3.

0 Material Modeling Guidelines Many but not all material models in LS-DYNA are able to include the effect of strain rate. It should be obvious that any materials that naturally exhibit strain rate sensitivity in the range of expected strain rates should be modeled to include that sensitivity, i.e., a material model (*MAT) that has the capability to capture strain rate effects should be utilized and material parameters and/or curves that control the strain rate sensitivity should be established via experimental data or by reliable and applicable data taken from the literature. For example, rate-sensitive metals are often modeled using *MAT_PIECEWISE_LINEAR_PLASTICITY in which the parameter LCSS refers to a table (*DEFINE_TABLE, *DEFINE_CURVE) that defines a family of stress vs. effective plastic strain curves, each curve corresponding to a particular effective strain rate. Curves used to describe any constitutive data, e.g., stress vs. strain or force vs. deflection, should be very smooth with no abrupt changes in slope. The range of data given along the abscissa should also be close to the working range of the material and not far beyond. To define an unnecessarily broad range of data will generally result in a loss of resolution in the range of interest (due to automatic rediscretization of the curve done for the sake of computational efficiency).

3.1 Elasto-Plastic Materials Elasto-plastic materials include metals and polymers. The most common material model used in modeling such materials is *MAT_024, which offers multi-linear strainstress behavior (isotropic), isotropic hardening, strain rate effects, and a plastic-strainbased failure criterion. Many other material models are available in LS-DYNA for modeling plasticity with anisotropic behavior, finite elastic strain, temperature sensitivity, damage models, different yield surface in tension and compression, etc. When strain rate effects are included in the material model (recommended), turn on the viscoplastic rate formlation by setting VP to 1. In doing so, the effective plastic strain rate is used in evaluating material behavior, as opposed to the often noisy total effective strain rate. Unless specifically noted otherwise in the LS-DYNA Users Manual, stress vs. strain curves required as input for plasticity material models should be expressed in terms of true uniaxial stress and true plastic strain which, in the uniaxial case, are equivalent to von Mises stress and effective plastic strain, respectively. Be aware that experimental data always includes some degree of error and, thus, tends to be somewhat noisy or erratic. When using *MAT_024, one should input a smooth stress-strain curve utilizing a minimal number of points.

The effective plastic strain values input in defining a stress vs. effective plastic strain curve in a LS-DYNA plasticity model should be the residual true strains after unloading elastically. Using experimental data from a true stress vs. true strain curve, the effective plastic strain input is calculated as effective plastic strain (input value) = total true strain - true stress/E Note that as the stress value increases, the recoverable strain (true stress/E) increases as well. For metals, E is very large compared to the yield stress so it is fairly common practice in the case of metals to just subtract off a constant value equal to the strain at initial yield from all subsequent strain values. For plastics/polymers, consideration should be given to the increase in recoverable strain as stresses increase (since the elastic component of strain may be quite large). In any case, the first plastic strain value should be input as zero and the first stress value should be the initial yield stress.

3.2 Low Density Foams Material scientists characterize foam as any material manufactured by some expansion process. For simulation purposes, any material with a Poisson coefficient close to zero can be characterized as a foam. Both definitions apply to low density foams with densities below approximately 200g/l. High density structural foams with densities above 200 g/l are not foams in the numerical sense, since they exhibit a non-negligible Poissons effect. Foams can be classified as elastic or crushable foams, with typical stress-strain responses shown in Figure 3.1. Crushable foams exhibit permanent deformation and are used in padding and some insulation. Seat foams, bumper foams, and some insulating foams are elastic. Seat foams are elastic, but can also be further differentiated as soft foams where dynamic test results are dependent on the size and shape of the sample as shown in Figure 3.2. (This is due to the soft foams open cell structure and air outflow.) Therefore, the size of soft foam test samples should approximate that of the part of interest.

Figure 3.1a Elastic Foam Response.

Figure 3.1b Crushable Foam Response.

Sample A

Sample A

f f' << ' A A

Figure 3.2 Soft Foam Sample Size Test Dependence. The size dependence problem does not occur in most insulation, padding or bumper foams. This is due their full or partly closed foam cell structure and to the much higher structural strength of these materials compared to seat foams. Crushable foams can be modeled using the following material models: *MAT_26 *MAT_63 *MAT_75 *MAT_126 *MAT_142 *MAT_163 *MAT_HONEYCOMB *MAT_CRUSHABLE_FOAM *MAT_BILKHU/DUBOIS_FOAM *MAT_MODIFIED_HONEYCOMB *MAT_TRANSVERSELY_ANISOTROPIC_CRUSHABLE_FOAM *MAT_MODIFIED_CRUSHABLE_FOAM

Elastic foams can be modeled using the following material models: *MAT_57 *MAT_73 *MAT_83 *MAT_179 *MAT_181 *MAT_LOW_DENSITY_FOAM *MAT_LOW_DENSITY_VISCOUS_FOAM *MAT_FU_CHANG_FOAM *MAT_LOW_DENSITY_SYNTHETIC_FOAM *MAT_SIMPLIFIED_RUBBER/FOAM

Most of these material models can make use of curve or table (for multiple strain rates) input of stress strain curves to define the foams behavior. All foams are to some

degree rate dependent and so preference should be given to material models which can include rate effects. Material models in which the rate effects are explicitly defined via tabulated input, such as *MAT_83 and *MAT_181, allow the user greater control without the requirement of curve fitting (see Figure 3.3). Because of this, *MAT_83 is the most frequently material model for the simulation of elastic foams.

& min

Figure 3.3 Tabulated Stress Strain Input Representing Strain Rate Dependent Behavior. *MAT_83 has rate independent unloading and is potentially unrealistic for foam with high damping. Figure 3.4 shows a comparison of the behaviors of *MAT_83 and *MAT_73 (a visco-elastic foam model) in a relaxation test. *MAT_73 is potentially a better choice for a foam with high damping. *MAT_181 allows for the definition of a (small), non-zero Poissons ratio which may be required for some applications. (Note that the signs of the *DEFINE_TABLE input for compression and tension for *MAT_83 and *MAT_181 are opposite of each other.)

d Relaxation Test t *MAT_83 *MAT_73

Figure 3.4 Rate-dependent Elasticity Versus Visco-Elasticity. *MAT_83 is numerically stable when the tabulated curves are correctly prepared. The only known remaining numerical problem occurs when the foam is impacted by a very sharp knife-like object that cuts deeply into the foam part. Setting SFLAG=1 so that

engineering strain rate is evaluated and RFLAG=1 so that the strain rates are evaluated in each principal direction aids realistic behavior and numerical stability in unloading. Set TFLAG=1 in order to indicate that both compressive and tensile response are defined in the load curves. Also set SRAF=0 to ensure adequate smoothing of the principal strain rate values during the computation. Since version 971 of LS-DYNA, rate dependent unloading can be used in conjunction with *MAT_83. The easiest way to do this consists of defining the parameters SHAPE and HU in the material cards and excluding the unloading curve from the table input (discussed in the next section). The fraction of the energy that is dissipated during hysteretic unloading is given by (1-HU) so for a seat foam HU is around 0.85 (little dissipation) and for a padding foam HU is around 0.15 (high dissipation). Generation of Engineering Stress Strain Curves for Table Input The basis for foam table input is quasi-static and dynamic uniaxial compression test data. This data must be repeatable, have relatively smooth stress strain curves, and contain the densification phase of the foam loading. The exact density of the test samples should be known. Figure 3.5 is sample dynamic test data of a bumper foam with a density of 53 g/L.
25
1-DD3-2 1-DD3-3 1-DD3-4

20

Compression Stress / MPa

15

10

0 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0

Compression

Figure 3.5 Dynamic Compression Test Data of an Elastic Foam. The test data must be aggressively smoothed so that no local variations, such as those visible in Figure 3.5, remain. This may be accomplished by averaging each data point with its adjacent 3 or 5 measurements. Smoothing by direct manipulation is also sometimes required. The smoothed curves should be plotted against the test data to insure that drift away from the test data has not taken place. In addition, each curve should contain no more than 100 points, equidistant along the abscissa. All of the quasistatic and dynamic curves should then be plotted together, as shown in Figure 3.6, and any intersections eliminated. These intersections are likely in the densification phase.

Figure 3.6 Smoothed Stress Strain Curves with a Close-up Check for Intersections. Each of the curves should then be differentiated to determine where their slope stops increasing in the densification phase. At the strain of the maximum derivative of both the quasi-static and dynamic curves, all stress strain curves are cut off. In the example shown in Figure 3.7, the derivative of the strain quits increasing at 92%, and all curves should be cut off at that strain level.

Figure 3.7 Differentiated Foam Stress Strain Curve to Determine Cut Off. The measurements from the test data in this example terminate at 20 MPa. Much higher stresses are likely to occur locally in numerical simulations. Therefore, extrapolation of the smoothed stress strain curves is necessary. A hyperbolic function of order n,

n+1
where n is defined as,

=n +

1 1 1 1 n

; n > 1

 2 1 ln ; > 1 n= 2 1 1 1 ln 1 2
and the definitions of 1 and 2 are shown in Figure 3.8

1 2 1 Figure 3.8 Definition of 1 and 2 .

are used. The stresses should be extrapolated to 200 MPa or to a stress value of the same order as the yield stress of supporting or interacting structure. Two stress strain data points are selected on the quasi-static curve where the derivatives of the stresses are increasing. For the curve shown in Figure 3.6, values of 82% and 91% strain were selected and resulting extrapolation exponent, n, was 3.570977. The results of that extrapolation are shown in Figure 3.9. Note that there is a continuous transition into the extrapolated section of the curve.

Figure 3.9 Extrapolated Quasi-Static Elastic Foam Stress Strain Curve (Two Scales). The dynamic stress strain curves are extrapolated in the same manner. The steepest dynamic curve should be used to calculate the extrapolation exponent and all the other dynamic curves should extrapolated using the same exponent to avoid intersections.

The extrapolated curves should then be plotted with each other to insure that there are no intersections as shown in Figure 3.10.

Figure 3.10 Extrapolated Elastic Foam Stress Strain Curves with No Intersections. One additional curve is defined using the unloading results from the quasi-static test data. In LS-DYNA the unloading follows the curve with the lowest strain rate. As a result, this curve is assigned the strain rate of zero in the *DEFINE_TABLE input. There is no rate dependency upon unloading. The remaining curves are assigned the appropriate strain rates and the LS-DYNA *DEFINE_TABLE input can be created from the stress strain curves represented on the *DEFINE_CURVE input. A constant strain interval of 1% varying from 0% to 99% resulting in 100 equidistant values should be used. Tension data may also be added and input on the same *DEFINE_CURVE cards. This data should also be prepared using smoothing and input using a constant strain interval of 1%, with intersections of the curves only allowed at the origin. Extrapolation is not required and failure of the foam may be defined. The most physically realistic way to define foam failure is to define failure only in tension using the *MAT_ADD_EROSION card. For *MAT_83, tensile data are represented by the negative part of the load curves using of the load curves for and can be replaced by a constant E-modulus if TFLAG is set to 0. For *MAT_181, tensile data are represented by the positive part of the load curves.

3.3 Thermoplastics Thermoplastics are types of polymers which are finding increasing applications in the aerospace industry. In addition to traditional uses within cabin interiors, usage in load bearing components is growing. However, a standard theory governing the mechanical behavior of thermoplastic materials has yet to be developed. In addition, some of the key parameters which govern behavior have yet to be fully determined. As a result, the current methods for modeling thermoplastics tend to be pragmatic and minimalistic, and treat them like soft steels using *MAT_024 (*MAT_PIECEWISE_LINEAR_PLASTICITY). A more sophisticated model, *MAT_187

(*MAT_SAMP-1, Semi-Analytical Model for Polymers) is also now available, which enables a closer match to actual thermoplastic behavior. Both approaches will be outlined in this section. Thermoplastics differ from metals in several key attributes. Among these are, that in addition to the yield surface, the Youngs modulus exhibits strain rate sensitivity; elasticity can be non-linear; necking is not localized; yielding differs between compression, tension, and shear (Figure 3.11); incompressible in compression (the Poissons ratio approaches .5); and, under tension the Poissons ratio approaches 0. As a result, Von-Mises behavior is not exhibited in thermoplastics. Despite this, using elastoplastic approaches are currently still the best choice for modeling thermoplastics because they provide stable, user friendly input, solutions which can successfully predict permanent deformation.

Figure 3.11 Comparison of Compression and Tension Material Response.

3.3.1 Using *MAT_024 *MAT_PIECEWISE_LINEAR_PLASTICITY is the current standard choice for modeling thermoplastics. It can be successfully used to predict the maximum deformation, but cannot be used to predict thermoplastic failure. In addition, it does not correctly capture the strain dependent unloading of thermoplastics. The usage of and preparation of input for *MAT_024 follows the standard process for material property table input. Beginning with engineering stress vs. engineering strain test data, curves are first converted to true stress vs. true strain. Multiple curves which represent response at different strain rates are required (Figure 3.12). An average Youngs modulus of all the strain rates is determined (Figure 3.13). (Multiple curves at the same strain rate can either be averaged, or a representative curve can be selected.)

Figure 3.12 Experimental Stress-Strain Data at Differing Strain Rates.

Figure 3.13 Determination of Average Youngs Modulus.

Each curve is then smoothed, creating a series of curves with no intersections, each with a maximum of 100 data points. If there is an intersection, then the equation = *e (- *), where * = (*), can be used when > * to eliminate the intersection. The onset of necking gives that value for * and is identified by finding the strain where (Figure 3.14).

d =0 d

The hardening curves are computed by using the relationships

y = , p =

(Figure 3.15). E

These curves are then formatted as

*DEFINE_TABLE input for use in the *MAT_024 material model definition.

Figure 3.14 Smoothed and Extrapolated Stress-Strain Curves.

Figure 3.15 Yield Stress versus Plastic Strain.

To verify that the *MAT_024 material model definition is adequate; the tension test which was used to create the data is simulated and compared. The correlation must be exact before the onset of necking. If it is not, the extrapolation used to avoid intersections, = * e(
*

, should be adjusted. For example the equation,

with the fitting parameters a and b may be used, and varied until a sufficiently accurate correlation is achieved (Figure 3.16).

= e

* a( * )b

Figure 3.16 Tensile Test Simulation Compared to Test Data.

3.3.2 Using *MAT_187 For typical thermoplastics, the tension yield point is less than the compression yield point. As a result, when a material model is based solely upon uniaxial tension tests, it results in too soft of a response under both bending and compression loading. Such a models yield point can be scaled to match expected loading, if that mode of loading is well understood, but such a model is not robust. In addition, the crazing failure of thermoplastics is not simulated successfully by an isochoric, effective plastic strain to failure criteria. Thermoplastics can be modeled more realistically by the use of *MAT_187, *MAT_SAMP-1, Semi-Analytical Model for Polymers, in which these characteristics can be represented. In addition, the strain dependent unloading of thermoplastics can also be modeling in*MAT_187 by the use of a damage model (see the LS-DYNA Users Manual). The yield surfaces of the thermoplastic being modeled in *MAT_187 are defined by separate compression, shear, and bi-axial tension stress-strain curves, as well as a tension table as described in the previous section for use with *MAT_024. Clearly, the use of this material model required a significant amount of test data to be utilized fully. The yield surface is the quadratic function fitted through values defined by the four load curves, as shown on the pressure versus Von-Mises stress (triaxiality) plot in Figure 3.17.

vm
shear tension shear

t
biaxial tension

biaxial tension

tension

compression

s c

f =0 3 1
compression

t
0 p
2 f ( p, vm , pl ) = vm A0 A1 p A2 p 2 0

Figure 3.17 Yield Surface Definition of *MAT_SAMP-1.

The yield surfaces at initial yielding can be fitted accurately using this approach. For example, comparisons of experimental data and the SAMP-1 curve fit are shown in the following three figures. The comparison for Polyvinyl Chloride (PVC) is shown in Figure 3.18, for Polystyrene (PS) is shown in Figure 3.19, and for Acrylonitrile Butadiene Styrene (ABS) is shown in Figure 3.20.

Figure 3.18 Yield Surface of PVC Compared to SAMP-1 Curve Fit.

Figure 3.19 Yield Surface of PS Compared to SAMP-1 Curve Fit.

Figure 3.20 Yield Surface of ABS Compared to SAMP-1 Curve Fit.

The process of converting engineering stress and strain is somewhat more complicated for compression than it is for tension. For tension, true strain is found by = ln (1 + 0 ) and true stress is found by = 0 (1 + 0 ) . In compression, the true strain is independent of Poissons ratio, so = ln (1 + 0 ) , but the true stress is not independent. If the Poissons ratio is constant then = 0 (1 + 0 ) , and if it is not constant, then the true stress must be solved for incrementally using the equation
2

&0 )2 ( ) dt . For simple shear the volume and the area cross section are = 0 (1 + d constant, and so = 0 = , = 0 = 0 , = 0 . l0 2
0

The identification of necking and extrapolation of the stress curve beyond necking also differs slightly from the method used for *MAT_024 data preparation. The strain at which the onset occurs is found by 2

d = 0 for compression and tension, and for d d shear the strain at necking onset is determined by = 0 . The stresses extrapolated d ( ) 2 beyond necking are defined as = * e for compression and tension, and = * for shear. The hardening curves for compression and tension are the same as those used for *MAT_024, y = , p = , and for shear are E y = , p = . 2G
*

3.4 Ice Ice is highly variable material with many different forms, crystal structures, and strengths. One of the strongest forms, defect free columnar or single crystal, can form under certain conditions on aircraft surfaces. As a result, in all safety analyses the properties worst case/highest strength ice should be used. However, at ballistic impact speeds approaching 1000 feet per second and greater, the mass of the ice becomes the primary determining factor in how much damage it causes (as opposed to its strength.) Ice is a linear elastic material which fails in a brittle fashion at strain rates of interest in aerospace problems. Upon impact, a stress wave will travel through the ice reducing its strength and the level of force it imparts to the impacted structure after an initial high peak. However, the mass of the reduced strength ice still imparts loads onto the impacted structure. Ice is much stronger in compression than it is in tension (approximately a factor of ten). The strength of ice also shows a significant amount of strain rate sensitivity in compression. A material model, *MAT_155, *PLASTICITY_COMPRESSION_TENSION_EOS, has been added to LS-DYNA to specifically incorporate all of the required features of modeling ice [3]. Ice material models which do not incorporate these features (some models incorporate a tuned level of plasticity) should be used with caution, as deviating from the conditions of the tuning would necessarily call into question the validity of the predictions. Table 3.1 provides material properties for single crystal ice at -10C.

Table 3.1 Sample Properties for Single Crystal Ice at -10C. E = 9.31 GPa = 1.35 E6 psi Compression Strength = 14.79 MPa = 2.147 E3 psi Tension Strength = 1. 3 MPa = 188 psi Density = 897.6 kg/m3 = 8.4 E-5 lbm/in3 Possions Ratio = .33 Strain Rate Strengthening Factor at 1000 sec-1 = 1.77

Because ice can continue to apply load as a projectile after it partially fails and after it deforms to a point where a Lagrangian mesh can significantly distort and tangle, representing the ice with a Eulerian mesh is an appropriate choice. Again, as in the case of a tuned plasticity material model, Lagrangian meshes have been used to represent ice in the past, but care must be taken to insure that mass is not removed prematurely, when actual ice would still be applying load to the structure.

3.5 Composites For sandwich or laminate composites modeled with mat_022, mat_054, or mat_055, laminated shell theory can be invoked by setting LAMSHT to 1 in *CONTROL_SHELL (recommended, particularly if material constants vary through the thickness of the shell through use of *INTEGRATION_SHELL or *PART_COMPOSITE). If each layer of the composite shell is comprised of an isotropic, elasto-plastic material, then *MAT_114 should be used since this material has built-in laminated shell theory. Invariant node numbering is particularly recommended for anisotropic composites. Simply set INN in *CONTROL_ACCURACY to 2 if there are composite shells, 3 if there are composite solids, or 4 if there are both composite shells and solids. To insure that stresses and strains are output in the orthotropic material coordinate system, set CMPFLG=1 in *DATABASE_EXTENT_BINARY. In this manner, x-stress will be fiber stress and y-stress will be matrix stress. Please also refer to the tutorials section of the web page mat_compos.pdf for an overview of composites modeling. Additional notes and examples relating to composite modeling in LS-DYNA are also available at ftp://ftp.lstc.com/outgoing/jday2.

3.5.1 Composite Delamination

Ply separation is caused by the bonding failure between plies due to the normal and shear stresses in composite laminate. The crack is assumed to run at the seam of two plies, although it often splits one ply near the fibers and may even jump from one ply to another. Fracture mechanics based Cohesive Zone Modeling (CZM) approach suits well to study the delamination. All the methods in LS-DYNA have fracture mechanics based separation law. (The well known and liked virtual crack closure technique (VCCT) is not implemented in LSDYNA). The user has to have the fracture toughness and some other material data for both Mode I and Mode II. These may not be readily available. The dynamic fracture toughness is measured using standard methods, but those are not included in any ASTM specifications. LS-DYNA offers several ways to model delamination, among them are: *CONTACT_TIEBREAK option 7, DYCOSS [4] *CONTACT_TIEBREAK.option 9, equivalent to *MAT_138 *CONTACT_TIEBREAK option 11 allows courser mesh [5] *MAT_COHESIVE_MIXED_MODE (138)8-node solid element with cohesive material *MAT_COHESIVE_ELASTIC (184)..8-node solid element with cohesive material *MAT_COHESIVE_GENERAL (186)8-node solid element with cohesive material As seen from the above list, there are two basic way to model delamination in LSDYNA, (1) use solid elements to explicitly model the bonding material between the plies; or (2) use the contact-tiebreak definition between the plies. With *MAT_138, *MAT_85, and *MAT_186, 8-node solids can model finite thickness of the bonding layer, or the solid elements can also represent zero thickness layers (the bottom 4 nodes are co-incident with the top 4 nodes). However, when the bonding layer is thin so that there is no need to consider the mass of the bonding layer, we prefer to use *CONTACT_TIEBREAK with option 7, 9, or 11. These options have the traction separation laws built into the contact definition. This reduces the input data significantly, and makes the modeling simpler. This is the case with most aerospace composites. *CONTACT_TIEBREAK option 6 with failure stress is also sometimes used for delamination. The separation occurs as soon as the contact stresses reach the user specified values. This option is a crude approximation and should be used only in desperation when better data is not available.

3.5.1.1 Cohesive Zone

The cohesive zone (CZ) is the area in front of the crack tip, shaded in Fig 3.21, where the bonding material undergoes deformation, and at some point the damage starts developing and ultimately the two plies are fully separated. The figure shows only separation due to tension i.e. in the direction normal to the plies or Mode I, but the relative displacement in the two tangent (shear, or Mode II) directions will cause separation as well. Compression does not cause separation. The energy dissipated in separating the plies is called the fracture energy and it is the most important variable which has to be computed correctly in the finite element analysis.

Figure 3.21 Cohesive Zone.

For typical graphite-epoxy composite materials, the length of cohesive zone is from one to two millimeters, or even less. Various authors give the cohesive zone length as G l cz = ME c 2

( )
0

where, M is a multiplier ranging from 0.21 to 1.0 depending on the author, E is the material elastic modulus, and 0 is the peak traction/stress which develops in the cohesive zone (point 2 in Fig 3.22). This formula is useful in estimating the CZ length. Numerical studies have suggested that at least three elements are needed to span the cohesive zone to adequately capture the fracture energy [6,7]. In other words, the elements size in the direction of the crack growth would have to be no more than 0.5 mm. This obviously puts a severe limitation to the use of the cohesive zone method in any practical engineering analysis. However, a practical solution to this problem exists and is presented in Section 3.5.1.4.

3.5.1.2 Constitutive Model The traction-separation law between the plies describes the cohesion zone behavior and the energy release in the separation process. The linear elastic/linear softening (bilinear) model is the simplest and also the most commonly used. Figure 3.22 shows

the bilinear constitutive model in tension (Mode I) [8]. The bottom diagram shows the stress-strain assumption with key points. The top diagram shows the corresponding points in the delamination progression. *MAT_COHESIVE_MIXED_MODE (138) and *CONTACT_AUTOMATIC SURFACE_TO_SURFACE_TIEBREAK options 7, 9, and 11 use the bilinear constitutive model [4, 5, 6, 9]. With option=11, _ONE_WAY SURFACE_TO_SURFACE_TIEBREAK is recommended [10]. It has been reported that with really high speed applications, the sharp corners in the bilinear law may induce numerical instability (ringing) due to sudden damage growth at Point 2 and sudden failure at Point 4. More advanced and complex separation laws are available in LS-DYNA. *MAT_COHESIVE_GENERAL (186) allows the users to define a load curve for a general shape separation law. Some authors have rounded the sharp corners with a cubic curve while keeping the total fracture energy (area under the curve) correct. *MAT_COHESIVE_ELASTIC (184) and *MAT_COHESIVE_TH (185) are also available for special applications of delamination, but the problem with all these advanced models is the lack is reliable data and reliable user experience. (Obviously it takes an advanced user to be able to determine what that shape is, by testing or other means).

Figure 3.22 Bilinear Constitutive Model.

In Figure 3.22, Point 1 is in the elastic part of the material response. Material has not suffered any damage and the unloading at this point would follow the elastic line. Point 2 represents the onset of damage and material softening (the damage growth) begins. Once the loading has progress to the Point 3 the material has suffered some

damage (damage parameter is greater than zero, but less than one), but the plies have not separated yet. If unloading happens at this point, it is assumed to follow the straight line from Point 3 to Point 0. The shaded area (Points 0, 2, 3) represents the energy dissipated to partial damage of the bonding and is not recoverable. At point 4 the plies separate permanently (damage parameter has reached unity). The total area under the triangle (Points 0, 2 ,4) represents the energy it takes to delaminate two plies and is known as the fracture energy. In LS-DYNA the fracture energy is the input parameter or fracture toughness or the energy release rate, G. It has units of energy/area. In addition, the elastic stiffness (slope) and the peak stress (Point 2) are required for complete definition of the bilinear law. Numerical studies have shown that the fracture toughness has to be accurate, but the initial stiffness and the peak stress do not need to be accurate, i.e. they can be changed without affecting the overall results. Camanho and Davila [8] use a constant value 10E+6 for all materials and call it penalty stiffness. Then, in order to keep the fracture toughness (area under the triangle) correct, the peak stress has to be adjusted accordingly.

3.5.1.3 Mixed Modes The above description of the constitutive law is shown for tension loading and ply separation in the normal direction. For general behavior, the relative normal and tangent displacements are computed and the law is extended for the 3-dimensional deformation, i.e. the Modes I and II are mixed for the 3-dimensional analysis. The relative displacements between the two plies are computed ( 1 , 2 , 3 ). Then the two tangent displacements are combined into one shear displacement:

II = 12 + 22
As stated earlier, the Mode I is governed by the normal displacement

I = 3
The total mixed mode relative displacement between the plies is then found as
2 m = I2 + II

and, the mode mixitity is then defined as

II I

The mixed mode concepts are illustrated in Figure 3.23. The T and S are the peak tractions in tension and shear, respectively. The shaded triangle shows the damage initiation, growth, and separation under general 3-dimeansional loading. The damage initiation, 0 and the mode mixitity, can be determined.

traction
S

I0
0 II

separation
IF
I

0
1

F II

F m

II

Figure 3.23 Mixed Mode Traction-Displacement Interaction.

The mixed mode delamination propagation (running crack) is most commonly predicted in terms of the power law

GI G IC

G II + G IIC

= 1.0

The subscripts I and II refer to the normal and shear, as before, and the subscript C refers to critical. The critical values are input to LS-DYNA. The ratio GI/GIC is the ration of the shaded triangle to the whole triangle in Figure 3.23. The exponent alpha is usually set to 2 (especially when the user does not have data to do otherwise) and, occasionally, 1. Messieurs Benzeggagh and Kenane have devised the so called B-K interaction law, which captures the dependence of fracture toughness as a function of the mode ratio better in epoxy composites. (Again the user must be aware and have accurate data). Both the power law and B-K law are available in LSDYNA.

3.5.1.4 Practical Solution with Coarse Mesh

In Section 3.5.1.1 we pointed out that the cohesive zone is about 1 to 2 mm long and for accurate modeling, at least three elements are needed in the cohesive zone. This means that the element size would have to be 0.3 to 0.6 mm long in the direction that the crack runs, which would would not be practical. Turon, Davila, Camanho and Costa have developed a technique to allow larger elements without sacrificing the overall accuracy [4]. They have demonstrated that their technique can yield as accurate results with element size 4 mm as those obtained with very fine meshes, element = 0.25 mm. As pointed out in Section 3.5.1.2, the energy release rates G IC and G IIC are critically important for successful analysis, but the interface (penalty) stiffness and the peak tractions for tension (T) and shear (s) are not. They can be varied without affecting the overall results much, as long as the fracture toughness is correct. If the peak traction is lowered (Figure 3.24), the failure strain (Point 4 in Figure 3.22) needs to be increased to keep the area under the curve correct. Consequently, the effective element length increases and fewer elements are needed. This is nothing but a useful trick, but it works because the overall energy balance is kept correct.

Figure 3.24 Lowering Peak Traction Lengthens the Element Effective Length.

Figure 3.25 Double Cantilever Beam (DCB) Mode I analysis: Left Mesh has 0.25 mm Elements and Right Mesh has 4.0 mm Elements.

50

40 0.25 mm 0.50 mm 1.00 mm 20 2.00 mm 3.10 mm 4.00 mm 10

Load [N]

30

0 0 2 4 6 8 10

Displacement [mm]

Figure 3.26 Uncorrected Load-Displacement Plots from DCB analysis with Different Meshes. Figure 3.25 shows two mesh sizes for the double cantilever beam (DCB) analysis. The element size on the left is 0.25 mm and on the right 4.0 mm (this is FE model of the Mode I fracture toughness testing). These plots show that the 0.25 mm mesh is fully converged, i.e. further mesh refinement would not improve the accuracy. Also, the 0.5 mm mesh seems to be adequately converged. On the other hand, the 4.0 mm mesh gives the peak load that is way too high. The reasoning is as follows: The cohesive zone length here is about 1.25 1.50 mm so five to six 0.25 mm elements is enough to model the cohesive zone correctly and about three 0.50 mm elements is enough to model the cohesive zone. The question now becomes: how to scale the peak tractions for a given element size for accurate results? We prefer the trial and error method as outlined by Erhart [5]. If the real loaddisplacement plot is available from tests, this is the correct (converged) solution, and the scaling of the peak traction is adjusted so that the results match the converged solution for selected element sizes (0.25, 0.5, 1.0, 2.0, 3.0, 4.0 mm). The correct scaling is found by trial and error. If the test plot is not available, we can use the equation in Section 3.5.1.1 to estimate the cohesive zone length and refine the progressively mesh till it is converged. This plot and the peak load then become the correct (converged) solution. Then we find the scaling factor by trial and error for the desired element sizes to match the converged solution. The normalized element size vs. scaling factors for both Mode I and Mode II are input as a load curves to LSDYNA. A typical load curve scaling factor plot is shown in Fig. 3.27.

The results are now independent of the element size as seen in Fig. 3.28.
1 scaling factor 0,8 0,6 0,4 0,2 0 0 1 2 3 4 element size [mm]

Figure 3.27 Scaling factor vs. the Element size for the Peak Traction.
30 25 20

Load [N]

15 10 5 0 0 2 4 6 8 10

Displacement [mm]
Figure 3.28 Corrected Load-displacement Plots from DCB Analysis with Different Meshes.

3.6 Material-Test Interaction Each material model, whether associated with the projectile or target (i.e. blade or containment structure) should be based on mechanical properties obtained by

mechanical property tests. The basic static properties of compression, tension, and shear should be measured in test. In the case of composites, or other non-isotropic materials, these properties must be obtained in each appropriate direction. Many, if not most materials used in aerospace applications show significant strain rate sensitivity, and, therefore, this behavior must also be determined in tests. The classification of the material into a general category, when dealing with an exotic material, is sometimes required, and, in this event, even basic properties, such as modulus and Poissons Ratio cannot be assumed, but must be obtained by test. A material model must also be selected that can match the basic physical behavior and properties of the actual material.

3.6.1 Model Creation If an increase in material strength due to rate effects is noted, then it should be modeled explicitly as strain rate sensitivity, using an appropriate model. (The static material properties should not be uniformly matched to the properties at high strain rate.) Most of the structure will not be undergoing high strain plastic deformation, and, therefore, to model the material with the static material properties representing only the high strain rate behavior will significantly over specify the strength of the material. After a provisional material model is created, a mesh of each test configuration that was used to obtain the mechanical properties should be constructed. General meshing guidelines are discussed in Section 2.6, but a mesh density sufficient to produce a uniform stress field in the elastic, pre-yield state is required. (Post-yield stress field and failure convergence is discussed in Section 3.7.) A careful comparison of the test and analysis results is required to insure that the physical material behavior is represented by the material model. At the same time, a good test-analysis comparison of the preyield stress state does not guarantee that accurate failure representation of the material will occur. The failure, and/or element erosion criteria, for most material models are usually based on some type of effective plastic strain at failure. The material model failure plastic strain is obtained by modeling the test configuration that was created to confirm the basic physical behavior of the material. The value of the failure plastic strain is mesh size dependent and, unlike elastic stress analysis, does not converge as the mesh size is reduced (see discussion in Section 3.7). The value of effective plastic strain can be adjusted to match the failure behavior of the mechanical property test. However, this means that the mesh size used to model a part of interest cannot vary significantly from that used to calibrate the material model with the test configuration. There are currently efforts underway to generalize and expand the element removal and failure criteria (i.e., mesh regularization such as used in *MAT_NONLOCAL and *MAT_TABULATED_JOHNSON_COOK). These technologies require additional test data,

which is currently not available for most alloys. As a result, the above failure definition procedure is still required for most analyses. The input definition for *MAT_TABULATED_JOHNSON_COOK requires a series of tests to determine failure in differing configurations, the description of which is beyond the scope of this section. With most materials, there exists a significant variation in mechanical properties that can be caused by processing details, such as heat treatment, grain orientation, or inherent randomness, which can also require mechanical property testing to be material, alloy, and potentially batch specific.

3.6.2 Model Validation and Verification A series of sub-scale ballistic tests, or representative sub-scale component tests, using the primary materials in the complete system analysis should be conducted to validate and verify that the basic material models, the contact algorithm, and the other many aspects of an explicit analysis are correctly defined. The elastic deflections, the plastic deflections, the boundary loads, the containment threshold velocity, and the final velocity of the projectile are some of the variables which can be measured and used to determine if the analysis is reliable. As much as possible, the basic parameters of the material model should not be changed on the basis of the results of the ballistic tests, especially when these changes would contradict the properties obtained in the mechanical property tests. If the test and analysis do not match, then the basic assumptions that went into the material model creation should be questioned, and additional mechanical property testing may be required. If additional work is required for the analysis to match the tests, then parameters to which there is some uncertainty should be the values that should be tuned. Ideally, no tuning should be required for the analysis to match the test, but, unfortunately, this practice can sometimes not be avoided. If tuning is used, then the analysis is necessarily only valid for conditions that are close to those tested and correlated to. The resulting correlated models should not be used for extrapolation. Full scale tests of systems are usually required for final verifications. Ideally the analysis model predication should match the test with no additional modifications to the analysis. However, most actual structures and systems are so complicated that some assumptions and compromises as to which components need to be modeled in detail are often made. If a system level test demonstrates that some component modeling needs to be enhanced, then care should be taken to insure that this enhancement is made using the same procedures for geometry, mesh, and material modeling as was used previously and shortcuts not be taken. For example, if a particular sub-structure that was assumed to be rigid is found to be dynamically active, then a full verification and validation of the mesh and material models will be required.

For analysis of containment problems, the represented structures are often simple and verification of the mesh is usually straight forward. However, as more challenging analyses with more complicated structures and meshes are undertaken, more methodical mesh verifications are required. An example of a system requiring more thorough verification is the analytical model predicting system response due to fan blade loss. In problems such as these, all available data should be leveraged to insure accuracy and completeness of the discretization of the structure. Examples of information which should be used to verify this type of structural system problems include various tests and data. The mass properties of the system model should be compared and correlated with the mass properties of the tested structure. Deflections and/or strains from system static tests, such pull tests on a full engine, should be compared to model predictions in order to verify stiffness. Dynamic properties should be checked with a vibration survey, or correlated to a modal survey, when available. In general, the verification of the non-linear fan blade loss model requires, at least, the same level of verification as the linear loads model.

3.7 Material Model Failure Calibration There is an inherent mesh dependency of the local strain value after localization or necking of a specimen occurs. This is demonstrated by looking at the following models. For a two element beam representation, the local strain after localization is

lo
l 2 l 0 2 = ln l0

l
For a four element beam representation, the local strain after localization is

lo

3l 4 l 0 4 = ln l0

For a beam with n number of elements, the local strain after localization is

lo

(n 1)l0 n l n = ln l0
l
Defining characteristic element length as lc = approximate local failure strain is

l0 and total length l = l0 + d , then the n

f ,local

(n 1)l0 n l d (l l0 ) + lc n = ln = ln = ln + 1 l0 lc lc

and the global failure strain is

f ,global = ln

d l = ln + 1 l l0 0

Plotting the ratio of the local failure strain to the global failure strain as shown in Figure 3.29 demonstrates mesh size dependency, non-convergence, and that a finer mesh size leads to a higher value of failure strain.

f ,local f , global

lc l0

Figure 3.29 Mesh Dependency on the Local Failure Strain.

The most common approach for the representation of metals in LS-DYNA are the plasticity models such as *MAT_024 or *MAT_PIECEWISE_LINEAR_PLASTICITY and *MAT_081 or MAT_PLASTICITY_WITH_DAMAGE, etc. In these models material failure, or erosion, is generally controlled by a user input effective plastic strain to failure. Correct use of the plastic failure strain parameter which can accurately be used to predict the threshold velocity for fan blade containment, or the ballistic limit of other impact analyses requires calibration with ballistic test data. The suggested approach is applicable to all material models which have element erosion controlled by this parameter. Despite the recent progress and development in the use of mesh regularization such as in *MAT_NONLOCAL and *MAT_TABULATED_JOHNSON_COOK, the calibration approach will continued to be required for many metals and alloys in which the significant test data requirements of the alternate approaches are not available. The effective plastic strain at which a metal ruptures is dependent on the type of loading. For example, the value of strain in a shear failure is typically lower than that in tension failure, and, in hydro-static compression, metals do not fail at all. Other factors that can influence the failure strain include temperature, strain rate, and geometry. As a result, the use of a failure strain from a standard tension dog bone test would only coincidently be the correct value to use for a ballistic impact test simulation, where the loading is at minimum bi-axial, but more typically tri-axial. Therefore, a value for effective plastic strain is required that is specific for the class of loading under consideration. An additional complication in determining the correct effective plastic failure strain to use in a simulation is that this value is element size dependent. This inherent mesh dependency begins after localization of the yielding (necking in a simple tension test). A finer mesh leads to a higher value of failure strain in a corresponding, correlated failure analysis. Even when a simple tension test is modeled with successfully finer meshes, convergence to a failure strain value found in handbooks is not to be expected as is shown in Figure. As a result, the correct value of failure strain is specific to the element size being used, as previously discussed.

Tensile Elongation = 6.8% Calibrated Value = 4%

Tensile Elongation = 7% Calibrated Value = 14%

Tensile Elongation = 20% Calibrated Value = 8%

V= 182 fps

V= 196 fps

V= 247 fps

V= 236 fps

V= 220 fps

V= 416 fps

Figure 3.30 Calibrated Failure Value is Not Scalable to Tensile Elongation. In order to determine the correct effective plastic strain to failure for a fan containment analysis, and for a specific element size, a set of ballistic impact tests are performed. These tests are performed using simple representative configurations which make them relatively simple and inexpensive tests. However, care must be taken in the design of the ballistic test to insure similar conditions and failure modes to the fan containment event under consideration. The conditions which should be appropriately represented or matched include: - Failure mode - Projectile impact angle - Projectile size, shape, and material - Projectile velocity - Target thickness and material - Target orientation The tests are modeled using the same input parameters as those that will be used in the fan containment simulation. The effective plastic strain to failure is simply adjusted so that the ballistic limit of the analysis matches that of the simple ballistic test results. The general parameters of the analysis which must be identical in the simple ballistic and the fan containment analyses include element size, element type, contact algorithm, and, of course, material model and corresponding failure strain. If the general parameters of the fan containment must be changed, then the calibration must be

repeated using the modified general parameters. These general parameters should be selected to represent the actual physics of the problem up to the point of failure as accurately as possible. The material model used should also be appropriate with strain rate effects and accurate stress-strain behavior included. In other words, adjusting the effective plastic strain to failure to compensate for other modeling inadequacies should be avoided. For example, a mesh size should be selected in which the elastic portion of the problem has converged. As an example, a cylindrical shell with a flange section was impacted with a 1 inch diameter steel ball with the resulting failure threshold velocity of 293 fps. The cylindrical shell and flange was modeled using four different mesh sizes as shown in Figure 3.31. For Meshes #1 and #2, correlation was not successful as the failure occurred in the flange, not the cylindrical section (Figure 3.32). The smaller element sizes of Meshes #3 and #4 were required for successful calibration as is shown in Figure 3.33 and Figure 3.34. The calibrated effective plastic strains to failure resulting from these examples also demonstrate both mesh sensitivity and the usefulness of the calibration approach.

Increasing Mesh Density

Mesh #1 689 nodes, 355 elements, coarse mesh in cylindrical portion and finer mesh in flange

Mesh #2 1421 nodes, 810 elements, more consistent mesh throughout

Mesh #3 4826 nodes, 2940 elements, consistent mesh throughout Mesh #4 14737 nodes, 9540 elements, consistent fine mesh throughout

Figure 3.31 Four Different Mesh Densities.

Failure Strain = 22%

Failure Strain = 5%

Failure Strain = 2%

Figure 3.32 Coarse Mesh Prevents Successful Calibration.

V= 174 fps

V= 221 fps

V= 293 fps

Figure 3.33 Mesh #3 with Calibrated Failure Strain of 15.4%.

V= 174 fps

V= 221 fps

V= 293 fps

Figure 3.34 Mesh #4 with Calibrated Failure Strain of 22.0%.