ARTICLE IN PRESS

Journal of Biomechanics 37 (2004) 127134

Design and numerical implementation of a 3-D non-linear viscoelastic constitutive model for brain tissue during impact
D.W.A. Brandsa,*, G.W.M. Petersb, P.H.M. Bovendeerdb
a

Department of Mechanical Engineering, Division of Computational and Experimental Mechanics, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands b Department of Biomedical Engineering, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands

Abstract Finite Element (FE) head models are often used to understand mechanical response of the head and its contents during impact loading in the head. Current FE models do not account for non-linear viscoelastic material behavior of brain tissue. We developed a new non-linear viscoelastic material model for brain tissue and implemented it in an explicit FE code. To obtain sufcient numerical accuracy for modeling the nearly incompressible brain tissue, deviatoric and volumetric stress contributions are separated. Deviatoric stress is modeled in a non-linear viscoelastic differential form. Volumetric behavior is assumed linearly elastic. Linear viscoelastic material parameters were derived from published data on oscillatory experiments, and from ultrasonic experiments. Additionally, non-linear parameters were derived from stress relaxation (SR) experiments at shear strains up to 20%. The model was tested by simulating the transient phase in the SR experiments not used in parameter determination (strains up to 20%, strain rates up to 8 s1 ). Both time- and strain-dependent behavior were predicted accurately R2 > 0:96 for strain and strain rates applied. However, the stress was overestimated systematically by approximately 31% independent of strain(rate) applied. This is probably caused by limitations of the experimental data at hand. r 2003 Elsevier Ltd. All rights reserved.
Keywords: Finite element modeling; Impact biomechanics; Non-linear viscoelastic constitutive model; Brain modelling

1. Introduction Traumatic brain injury (TBI) caused by a mechanical insult on the head, for example during trafc accidents, sport accidents or falls, causes high mortality and disability (Brooks et al., 1997; Viano et al., 1997; Waxweiler et al., 1995). TBI occurs when the local mechanical load, exerted on the brain tissue, exceeds certain tolerance levels. Understanding how an external mechanical load on a head is transferred to a local mechanical load in the brain is needed to improve injury protecting devices and diagnostic methods. To obtain this understanding, nite element (FE) modeling is often used (e.g. in Bandak and Eppinger, 1994; Claessens et al., 1997; Turquier et al., 1996; Zhang et al., 2001). Current FE head models contain a detailed geometrical description of the intracranial contents but lack an accurate description of brain material behavior.
*Corresponding author. Tel.: +31-40-247-3135; fax: +31-40-2447355. E-mail address: d.w.a.brands@tue.nl (D.W.A. Brands). 0021-9290/$ - see front matter r 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0021-9290(03)00243-4

At strain and strain rate levels associated with TBI (approximately 20% (Bain and Meaney, 2000; Galbraith et al., 1993; Schreiber et al., 1997) and typically 20 s1 ), brain tissue behaves as a non-linear, viscoelastic material (Bilston et al., 2001; Estes and McElhaney, 1970; Peters et al., 1997). In stress relaxation (SR) experiments shear softening occurs, i.e. the stiffness decreases as strain increases (Arbogast et al., 1995; Bilston et al., 2001; Brands et al., 2000; Prange et al., 2000, 2002) while Darvish and Crandall (2001) found that shear hardening occurred when increasing strains in oscillatory experiments at frequencies exceeding 44 Hz; indicating full nonlinear material behavior. The bulk modulus of brain tissue is about 106 times higher than the shear modulus (Etoh et al., 1994; Goldman and Hueter, 1956) indicating nearly incompressible material behavior. This provides the following requirements for a material model for use in FE modeling brain tissue during impacts: accurate replication of the non-linear viscoelastic behavior in shear-like deformations for strain(rate)s up to 20% 20 s1 and special precautions for accurate modeling of the nearly incompressible behavior.

ARTICLE IN PRESS
128 D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134

In this paper the development and FE implementation of a new non-linear viscoelastic constitutive model for modeling brain tissue is presented. The model is written in a differential formulation as opposed to the QLV formulations currently implemented in explicit FE packages. Furthermore, the formulation is such that the nearly incompressible material behavior is modeled with sufcient numerical accuracy. Material parameters are determined from two sets of shear measurements with porcine brain tissue on a rotational viscometer and from ultrasonic experiment data. The model and its implementation are tested using a three-dimensional FE simulation of transient shear experiments.

For later use in the constitutive model, we introduce the, recoverable, elastic Finger tensor Be ; and its invariants I1;2;3 : Be F e F c e; I1 traceB e ;
2 2 I2 1 2traceB e traceB e ; I3 detB e :

To separate changes in volume from changes in shape, % e and we introduce the isochoric elastic Finger tensor B its associated invariants as
1=3 % e I3 B Be ;

% e ; I %1;2 I1;2 B %3 1: I

2. Methods 2.1. Kinematics The stress in an arbitrary solid material is determined by changes of volume and shape, described by the deformation gradient tensor F : Deformation gradient tensor F is split multiplicatively into an elastic part, F e ; and an inelastic part, F p : d~ x F d~ x0 ; F F e F p; 1

The description of the rate of deformation is based on the velocity gradient tensor, L which, is decomposed additively into an elastic part, Le and an inelastic part, Lp ; F 1 ; L Le Lp ; LF e F 1 ; Le F e L D W; p F 1 F 1 : Lp F e F p e
T D1 2L L ; T W 1 2L L

Both parts, Le and Lp ; are decomposed additively as 5 with D; the symmetric rate of deformation tensor (LT denotes the transpose of tensor L) and W ; the skewsymmetric spin tensor. To obtain a unique relaxed stress-free state, Cp ; it is assumed that the inelastic deformation occurs spin-free, We W and W p 0: 6

x represent a material line element in where d~ x0 and d~ undeformed state C0 and deformed state Ct ; respectively (see Fig. 1). The inelastic contribution refers to the deformation (with respect to the undeformed state), of the relaxed stress-free conguration Cp ; which is dened as a ctitious state that would be recovered instantaneously when all loads were removed from the material element.

2.2. The constitutive model For accurate modeling of the nearly incompressible brain tissue, the Cauchy stress, r; is written as the sum of a volumetric part, rv ; which depends on volumetric changes only, and a deviatoric part, rd ; which depend on change of shape only: r r v rd : 7 The volumetric part of the stress, rv ; is assumed linear elastic, implying: p rv K I3 1I 8 with unitary tensor I and bulk modulus, K : The deviatoric part of the stress is modeled non-linear viscoelastic. It is decomposed in a number of viscoelastic modes, rd i: rd
N X i1

F C0

Ct

Fp CP

Fe

Fig. 1. Graphical representation of the multiplicative decomposition of the deformation gradient tensor F : The inelastic part F p of F transforms the undeformed state C0 to a relaxed stress-free conguration, Cp ; which is a ctitious state that would be recovered instantaneously when all loads were removed from the material element. The elastic part F e of F transforms the stress free state, Cp ; to the deformed state Ct : To obtain a unique ctitious stress free state, Cp ; it is assumed that the inelastic deformation occurs spin free i.e. all rotations must be accounted for in the constitutive model governing the elastic part.

rd i:

The number of modes used, N ; is determined by the frequency range for which the model has to be valid. To derive rd i we apply Eq. (1) for each mode separately obtaining F F e;i F p;i : We then dene the strain rate

ARTICLE IN PRESS
D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134 129

dependent behavior, governed by viscosity, Zi ; as Dp;i rd i 2Zi : 10

density r; using: K c2 p r: 15

The elastic behavior is modeled using a strain energy density function (SEDF), Wi : %1;i 3 C01;i I %2;i 3 C20;i I %1;i 32 Wi C10;i I %2;i 32 ; C02;i I 11 % e;i %1;i and I %2;i represent invariants of the tensor B where I (Eq. (3)). The deviatoric part of the Cauchy stress tensor, follows from W : 2 %d % rd i pfC10;i 2C20;i I1;i 3gB e;i I3;i 2 1 d % %2;i 3gB pfC01;i 2C02;i I e;i : I 3 ;i 2.3. Numerical implementation The time evolution of stress and strain in each mode, is described by an evolution law, based on kinematics only. For each mode, the inelastic right CauchyGreen tensor, C p ; is dened as
T 1 C p FT p F p F Be F :

12

Linear parameters C10;i ; C01;i and Zi were determined from data of oscillatory shear experiments or dynamic frequency sweeps (DFS) with porcine brain tissue on a rotational plate-plate viscometer (ARES, Rheometric Scientic, 1993) published previously (Brands et al., 2000). A strain amplitude of 0.01 was applied in a frequency range of 1:6216 Hz at various temperatures and a master curve, valid for frequencies ranging from 1.6 to 684 Hz at 37 C; was constructed using the time temperature superposition principle (TTS) rst applied to brain tissue in (Peters et al., 1997). The results of these experiments were presented in terms of the storage modulus , G 0 and the loss modulus, G00 ; as a function of frequency o: For small shear strains, the non-linear viscoelastic constitutive equations (10) and (12) reduce to a linear multimode Maxwell model for which, G 0 and, G 00 ; are found as: n 2 X l2 io G0 Gi ; 16 2 1 l2 io i 0 G 00
n X i0

Gi

li o 2 1 l2 io

13

Taking the time derivative of C p and using the requirement of spin free inelastic deformation, Eq. (6), provides a evolution equation insensitive for large rigidbody rotation and translation p 2C p F 1 Dp F : C 14

This evolution equation is used for updating C p numerically in the time integration procedure. At the new time increment, the updated elastic Finger tensor 1 T then follows from Be F C p F : Application of B e in constitutive equations (8) and (12) yields updated rd i and rv : Finally, the inelastic rate of deformation Dp is determined from Eq. (10) and serves as basis for proceeding to the next time step using Eq. (14) again. This procedure is implemented in an explicit FE Code typically used for crash impact simulations (madymo , TNO-Automotive, 1999). 2.4. Determination of material parameters The multi-mode model contains one material parameter for the volumetric behavior K ; and ve parameters for each viscoelastic mode C10;i ; C01;i ; C20;i ; C02;i and Zi ). The bulk modulus, K ; is determined from the velocity of dilatational waves, cp ; measured in brain tissue (Etoh et al., 1994; Goldman and Hueter, 1956) and mass

with Gi 2C10;i C01;i the shear modulus and li Zi =Gi the relaxation time of mode i: Note that, only the sum C10;i C01;i can be obtained from the shear experiments. We uniquely determined C10;i and C01;i by assuming them to be equal as in Miller and Chinzei (1997). Next, the parameters describing non-linear behavior, C20;i and C02;i ; were derived from large strain stress relaxation experiments (SR), using the linear parameters determined before. These were performed, with the same sample, within minutes after nishing the DFS at shear strains of 0.05, 0.1 and 0.2. Experiments were nished within 4 h after sacrice. In the SR experiments it was found that, once the strain had obtained a constant value, (within 0:1 s), the stress could be written as a function of the shear strain at the plate edge, gR ; a strain-dependent normalized stiffness (damping function), ha gR ; and the linear relaxation modulus, Gt: ta t; gR G tha gR gR t; where ha g R
t a g R gR t a g R gR

17

limgR -0

18

Before using these data in the present work, they were corrected for the radial inhomogeneity in the strain eld, according to the procedure described in the appendix, to obtain the true damping function htrue gR :

ARTICLE IN PRESS
130 D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134 Table 1 Material parameters obtained from tting the 4-mode non-linear viscoelastic constitutive model on brain tissue DFS and SR shear data (sample G2 in Brands et al. (2000)) Mode i 0 1 2 3 4 MooneyRivlin parameters Viscosity Bulk modulus K (GPa) 2.3

Evaluation of Eq. (12) for simple shear strain provides the following expression for normalized stiffness for each mode: hi g 1 2g2 fnls;i ; fnls;i C02;i C20;i : C01;i C10;i 19

The non-linear shear parameter, fnls;i describes the nonlinearity in mode i: Since htrue gR is independent of time, we assume equal nonlinearity for each mode, i.e. hi g htrue gR providing fnls;i fnls : As for the linear parameters, also for the nonlinear parameters we uniquely determined C20;i and C02;i by assuming them to be equal. 2.5. Test of the constitutive model The model is tested by a three-dimensional FE simulation of the SR experiments on the rotational rheometer including the transient strain onset that has not been used for material parameter determination. In this manner we test the constitutive model and its numerical implementation as well as the validity of the strain correction method applied to the experimental data. The geometry of the brain sample is modeled by a quarter cylinder with diameter 25 mm and thickness 2 mm corresponding to the typical sample size used in the experiments. Spatial discretisation is obtained using 1288 brick elements with linear interpolation functions (Fig. 2). A preliminary convergence study revealed a maximum deviation of 2%. To prevent mesh locking, likely to occur due to the nearly incompressible material behavior, reduced spatial integration is used. Symmetry boundary conditions are applied on the cross-sectional planes. The lower plane is rotated according to the

C10;i C01;i (Pa) C20;i C02;i (Pa) Zi (Pa s) 85.96 67.27 80.66 106.8 824.9 386:0 302:0 362:2 479:5 3704 N 18.9 2.46 0.606 0.0403

experimental data. The upper-plane is rigidly supported. To compare numerical and experimental results, the reaction torque at the upper plate, T ; is determined and the apparent stress, ta t; gR ; is calculated, as in the experiments, using ta t ; gR 2T pR3 20

with, R the radius of the sample. Brain tissue material parameters shown in Table 1 are applied. The time step used is limited by the conditional stability of the explicit central difference time integration scheme used in the code and is set to 1:66 107 s:

3. Results 3.1. Determination of material parameters With cp 155972 m=s (average7range in literature values) and r 1040 kg=m3 Eq. (15) provides a bulk modulus of 2:3 GPa: The small strain DFS experimental data could be tted well over the complete frequency range using four viscoelastic modes with rst-order material parameters shown in Table 1 (see Fig. 3). The error in G 0 equals 2:079:6% (average7standard deviation) and reaches maximum absolute values of approximately 20% at upper and lower end of the frequency spectrum (684 and 1:6 Hz; respectively) . The loss modulus is tted with an average error of 3:574:6%: The error values remain less than 10% and are approximately distributed at random. Application of more modes yielded no improvement to the results while less modes deteriorated the accuracy of the solution. The damping function obtained from the SR data shows that shear softening is more prominent in the normalized stiffness data which has been corrected for strain inhomogeneity than in the raw experimental data (Fig. 4). The shear softening equals 37% at 20% strain. Fitting Eq. (19) to this data with fnls 4:49 provides a relative error less than 2%.

Fig. 2. Graphical representation of the three-dimensional mesh (Top view and side view of cross-section shown). A quarter cylinder is modeled using 1288 brick elements with 1672 nodes. Symmetry boundary conditions are applied on the cross-sectional planes.

ARTICLE IN PRESS
D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134
10
4

131
100 80 Num Exp =0.2
0 0

10 Exp Fit

30 Stress [Pa]

Strain [%]

20

10

G [Pa]

G [Pa]

60 40 20

10

10

=0.1 0=0.05 0.1 0.2 0.3 Time [s] 0.4

10 0 10

10 10 Frequency [Hz]

10

10 0 10

10 10 Frequency [Hz]

10

0 0

0.1

0.2 0.3 Time [s]

0.4

0 0

Fig. 3. Master curves of storage and loss modulus G0 and G00 from sample G2 obtained from DFS with TTS applied, at 1% shear strain in Brands et al. (2000) together with four mode t showing good resemblance of small strain behavior.

Fig. 5. Experimental apparent shear stresses (sample G2 in Brands et al. (2000)), at 5%, 10% and 20% maximum edge strain, together with the FE simulation results. Left: sample edge shear strain histories, Right: apparent shear stresses, Num result of three-dimensional FE model, Exp experiment: The numerical predictions are systematically higher than the experimental results regardless of strain and time.

1
80 80 Num Exp =0.2 40 20 0 0 0=0.05 0.1 0.2 0.3 Time [s] 0.4
0

h()

0.8
Stress [Pa]

Stress [Pa]

60

60 40 20 0 0

0=0.2 =0.1
0

0.6

Raw data Corrected data Fit result

5 10 15 20

0=0.1

0.4 0

0=0.05 0.05 Time [s] 0.1

[%]
Fig. 4. Normalized stiffness, hg; of sample G2 in Brands et al. (2000) obtained from the constant strain part of SR experiments on a plate plate viscometer. The raw experimental data is compared with data corrected for the non-homogeneous radial strain eld by applying second-order MooneyRivlin (MR2) model in Eq. (A.1). To obtain realistic ts of the normalized stiffness, (i.e. limg-0 hg 1), the normalized stiffness value at 5% strain is assumed to be valid at 1% strain also. Shear softening is more prominent in corrected data while tted result shows that the MR2 model provides excellent t.

Fig. 6. Investigation of error source of simulation results versus experimental data. Left: simulation results multiplied by a constant factor X 0:7 showing much better simulation results. Right: close up of the stress history during transient part of the deformation (scaled result).

4. Discussion A non-linear viscoelastic material law for brain tissue is developed and implemented in a FE code, to improve capabilities of FE head models to predict TBI in trafc accidents, sport accidents or falls. To prevent that errors in the computation of the hydrostatic stress interfere with deviatoric stress, we decoupled hydrostatic and deviatoric behavior by using isochoric strain measures for the deviatoric behavior. This is important since, due to the nearly incompressible material behavior, deviatoric stresses are typically six orders of magnitude smaller than hydrostatic stresses. A drawback of this approach is that it introduces the additional assumption that shear behavior is independent of volumetric compression. However, to our knowledge, evidence of such dependency is not reported in literature. The hydrostatic stress is modeled linearly elastic. Linearity seems valid as volumetric strains are small at pressures expected during trafc impacts (typically 105 Pa (Nahum et al., 1977)). Viscous effects are neglected, since ultrasonic experiments indicate signicant damping due to hydrostatic deformation only at

3.2. Test of the constitutive model The three-dimensional FE simulation of the rotational viscometer experiments, with material parameters in Table 1, provides predicted stress values which are systematically higher than experimental results (Fig. 5). A linear regression analysis (Statgraphics 5.1), shows good correlation between numerical and experimental results R2 > 96%: The ratio of predicted over experimental stress equals 1:3370:28; 1:3470:25; 1:2670:18 for 5%, 10% and 20% strain, respectively (average7 standard deviation) but does not statistically depend on strain at the 95% condence level (Students-T -test). The model thus provides good prediction of timedependent behavior but an average over estimation of the stress by 31% which does not depend on strain. Scaling the simulation results by a single factor of 0.7 indeed provides much better simulation results (strain averaged maximum relative error equals 7% at 0:2 s (Fig. 6)).

ARTICLE IN PRESS
132 D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134

frequencies above 50 kHz (Etoh et al., 1994), well above 1000 Hz typical for impacts of interest in this study. The deviatoric behavior is described using a differential formulation. Differential formulations have been proposed before (Bilston et al., 2001; Pamidi and Advani, 1978). However, these predict innite stress at instantaneous loading (Pamidi and Advani, 1978) or where not implemented in a FE package (Bilston et al., 2001; Pamidi and Advani, 1978). Based on the experimental nding of strain and time separability in stress relaxation (SR) experiments, we could have opted for a quasi-linear viscoelastic (QLV) model (Mendis et al., 1995; Miller and Chinzei, 1997; Prange et al., 2002). However, at frequencies above 44 Hz fully non-linear material behavior might be present (Darvish and Crandall, 2001) which cannot be described with QLV theory but can be included in the present formulation. The elastic behavior was modelled by a hyperelastic SEDF written in polynomial form with integer powers of rst and second invariants of the isochoric Finger tensor. We took a second-order model, the simplest model which predicts non-linear shear behavior. For correct prediction of shear softening observed (Fig. 4) negative second-order parameters were required. If the model is used outside its range of validity (i.e. strains more than 20%) this might result in negative stiffness when shear strains exceed 27%. Negative stiffness in simulations in a head model, in which it is unknown beforehand which strain levels will occur, can be avoided by extending the model with a third order MooneyRivlin term with suitable parameter settings as done in Brands (2002) and Brands et al. (2002). However, for realistic parameter settings, experiments at higher strains must be performed. Negative stiffness can also be avoided by using an Ogden SEDF with fractional powers of stretches Ogden (1972), applied to brain tissue in Miller and Chinzei (2002) and Prange et al. (2002, 2000). However, we found that this SEDF cannot predict the amount of shear softening observed in our experiments. The inelastic behavior is modeled by a simple linear Newtonian law. To describe the viscoelastic behavior in a broad frequency range, we used a multi-mode approach with discrete time constants as opposed to a relaxation spectrum, i.e. some arbitrarily continuous function of relaxation time (Macosko, 1994). A drawback of this method that each additional mode introduces ve new material parameters in the model. To reduce the number of independent parameters we assumed the ratio between rst- and second-order MooneyRivlin parameters governing the elastic strain-dependent behavior in each mode to be constant, based on the observation of (approximately) time-strain separability in SR data (Bilston et al., 2001; Brands et al., 2000; Prange et al., 2002).

We presented a new approach for determining material parameters of brain tissue on a rotational viscometer. Often all material parameters are determined from of SR experiments only while assuming perfect instantaneous strain application (Bilston et al., 2001; Prange et al., 2002). A drawback of this approach is that time constants valid during the rise time of the strain cannot be determined. Instead, we used two data sets from a single sample (Brands et al., 2000). Linear viscoelastic material parameters were tted to DFS results valid for small strains (0.01) but high frequencies (up to 684 Hz) while parameters C20 C02 ; describing the strain-dependent decrease of stiffness, were determined from the constant strain part of SR experiments. This method implies that time constants, determined for small strains and high frequencies, are also valid at large strains. Evaluating the predicted material behavior during transient strain application in the SR serves as a test of this assumption. During simulation of the SR experiments, stress values were overestimated by 31%, independent of the strain level applied, also during transient onset of the strain. A potential explanation is that we assumed no shear softening for strains between 1% (at which the DFS is performed) and 5%, the lowest strain value in the SR experiments, resulting in a normalized stiffness of 1.0 (Fig. 4). As a result, the tted model predicts only 2% shear softening at 5% strain. This softening is low compared to the 3050% softening reported in literature, when increasing shear strains from 1% to 5% (Brands et al., 1999; Bilston et al., 2001). Correct strain-dependent behavior shows that the correction method for the radially inhomogeneous strain eld is indeed valid. Correct time-dependent behavior indicates that time constants obtained from the DFS results are valid for frequencies over 20 Hz also. The DFS results were composed from isothermal DFS results up to 16 Hz; using the time temperature superpositioning (TTS) principle. This provides evidence on the validity of TTS for determining time constants of brain tissue for given model, as well as the validity of these time constants determined at small strains, for large strains and high frequencies. However, strain rates applied range up to 8 s1 which is less than the rate values expected during injurious impacts (15 21 s1 Brands (2002)). The behavior at higher strain rates remains to be investigated. It is impossible to fully characterize the threedimensional non-linear viscoelastic behavior of a material using simple shear experiments from a rotational viscometer. Effects of anisotropic material behavior (Prange et al., 2002) cannot be determined. Also, MooneyRivlin parameters C10;i ; C01;i ; C20;i and C02;i could not be dened uniquely. The ratio ratios C10 =C01 in a rst-order MooneyRivlin model did have a signicant effect on stress response in free compression

ARTICLE IN PRESS
D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134 133

experiments by Mendis et al. (1995). Adjusting the ratios Cj 0;i =C0j;i while keeping Cj0;i C0j ;i constant provides a powerful tool to tune free compression results without inuencing shear behavior. The performance of the model for other deformation modes remains to be investigated. In this paper, we developed a non-linear viscoelastic material law for brain tissue and implemented it in a FE code, to improve capabilities of FE head models to predict TBI in trafc accidents, falls or sports. We provided an approach to determine material parameters for the model and found that the model is capable of predicting realistic shear material behavior of brain tissue observed in SR experiments including the transient application of the strain. However, full characterization of the material properties is not possible using experimental shear data at hand and requires extra experiments.

Acknowledgements The authors would like to acknowledge TNO Prins Maurits Laboratory and Ford Motor Company for their nancial support of this research and P. Nauta of TNO Automotive for his assistance during the numerical implementation of the model.

Appendix. A Brands et al. (2000) determined the stress assuming linear viscoelastic theory, thus neglecting the effect of the radial inhomogeneous strain eld between the plates of the rotational viscometer. For any isotropic material, the true experimental normalized stiffness htrue gR can be obtained when the strain rate is zero, using Soskey and Winter (1984),   q ln ha gR htrue gR ha gR 1 : A :1 4q ln gR When experimental data at sufcient strain levels is present, q ln ha gR =4q ln gR can be estimated without making assumptions on material behavior. As we have data at three strain values only, we tted Eq. (19) to experimentally found ha g and applied Eq. (21) to obtain htrue g which is then tted to Eq. (19) again to obtain the correct material parameters.

References
Arbogast, K.B., Meaney, D.F., Thibault, L.E., 1995. Biomechanical characterization of the constitutive relationship for the brainstem. In: Proceedings of the 39th Stapp Car Crash Conference, SEA Paper 952716, pp. 153159.

Bain, A.C., Meaney, D.F., 2000. Tissue-level thresholds for axonal damage in an experimental model of central nervous system white matter injury. Journal of Biomechanical Engineering: Transactions of the ASME 122, 615622. Bandak, F.A., Eppinger, R.H., 1994. A three-dimensional nite elements analysis of the human brain under combined rotational and translational accelerations. In: Proceedings of the 38th Stapp Car Crash Conference, SAE Paper 942215, pp. 148163. Bilston, L.E., Liu, Z., Phan-Thien, N., 2001. Large strain behaviour of brain tissue in shear: some experimental data and differential constitutive model. Biorheology 38, 335345. Brands, D.W.A., 2002. Predicting brain mechanics during closed head impact: numerical and constitutive aspects. Ph.D. Dissertation, Eindhoven University of Technology, Eindhoven, The Netherlands. Available on internet: http://alexandria.tue.nl/extra2/ 200210809.pdf. Brands, D.W.A., Bovendeerd, P.H.M., Peters, G.W.M., Paas, M., van Bree, J., Wismans, J.S.H.M., 1999. Comparison of the dynamic behaviour of brain tissue and two model materials. In: Proceedings of the 43rd Stapp Car Crash Conference, Paper no. 99sc21, pp. 5764. Brands, D.W.A., Bovendeerd, P.H.M., Peters, G.W.M., Wismans, J.S.H.M., 2000. The large strain dynamic behaviour of in vitro porcine brain tissue and a silicone gel model material. Stapp Car Crash Journal 44, 249260. Brands, D.W.A., Bovendeerd, P.H.M., Wismans, J.S.H.M., 2002. On the potential importance of non-linear viscoelastic material modelling for numerical prediction of rain tissue response: test and application. Stapp Car Crash Journal 46, 103121. Brooks, C.A., Gabella, B., Hoffman, R., Sosin, D., Whiteneck, G., 1997. Traumatic brain injury: designing and implementing a population-based follow-up system. Arch Phys Med Rehabil 78 (8), S26S30. Claessens, M., Sauren, F., Wismans, J., 1997. Modelling of the human head under impact conditions: a parametric study. In: Proceedings of the 41st Stapp Car Crash Conference, no. SAE 973338, pp. 315328. Darvish, K.K., Crandall, J.R., 2001. Nonlinear viscoelastic effects in oscillatory shear deformation of brain tissue. Medical Engineering and Physics 23 (9), 633645. Estes, M.S., McElhaney, J.H., 1970. Response of Brain Tissue to Compressive Loading. No. ASME Paper, 70-BHF-13. Etoh, A., Mitaku, S., Yamamoto, J., Okano, K., 1994. Ultrasonic absorption anomaly of brain tissue. Japanese Journal of Applied Physics 33, 28742879. Galbraith, J.A., Thibault, L.E., Matteson, D.R., 1993. Mechanical and electrical responses of the squit giant axon to simple elongation. Journal of Biomechanical Engineering 115, 1322. Goldman, D.E., Hueter, T.F., 1956. Journal of the Acoustical Society of America 28, 35. Macosko, C.W., 1994. Rheology: Principles, Measurements, and Applications. Advances in Interfacial Engineering Series. VCH Publishers Inc., New York, NY, USA. Mendis, K.K., Stalnaker, R.L., Advani, S.H., 1995. A constitutive relationship for large deformation nite element modelling of brain tissue. Journal of Biomechanical Engineering: Transactions of the ASME 117, 279285. Miller, K., Chinzei, K., 1997. Constitutive modelling of brain tissue: experiment and theory. Journal of Biomechanics 30 (11/12), 11151121. Miller, K., Chinzei, K., 2002. Mechanical properties of brain tissue in tension. Journal of Biomechanics 35, 483490. Nahum, A.M., Smith, R.W., Ward, C.C., 1977. Intracranial pressure dynamics during head impact. In: Proceedings 21st Stapp Car Crash Conference, SAE Paper 770922, pp. 339366.

ARTICLE IN PRESS
134 D.W.A. Brands et al. / Journal of Biomechanics 37 (2004) 127134 41st Stapp Car Crash Conference, no. SAE 973335, SAE, pp. 277291. Soskey, P.R., Winter, H.H., 1984. Large step shear strain experiments with parallel-disk rotational rheometers. Journal of Rheology 28 (5), 625645. TNO-Automotive, 1999. MADYMO Users Manual 3D. Version 5.4 Edition. TNO-Automotive, Delft, The Netherlands. Turquier, F., Kang, H.S., Trosseille, X., Willinger, R., Lavaste, F., # Tarriere, C., Domont, A., 1996. Validation study of a 3D nite element head model against experimental data. In: Proceedings of the 40th Stapp Car Crash Conference, SAE Paper 962431, pp. 283294. Viano, D., von Holst, H., Gordon, E., 1997. Serious brain injury from trafc related causes: priorities for primary prevention. Accident Analysis & Prevention 29 (6), 811816. Waxweiler, R.J., Thurman, D., Sniezek, J., Sosin, D., Niell, J.O., 1995. Monitoring the impact of traumatic brain injury: a review and update. Journal of Neurotrauma 12 (4), 509516. Zhang, L., Yang, K.H., King, A.I., 2001. Comparison of brain responses between frontal and lateral impacts by nite element modeling. Journal of Neurotrauma 18 (1), 2130. Ogden, R.W., 1972. Large deformation isotropic elasticityon the correlation of theory and experiment for incompressible rubberlike solids. In: Proceedings of the Royal Society London, no. Series A 326, London, UK, pp. 565584. Pamidi, M.R., Advani, S.H., 1978. Nonlinear constitutive equations for human brain tissue. Transactions of the ASME: Journal of Biomechanical Engineering 100, 4448. Peters, G., Meulman, H., Sauren, A., 1997. Application of the time temperature superposition theory on brain tissue. Biorheology 34, 127138. Prange, M.T., Meaney, D.F., Margulies, S.S., 2000. Dening brain mechanical properties: effects of region, direction and species. Stapp Car Crash Journal 44, 205214. Prange, M.T., Meaney, D.F., Margulies, S.S., 2002. Regional, directional, and age-dependent properties of the brain undergoing large deformation. Journal of Biomechanical Engineering: Transactions of the ASME 124 (2), 198207. Rheometric Scientic, 1993. Advanced Rheometric Expansion System, with 2K FRTN1 Force Transducer and Liquid Nitrogen Air Temperature Control, Instrument Manual. Piscataway, NJ, USA. Schreiber, D.I., Bain, A.C., Meaney, D.F., 1997. In vivo thresholds for mechanical injury to the blood-brain barrier. In: Proceedings of the