256] hypothesized that the onset of yielding of single crystals of bcc metals is governed by the yield criterion that is represented as a linear combination of the Schmid stress and three other (non-Schmid) shear stress components, and that the individual non-Schmid stresses contribute differently towards the effective stress in tension and compression, which results in the experimentally observed tension-compression asymmetry of the yield stress in certain bcc metals.

The tension-compression asymmetry of hcp metals was also recently experimentally investigated for magnesium alloy sheets [10, 257,258,259,261], pure titanium [142, 199, 262], and titanium alloy sheets [200, 263, 264].

Dietrich et al. [265] designed a new anti-buckling fixture that allows tension–compression cyclic loading tests on sheet metals. The fixture can be mounted on a conventional tensile testing machine.

Marcadet and Mohr [266] investigated the effect of loading direction reversal on the onset of ductile fracture of DP780 steel sheets using continuous compression-tension experiments on flat notched specimens to identify the parameters of several plasticity models for reproducing the reverse loading stress-strain curves. Moreover, the model is used to estimate the local strain and stress fields in monotonic fracture experiments covering plane stress states ranging from pure shear to plane strain tension.

A new test equipment has been developed to measure the in-plane cyclic behavior of sheet metals at elevated temperatures up to 400 °C [267].

Zecevic et al. [268] developed a comprehensive polycrystal plasticity model capable of predicting the elasto–plastic cyclic deformation behavior of multi-phase polycrystalline metals deforming by crystallographic slip. The model successfully captures the flow response of a DP590 steel sheet under cyclic tension–compression–tension deformation to various levels of plastic pre-strains.

Stress measurement using X-ray diffraction

Jeong et al. [269] investigated the equibiaxial flow behavior of an IF free steel sample was investigated using Marciniak punch test with in situ X-ray diffraction for stress analysis. An experimental technique using a combination of in-situ X-ray diffraction and digital image correlation (DIC) was applied for an IF steel sheet to measure contours of plastic work in stress space up to strain levels of 0.1 [270].

Formability of metallic materials

Formability describes the capability of a sheet metal to undergo plastic deformation in order to get some shape without defects. During the last decades different assessment methods of metal sheets formability have been developed. The most useful tool used to assess the formability is the Forming Limit Diagram (FLD). This method meets both manufacturer and user’s requirements and is widely used in factory and research laboratories. Due to the vastness of the analyzed domain, this paper does not cover all types of defects that limit the formability of sheet metals (necking, fracture, sheared edge, wrinkling, modification of the roughness, etc.). Four main mechanisms limit the deformation of metal sheets: necking, fracture, shearing and wrinkling [271, 272] (see Fig. 8 [272]). The appearance of one or the other of the four ways depends on the type of material and on the deformation process of the sheets. The interplay between necking and fracture is still not well understood and needs further research. We have limited ourselves in this paper only to necking as a defect that limits deformation. A detailed presentation of the FLD method can be found in the literature [32, 92, 120, 273,274,275,276,277,278,279,280].

Fig. 8
figure 8

Forming limit curves with respect to shear fracture/necking and wrinkling [272]

During the last decades, the development in the field of sheet metal forming has been focused on the use of new materials (high strength steels, new aluminum alloys, magnesium alloys, sandwich materials, etc.), materials exhibiting phase transitions during the forming process, innovative forming procedures (nano-micro forming, hydroforming, incremental forming, etc.), use of neural networks, intensive use of numerical simulation, etc. In order to follow this trend, both experimental and theoretical research has been performed in the field of formability. The experimental research has been focused on refining the experimental methods used for detecting and measuring the limit strains (with the aim of enhancing the accuracy and robustness of experimental methods), development of new laboratory tests for determining the Forming Limit Curves (FLC), as well as on studying the influence of various material or process parameters on the limit strains. The theoretical research aimed to improve the predictive capabilities of plasticity models by refining the numerical techniques used for their solution, as well as by incorporating ductile fracture and crystal plasticity models in FLC approaches. The influence of some process parameters (such as temperature or normal pressure) on the limit strains has been also analyzed from a theoretical point of view. The most significant results obtained during the last decade in this domain will be presented in the following subsections.

Advances in experimental investigation of FLC

New concepts related to the manner of defining the Forming Limit Curves have been introduced during the last decade: the Generalized Forming Limit concept (GLFC) [281], as well as a concept based on combining a limit strain model with a fracture model [282]. In the same time, the concepts of Stress-Based Forming Limit Curve (σFLC) (introduced for the first time by the research group of Embury [283, 284] and later developed by Arrieux [285]) and Extended Stress-Based Forming Limit Curve (XSFLC) [286, 287] defined at the end of 1970’s and in 2007, respectively, have got further developments. A detailed description of these concepts can be found in [32]. The most significant limitation of the Strain Forming Limit Curve consists in the difficulty of applying it to non-linear load paths. The FLSD concept therefore appears to be useful, particularly in multistage forming, for predicting the failure of metal tubes and sheets.

In order to extend the application of stress limit curves to a 3-D stress state (presence of through-thickness components of compressive stress), Simha et al. [286] has introduced a new concept, namely Extended Stress-Based Limit Curve (XSFLC). The XSFLC represents the equivalent stress and mean stress at the onset of necking during in-plane loading. Figure 9 shows the three formulations of the Forming Limit Curve concept, namely: strain-based FLC (εFLC), stress-based FLC (σFLC) and Extended Stress-Based FLC (XSFLC), respectively. Figure 9 also presents the loading paths for the three cases: uniaxial stress, plane strain and biaxial stress. A thorough analysis of the conditions for the use of the XSFLC as a Formability Limit Curve under three-dimensional loading is presented in [286].

Fig. 9
figure 9

Schematic of the Strain-Based Forming Limit Curve (εFLC), the Stress-Based Forming Limit Curve (σFLC) and the Extended Stress-Based Forming Limit Curve (XSFLC) [287]

The most important drawback of the FLC is the strain path dependency. During the last decade, the efforts of the scientists have been oriented towards finding a model that overcomes this disadvantage. The Generalized Forming Limit concept (GFLC) has been proposed by Müschenborn and Sonne [288] and revisited by Volk [29, 30, 281, 289] and consists of the conventional FLC and an acceptable number of experiments with bi-linear strain history. The concept is based on the idea that all points with an equivalent pre-strain have the same remaining formability at the defined post-forming direction. This concept allows an easy evaluation of the formability for a multi-step deformation history with different changes of forming direction. This concept is currently implemented in the AutoForm FE commercial code.

By combining the stress-based forming limit curve (σFLC) and the maximum shear stress (MSS) criterion, Stoughton and Yoon [282] proposed a new criterion for failure prediction. This criterion is able to take the stress distribution through the thickness of the sheet into account. Based on the ideas of Yoshida et al. [290] and Zeng et al. [291], Stoughton and Yoon [31] proposed a new concept, the so-called “polar diagram of the effective plastic strain- PEPS diagram”. This diagram is a polar diagram with effective plastic strain EPS as the radial variable and the arc tangent of the ratio of the principal plastic strain rates as angle. The graphical illustration of the PEPS diagram for a complex strain path is presented in Figure 10. The advantages of this new representation of the limit strains are: its insignificant path dependence, its independence on the stress-strain relationship, and its similar shape with the strain FLC [32]. The PEPS method has been successfully implemented into user material interfaces for PAMSTAMP and LS-DYNA 3D commercial codes [292].

Fig. 10
figure 10

Graphical illustration of the PEPS diagram for a complex strain path [31]

Both stress-based FLC (σFLC) and the polar diagram of the effective plastic strain (PEPS) are very efficient in the FE simulation code because the path dependency of SFLCs is much less significant than that of strain-based FLCs. Stoughton [293] present the methodology to convert non-linear strain paths FLC into a single stress-based forming limit curve (σFLC) (see Fig. 11 [125]).

Fig. 11
figure 11

From a path-dependent FLD to a non-path dependent FLSD [125]

Advanced methods used to determine the limit strains

Since the proposal of the FLC concept, many researchers have been actively involved in the development of experimental methods for the accurate and objective determination of the limit strains. These experimental aspects have been the most important obstacles limiting the practical use of the FLC’s. During the last years, the digital processing of the images has allowed the development of refined methods for the experimental measurement of the limit strains. These methods aim to remove the subjective perturbations induced by the human operator in the process of image analysis. More precisely, new algorithms for the detection of the defect occurrence on the deformed part have been developed. They have contributed to the increased accuracy of the limit strain measurement and to the reduction of the discrepancy between the experimental data obtained in different laboratories. In the following, we shall present some of these methods. Further details related to this problem can be found in [32, 92, 120, 273,274,275,276,277,278,279,280].

The first methods used for determining limit strains were proposed by Takashina [294] (the so-called “three circle method”), Veerman [295], Bragard [296] and d’Hayer [297] (the so-called “double profile method”), Hecker [298], Kobayashi [299], IDDRG Group [300] (known as the “Zürich Nr.5 method”). A review of these methods can be found in references [92, 273].

Together with the development “of online” video strain measuring methods, new methods of determining the limit strains have been proposed in the last years. A new criterion based on the evolution of the strain rate as a function of time during the forming process has been proposed by the SOLLAC team [301]. The method is based on the observation that the beginning of the necking is accompanied by a considerable increase of the strain rate (see Fig. 12a).

Fig. 12
figure 12

Methods to determine the limit strains: a Strain-rate versus time; b IDDRG method

According to this method the start of necking corresponds to the dramatic change in the strain-rate versus time variation (characteristic point). This point could be determined by intersecting the straight lines corresponding to the first and the last sector of the curve. The strain – rate evolution is automatically determined by image analysis. The strain-rate method has been intensively studied over the past decade to improve the accuracy of detecting the limit strains and to increase the robustness of the method Volk [302,303,304], Merklein [34, 35], Wang [38], Hashemi [305], Dicecco [39], Lumelskyj [306, 307]. Situ et al. [308] proposed a method based on the major strain acceleration criterion. Volk and Hora [36] proposed an improved method based on recording the strain rate history with the aim of enhancing the capability of detecting the necking initiation and, by consequence, the accuracy of determining the limit strains. The improvement consists in applying a three-stage procedure: The necking region is identified in the first stage (using a 2D evaluation of the strain distribution); the necking initiation is identified in the second stage (by analyzing the strain-rate versus time diagram); finally, in the third stage, the limit strains are evaluated. The method proposed by Volk and Hora is physically founded and ensures an enhanced robustness of the limit strain measurement.

The Nakajima workgroup of the IDDRG developed a new procedure [309], the so-called “in-process measurement” method (see Fig. 12b). A guideline for the FLC determination based on this method is presented in [310, 311]. The method is similar to that proposed by Bragard (see [296]). Using a video camera system, a film of the forming process is made. Based on the film of the forming process, the development of the strain distribution starting from the onset of necking and ending with the fracture occurrence is analyzed. The method is very robust and leads to a good repeatability of the results. Based on this achievement, the Nakajima workgroup proposed a revision of the ISO 12004-2 standard “Metallic materials-sheet and strip-Determination of the forming limit curves” [32]. The formability of a wide range of materials was assessed using traditional Nakajima FLD testing at different labs and compared with the results obtained using the analysis method in the revised ISO/DIS 12004–2 standard, for position dependent testing after fracture [312]. Leppin et al. [313] proposed a method to correct the effect of non-proportional strain paths on the Nakajima test and Hogström at al [314] optimized the ISO method. A comparison of different algorithms used to detect the necking onset was made by Hotz et al. [34]. The limitations of these methods are that they are static, do not describe the development of the strains, and are not applicable to materials characterized by a heterogeneous evolution of the thickness strain (for example in the case of small-radius bending). Vysochinskiy et al. [40] proposed thickness-control method, a robust method for the case of multiple local necks in the sheet. According to this method, the onset of local necking occurs when the local thinning exceeds the allowable tolerance. Dicecco et al. [315] proposed a curvature-based necking criterion to determine the limit strains for aluminium alloy at elevated temperature.

Vallellano and his co-workers [37, 316, 317] developed a method to determine the limit strain based on the strain rate variations near the necking zone. A schematic of the method proposed by Vallellano et al. [37] is presented in Fig. 13. The method has been successfully used later by Zahedi et al. [318]. Recently, Iquilio et al. [319] improved the method proposed by Vallellano and proposed a novel experimental method to determine the limit strain by capturing the heterogeneity of the strain distribution at the local necking site. The limit strains are determined when a non-homogeneous decrease in the thickness necking area begins.

Fig. 13
figure 13

Schematic of the time dependent methodology [37]

Min et al. [320] propose an original method to compensate the process depend effects, based on the measured surface geometry of the test specimen using digital image correlation (DIC) techniques. This method has been improved by Min et al. [321, 322] by using a change in the surface curvature in order to detect a geometric effect associated with the onset of localized necking. Figure 14 [321] presents the curvature evolution during the necking and the change of surface curvature CAB between the points A and B. The method captures the effects of curvature, non-linear strain path and contact pressure. This enables compensating for these effects and reconciling the forming limits determined using the methods of Marciniak (in-plane test) and Nakazima (out of plane test). By using this method, both tests (Marciniak and Nakazima) give identical limit curves.

Fig. 14
figure 14

Cross-sectional views showing the evolution of surface curvature during localized necking. a end of diffuse necking, b initiation and c expansion of a local dimple, d illustration of the change of surface curvature [321]

Elangovan et al. [323, 324] used the Artificial Neural Network (ANN) model to predict the Forming Limit Curves. Abspoel et al. [226] proposed a new semi-empirical method for predicting Forming Limit Curves from mechanical properties namely, total elongation, anisotropy coefficients and sheet thickness.

New tests used to determine the FLC

Banabic et al. [41] proposed a new procedure for the experimental determination of the FLCs. The methodology is based on the hydraulic bulging of a double specimen (Fig. 15).

Fig. 15
figure 15

Schematic view of the new test [41]

The most important advantages of the method are the capability of investigating the whole strain range specific to the sheet metal forming processes, simplicity of the equipment, reduction of the parasitic effects induced by the friction, as well as the occurrence of the necking in the polar region. The comparison between the FLCs determined using the new procedure and the Nakazima test shows minor differences. Figure 16 compares the FLCs obtained using the methodology proposed by Banabic and the Nakazima test (according to the specifications of the international standard ISO 12004-2) [41]. The test was performed by Mitukiewicz [325] and Karadogan [326].

Fig. 16
figure 16

FLD of the AA6016-T4 alloy [41]

Another research direction in this field was oriented toward the use of biaxial traction tests performed on cruciform specimens. This type of mechanical test was proposed at the beginning of 1950’s as a method for determining the biaxial yield stress [327]. The development of the biaxial traction test became more intense at the end of 1990’s (see Section Experimental validation of the anisotropic models) of the paper. Yu at al [328] designed and optimized cruciform specimens to be used for FLC determination. The NIST team designed cruciform specimens and built a biaxial traction machine [329]. During the last decade, this type of mechanical test has been used and developed by Letoing et al. [42, 175, 176, 178, 196, 197], Merklein et al. [35], Shao et al. [330, 331]. In this test the strain path is very easily controlled by the displacements in the two main directions of the cruciform specimen. The main advantages of the test are the flexibility, the fact that it is an in-plane test which is not perturbed by friction.

Influence of different factors on the limit strains

Influence of bending

A comparative study between the Nakazima and Marciniak tests for aluminum killed steel, brass and cold rolled aluminum [332] showed a clear overestimation of the limit strains obtained from Nakazima tests. It is however too early to conclude from this study that sheet curvature during forming is the reason for the overestimation. Charpentier [333] reported the influence of punch curvature on the stretching limits of sheet steel. Several researchers have investigated the influence of the bending radius on the limit strains during the last decade. The results obtained by the groups coordinated by Vallellano [37, 316, 334,335,336,337,338], van den Boogaard [339,340,341], Stoughton [320, 322], as well as by other groups [155, 342,343,344,345], etc. are relevant from this point of view. Figure 17 illustrates the influence of the bending radius on the limit strains [336].

Fig. 17
figure 17

Experimental and predicted forming limit strains versus t0/R [336]

Influence of temperature

During the last years, the intense utilization of advanced high strength steels (AHSS), aluminum alloys, magnesium alloys, and titanium alloys has determined an active investigation of the temperature influence on the limit strains. Significant results in this domain were reported by a number of researchers: [346, 347] with reference to different steel sorts; papers published by the groups of Dean and Lin ([44, 330, 331, 348,349,350] etc.), Bariani and Bruschi ([125, 351]), Leotoing and Guines ([43, 352]), Worswick et al. ([315, 353]), Wang et al. [45], etc., as well as some groups from China [354,355,356], etc. with reference to aluminum alloys. The influence of the temperature on the formability of magnesium alloys was investigated by Sorgente et al. [357], Ghiotti [358], Zhang et al. [359], Huang et al. [360], Ambrogio et al. [361], Boba et al. [362], Berge et al. [363], Koh et al. [364], Kim et al. [365], Stutz et al. [366], Mekonen [367] etc. Yang et al. [368], Bodunrin [369], etc. studied the influence of the temperature on the formability of titanium alloys. Further details on this topic can be found in the synthesis papers [125, 350, 370,371,372]. Figure 18 [44] illustrates the influence of the temperature on the formability of an aluminum alloy.

Fig. 18
figure 18

Predicted versus experimental FLCs for AA2060 at different temperatures [44]

Influence of strain rate

The strain-rate effects should consider both room temperature and elevated temperature. Drewes and Martini [373], Ayres and Wenner [374] and Percy [375] initiated the research concerning the influence of the strain rate on the limit strain at room temperature. In the last decade this influence has been analyzed especially at elevated temperature by Chu [43], Gao [44], Ding [352], Koh et al. [364], Li [376], Yamashida [377], etc. In general, the authors reported that with the increase of the strain rate the FLCs move downward (which means the formability decreases), both for steels and aluminum alloys. The influence of the strain-rate on the FLC for an aluminum alloy deformed at elevated temperature is presented in Fig. 19 [44]. In this case the influence of the strain-rate on the FLC is significant due to the high strain-rate sensitivity of the tested material at warm. At room temperature, the strain rate effect is important for the high speed forming processes (for example, electromagnetic forming [378]. Increases in formability for several materials have been reported in high-speed forming processs ([379,380,381]). Jacques [382] extended the Marciniak model to include the contribution of inertia. For this purpose, it was introduced in the model the Bridgman correction factor [70] in order to account for triaxiality effects.

Fig. 19
figure 19

Predicted versus experimental FLCs for AA2060 for different speeds [44]

Influence of strain path

Several researchers Nakazima [383], Müschenborn [288], Graf [384] have noticed that the strain path has a strong influence on the shape and position of FLC. This topic was recently investigated by Volk [29, 30, 281, 289] who developed a theoretical model based on Müschenborn and Sonne [288] proposal for calculating the limit strains along complex strain paths [281]. Stoughton and his coworkers [31, 32, 292], also contributed to this domain by developing an original method for determining FLC’s along complex strain paths [31]. Contributions on the determinations of forming limit curves for complex strain paths were also reported by Gao [44], Erfanian [51], Leotoing [175], Rojek et al. [385], El Fakir [386], Mattiasson [387], Suttner [388], Nurcheshmeh [389], Wang [355],, etc.

Advances in the FLC prediction

Various theoretical models have been developed for the calculation of forming limit curves (FLC). The first ones were proposed by Swift [390] and Hill [391] assuming homogeneous sheet metals (the so-called models of diffuse necking and localized necking, respectively). The Swift model has been later developed by Hora (the so-called Modified Maximum Force Criterion - MMFC) [392, 393]. Marciniak and Kuczynski (MK model) [394] proposed a model assuming that sheet metals were non-homogeneous from both the geometrical and the microstructural point of view. Barata and Jalinier [395] extended the MK model for non-linear strain path. Stören and Rice [396] developed a model based on the bifurcation theory. Dudzinski and Molinari [397] used the method of linear perturbations for analyzing the strain localization and computing the limit strains. Bressan and Williams [398] introduced the so-called “Through Thickness Shear Instability Criterion” in order to take the shear fracture mode into account. Based on the analysis of the influence of the stress distribution through the thickness on the mode of failure, Stoughton [293, 399] proposed a generalized failure criterion. Since the theoretical models are rather complex and need a profound knowledge of continuum mechanics and mathematics while their results are not always in agreement with experiments, some semi-empirical models have been developed in recent years. The models used for FLC prediction are detailed (formulation of the model, solution methods, numerical aspects, advantages and limitations) in [92].

At present, the most widely used models for the computation of the limit strains are those proposed by Marciniak and Kuczynski [394] and Hora [392, 393], respectively. As a consequence, the models previously mentioned will be briefly discussed in the following (see also [92, 400]). The Forming Limit Stress Diagram (FLSD) proposed by Arrieux et al. [285] has been also intensively studied during the last decade.

During the last decade the research in the field of forming limits prediction has been oriented along the same directions as in the previous decade. As an observation we mention that the research regarding the physical understanding of the phenomena of strain localization using polycrystalline, ductile damage models and the influence of the through-thickness stress have more significant weight. Therefore, we will focus our attention only on these three areas.

Because a book chapter devoted to the description of advanced models used for FLC prediction has been recently published [275] we will give here only a synthetic presentation of the research papers published during the last decade, with an emphasis on the papers published after the issuance of the book chapter mentioned above.

Crystal plasticity based FLC prediction

During the last years, two different approaches have been mainly followed to couple crystal plasticity calculations to sheet formability prediction, namely the MK framework and FE modelling.

The adaptability of the texture-based models to the MK theory of the strain localization has been proved in the 1980’s by Bassani [401], Bate [402], Asaro [403], Barlat [8, 101] and later by van Houtte [404] and Neale [405] teams. The early studies often adopt the Taylor-Bishop-Hill plasticity model. During the last decade, Signorelli’s team [47, 48, 406,407,408] has showed that FLDs predicted using coupled models of VPSC and M–K method (MK-VPSC) [110] are able to provide a better agreement with the measured FLDs as compared with the Taylor based M–K models (see Fig. 20 [47]).

Fig. 20
figure 20

Experimental and predicted FLDs for the DP-780 steel samples taken along the RD [47]

Apart from the above models, the full-field approaches, namely crystal plasticity–finite element (CPFEM) have been used by Inal’s group to predict the limit strains [409, 410]. Recently, this group has used the rate tangent based crystal plasticity–fast Fourier transform model (RTCP-FFT) [411]. This model, which may account for the 3D microstructure, grain interactions, and intragranular texture evolution, has been implemented recently in the M–K framework [46] to improve the accuracy of FLD predictions as compared with Taylor models (a homogenous deformation is assumed throughout the material) and VPSC.

Ductile damage and fracture models

It is well accepted that in metals, damage and ultimately failure is the result of nucleation, growth, and coalescence of voids (e.g. [412]). Since voids grow due to the plastic deformation of the fully-dense material, early on it has been recognized that the rate of void growth can be deduced once the plastic potential of the porous solid is known. One of the most widely used models are those derived by Gurson [413, 414] by conducting limit-analysis of a hollow sphere or hollow cylinder obeying von Mises yield criterion. It was proven that the presence of voids in a von Mises matrix induces dependence of the plastic flow on the mean stress, or hydrostatic pressure. To better account for void evolution, various modifications of Gurson’s criteria have been proposed (e.g. [415,416,417], etc.), the most used being the Gurson-Tvergaard-Needleman (GTN) model. Recently, efforts have been devoted towards describing the role played by the particularities of the yielding behavior on void evolution. Cazacu et al. [418, 419] demonstrated that if the yielding behavior of the void-free material is described by either von Mises, Tresca, or Drucker criterion (see Eq. (1)), the dependence of the plastic potential and ultimately the fracture locus on the stress invariants cannot be arbitrary Specifically, it was mathematically proven that with the exception of purely hydrostatic and purely deviatoric loadings (triaxiality zero), there is dependence on J3and moreover there exists a very specific coupling between the signs of the pressure and that of the third invariant J3such that:

$$ \mathrm{F}\left({\sigma}_m,{J}_2,{J}_3\right)=\mathrm{F}\left(-{\sigma}_m,{J}_2,-{J}_3\right) $$
(19)

Furthermore, it was shown that for positive triaxialities the rate of void growth is faster for loadings at J3>0 (e.g. uniaxial tension) than for loadings at J3 ≤ 0 (e.g. equibiaxial tension); the reverse holds true in the case of void closure (i.e. loadings corresponding to negative triaxialities). It was also shown that the rate of void growth is much faster in a Tresca material than in a von Mises material. Recently Cazacu and Stewart [420] used limit-analysis to derive the plastic potential for isotropic materials with matrix displaying tension-compression asymmetry. This model predicts that irrespective of loading there is an effect of the sign of the mean stress while the effect of J3 is very strong for any loadings except hydrostatic tension or compression. In the absence of porosity, this model reduces to the isotropic Cazacu et al. [111] criterion. For materials with k > 0 (matrix softer in compression than in tension), the rate of void growth is much faster than for materials characterized by k < 0 (matrix harder in compression than in tension).

In the case of a textured material, it is extremely difficult or impossible to obtain the plastic potential of the damaged material in closed-form. Benzerga and Besson [421] developed such a potential for the case when the matrix is described by the Hill [88] while Stewart and Cazacu [422] developed an orthotropic plastic potential for the case when the matrix is described by the orthotropic Cazacu et al. [111] criterion .

With the availability of High-Resolution X-ray Computed Tomography (XRCT), in the past decades it has become possible to quantify the porosity in the bulk of a given material. Specifically, void nucleation and growth can be clearly visualized during continuous or interrupted in-situ tests. Recently, Maire and collaborators used this technique to measure porosity evolution in several commercial aluminum alloys. For example, for AA 6016-T4 tensile test results on both smooth and notched samples of the material as well as damage measurements by XRCT for several orientations in the plane of the sheet were reported in [423]. It was shown that the observed rate of void growth is well predicted using the Cazacu and Stewart [422] model for all notch acuities (see Revil et al. [424]. For XCRT data on 2000, 5000 and 7000 series aluminum alloys, the reader is referred to Maire et al. [425].

In the case of hcp materials, every aspect of damage and its evolution are completely different than in materials with very little tension-compression asymmetry. As an example, Fig. 21a shows the reconstructed 2-D views in the (TD, TT) plane, and in the (RD, TD) plane of a smooth RD specimen of a strongly textured Ti material, respectively, which was subjected to uniaxial tension very close to failure. For comparison purposes, in Fig. 21b are shown the respective views extracted from an XCMT scan done for a copper specimen, scan which was taken at the same level of axial displacement (see Revil et al. [142]). Note that for the same axial displacement, the Cu material is much damaged than the Ti material shows. Specifically, for the Ti specimen the average porosity over the specimen volume and minimal cross-section is of 0.052% and 0.25%, respectively.

Fig. 21
figure 21

Post-test XCMT scans of smooth specimens subjected to uniaxial tension: a Ti specimen; b Cu specimen (after [142]). (a)Anisotropic model with k = 0 (b) Stewart and Cazacu [422] model

In situ scans on a notched RD specimen subjected to uniaxial tension were also reported in [142]. It was observed that damage initiates at the outer surface of the specimen and further grows from the outer surface to the center of the specimen. Stewart and Cazacu [422] model predicts these trends as well as the fact that damage accumulates differently in the TD and ND directions of the cross section (see [136, 142] for comparisons between model and data). To evaluate the importance of the consideration of the tension-compression asymmetry on the damage predictions, these authors also performed simulations with k = 0, as shown in Fig. 22a. Note that in this case, the maximum porosity is at the center of the specimen. Only if the tension-compression asymmetry in plastic deformation is accounted for, the experimental porosity distribution is captured (see Fig. 22b).

Fig. 22
figure 22

Importance of accounting for tension-compression asymmetry: comparison of the F.E. isocontours of porosity in the (RD-TD) of a notched Ti specimen subjected to uniaxial tension along RD for an axial displacement of 1.2 mm: a predictions according to anisotropic model with no tension-compression asymmetry (k = 0); b Stewart and Cazacu [422] model (after [142])

For examples of other damage models and approaches for modeling damage in metallic materials the reader is referred to the monographs [79, 426].

The GTN damage model has been used by many researchers interested in the prediction of FLC’s. The papers published by Zahedi [318], Hu [346], Lin [427], Malcher [428], Yang [429], Tang [430], etc. can be mentioned here. Recently, the CERTETA team used the Gurson’s model with some recent extensions to model porous materials, following both the evolution of a homogeneous sheet and the evolution of the distribution of voids [431]. The Gurson-Tvergaard-Needleman (GTN) model has been used by Kami et al. [49, 432] to determine the FLC of AA6016-T4 aluminum alloy. Figure 23 shows that the results obtained by numerical simulation using the GTN model are in good agreement with experimental data. The comparison becomes even more favorable when confronted with the predictions of the Marciniak-Kuczynski (M-K) model and the modified maximum force criterion (MMFC) (see Fig. 23) [432].

Fig. 23
figure 23

Comparison between the FLC calculated with different methods and experimental data [432]

Fracture laws that expressed by relations between the fracture strain and the stress triaxiality have been frequently used during the last decade. Furthermore, Wierzbicki and his group introduced a new concept to define the formability based on the dependence of the equivalent strain at fracture on the stress triaxiality [433,434,435,436]. This has been intensively studied in the last decade by the groups of Wierzbicki [437,438,439,440,441,442] anh Huh and Yoon [443,444,445,446] and also by Li [447], Lou [448] etc. A review of the most popular laws used for the prediction of fracture forming limit curves was conducted by Habibi [344]. Lou proposed in 2012 [443] a ductile fracture condition extended in 2014 [444] and 2017 [448]. The fracture locus for plane stress proposed by Lou is presented in Fig. 24 [444]. A fracture polygon based on the maximum shear stress (MSS) fracture criterion has been proposed by Stoughton and Yoon [282] and used to predict the necking limit. To describe yielding and fracture of pressure-sensitive metallic materials [449] introduced in the expression of the orthotropic yield function of Nixon et al. [135] an additional hydrostatic pressure term.

Fig. 24
figure 24

Fracture locus for plane stress in the in the space ε1-εh determined by using Lou criterion [444]

Effect of the normal pressure on the formability

The effect of the normal pressure on the formability of sheet metals is well known and has been exploited in industry for a long time [450]. Banabic and Soare [451], Wu et al. [452] and Allwood and Shouler [453] analyzed independently the influence of the normal pressure on the Forming Limit Curve using an enhanced MK model. The effect of the through-thickness normal stress on the forming limit diagram has been studied extensively in the last 5 years especially by Iranian and Chinese groups: Assempour et al. [50], Liu [454], Zhang et al. [54, 455, 456], Nurcheshmeh and Green [457], Lang et al. [458], Hashemi [52, 459, 460], Mirfalah [461, 462], Ma [53, 463], Bettaieb [464], Shi [465], Erfanian [51], Hu [466] etc. The influence of the hydro-static pressure on the yield and flow of metallic materials was also investigated by Soare and Barlat in [467]. The study of the influence of through-thickness shear on the limit strains was initiated by Eyckens et al. [468] and continued in the last decade by Eyckens [469], Fatemi [470], Mirfalah [461, 462], etc.

Besides the above-mentioned domains, systematic investigations have been recently performed along the following directions: formability of multi-layer (sandwich) sheets ([55,56,57,58,59,60,61, 471,472,473], etc.), extension of the Modified Maximum Force Criterion-MMFC model [61, 63, 260, 392, 393, 474,475,476] etc.), formability of sheet metals at micro-meso scale [477], as well as revisiting the perturbation approach ([64, 65, 478, 479]), etc.

Concluding remarks and future directions

In this paper we have presented and discussed recent advances in modeling the yielding and hardening behavior, and damage of metallic materials, innovative experiments aimed at providing verification/validation of these models, and illustrative examples and applications to metal forming. Further discussion, concluding remarks, and future directions are presented in the following.

At present, there are 3-D yield functions (or strain rate potentials, e.g. see [79, 100],) for single crystals, and polycrystalline FCC, BCC or HCP materials that take the distinctive features of the mechanical behavior of these materials into account (see also [136]). For instance, for HCP materials there exist pressure-insensitive yielding formulations that account for the difference in yielding behavior in tension and compression observed experimentally for these materials. Due to a correct modelling of material symmetries, anisotropic yield functions available nowadays are more versatile than those used in the past. Examples of the capabilities of representative plane stress or general three-dimensional formulations were presented.

Any isotropic or anisotropic yield condition, including any of those described in this article, can be used as a plastic potential though the associated flow rule (AFR) in order to determine the plastic strain increment. Many authors have used the AFR, also called normality rule, based on the strain hardening stability expressed in the Drucker postulate [480]. Critical experimental investigations as reviewed by Hecker [481] could not find any clear violation of the rule. Nevertheless, the non-associated flow rule (NAFR) made a resurgence during the last two decades and its applications to sheet metal forming simulations have been increasing because the NAFR allows the use of simpler functional forms with improved calibration accuracy [482]. In this respect, the NAFR is interesting for industrial applications where simpler approaches are preferred. Moreover, the work of Stoughton and Yoon [483] on the development of stability constraints has also been essential for allowing a wider application of the NAFR. In fact, the recent version of the homogeneous anisotropic hardening model HAH20 is non-associated in the direction of the hydrostatic stress to preserve plastic incompressibility but associated in the deviatoric hyperplane as implied in [159, 484] and demonstrated in Soare and Barlat [467]. This is because, for metals, the AFR is supported by theoretical arguments based on crystal plasticity set forth by Bishop and Hill [82] with a minimum number of assumptions. Although AFR models might appear more complex, numerical optimization methods are available through commercial software to calibrate constitutive models with high accuracy. Moreover, the use of numerical derivation techniques such as in [485, 486], which do not require the analytical derivation of the flow rule from the potential, are very accurate and greatly simplify the finite element implementation of the AFR-based models.

In the future, the evolution of yield function coefficients will be self-contained in the formulation for any linear and non-linear loading. Therefore, as in the case of crystal plasticity, it will be possible to describe the evolution of anisotropy as deformation proceeds. However, the computation time will be much lower than what is at present required for calculations when crystal plasticity models are directly linked to FE codes.

In conventional multiaxial stress tests many linear stress paths are applied to test samples to determine the contours of the plastic work and the directions of the plastic strain rates. Proper material models (anisotropic yield functions) are identified using the experimental data, assuming that the normality flow rule (NFR) associated with the work contours is satisfied. When the contours of plastic work are similar with the increase of plastic work an isotropic hardening (IH) model is applied to the material while if it is not the case a differential hardening (DH) model should be used. To the author’s knowledge this procedure of material modeling is effective to enhance the predictive accuracy of sheet metal forming simulations, although IH is assumed in most industrial forming simulation.

In reality, however, most materials exhibit DH [12, 13, 15, 16, 183, 184, 187, 189, 195, 198, 199, 247, 267]. Moreover, some materials do not follow NFR with respect to work contours [16, 221]. To quantitatively evaluate the DH of sheet metals and the validity of NFR for a lager strain range up to immediately before a localized neck occurs, MTET is a useful testing method. In particular, it is capable of applying arbitrary nonlinear stress/strain paths to a specimen, which sheet metals very often undergo in industrial forming operations. The main issue of MTET is that it requires a special testing apparatus that can simultaneously apply internal pressure and axial force to a tubular specimen using a feedback control system. Even preparing a tubular specimen could be problematic for high strength materials and materials with poor weldability. Therefore, the development of a systematic and user-friendly multiaxial stress testing method is crucial to the identification and the development of highly-accurate material models. The VFM integrated into a DIC platform will be a powerful experimental tool for the establishment of the testing method.

Regarding the formability of metallic materials, in the past, the FLC models provided an approximate description of the experimental results. Such models were used especially for obtaining qualitative information concerning the necking/tearing phenomena. At present, the FLC models allow a sufficiently accurate prediction of the limit strains, but each model suffers from its own limitations (see Section Advanced methods used to determine the limit strains). There is no model that can be applied to any sort of sheet metal, any type of crystallographic structure, any strain-path or any variation range of the process parameters (strain rate, temperature, pressure, etc.).

During the last decade, the experimental research has been mainly focused on refining the experimental methods used for detecting and measuring the limit strains, developing new laboratory tests for FLC determination, as well as on investigating the influence of various material and process parameters on the limit strains. The theoretical research has been oriented toward improving the predictive capabilities of various models by refining the numerical methods used to solve the models and including new material or process parameters in the models (temperature, strain rate, pressure, shear stresses, strain path effect, bending effect, etc.). An intense research activity can be noticed in the field of investigating the effect of the normal pressure and through-thickness shear on the formability. The research on ductile fracture and crystal plasticity models has gained an increased weight during the last decade. These models have contributed to the clarification of some phenomena taking place at micro-meso scale directly linked to the necking and fracture processes, thus ensuring an enhanced accuracy in the FLC prediction. The industrial utilization of sandwich materials has also intensified the research regarding their formability.

The future research will be focused on a more in-depth analysis of the phenomena accompanying the necking and fracture of the sheet metals. On the basis of the analysis, more realistic models will be developed in order to obtain better predictions of the limit strains. New models will be developed for prediction of the limit strains for special sheet metal forming processes: superplastic forming, forming at very high pressure, incremental forming etc. Commercial codes allowing the quick and accurate calculation of the FLC’s both for linear and complex strain-paths will be developed. The texture models will be also implemented in such commercial programs. The FLC computation will be included in the finite element codes used for the simulation of the sheet metal forming processes. The aim is to develop automatic decision tools (based on artificial intelligence methods) useful in the technological design departments. The stochastic modeling of the FLC’s will be developed in order to increase the robustness of the sheet metal forming simulation programs. Comparing different FLC detection methods for the same material is a usefull research topic. More refined, accurate and objective experimental methods for the experimental determination of the limit strains (e.g. methods based on thermal, acoustic effects, X-ray tomography) will be also developed.