Sunteți pe pagina 1din 12

International Journal of Impact Engineering 38 (2011) 252e263

Contents lists available at ScienceDirect

International Journal of Impact Engineering


journal homepage: www.elsevier.com/locate/ijimpeng

Failure criterion for laminated glass under impact loading and its application in nite element simulation
T. Pyttel a, *, H. Liebertz b, J. Cai c
a

University of Applied Science Giessen-Friedberg, Wilhelm-Leuschner-Str. 13, 61169 Friedberg, Germany Volkswagen AG, 38436 Wolfsburg, Germany c ESI GmbH, Mergenthaleralle 15-21, 65760 Eschborn, Germany
b

a r t i c l e i n f o
Article history: Received 29 April 2010 Received in revised form 26 October 2010 Accepted 28 October 2010 Available online 10 November 2010 Keywords: Glass Windshield Finite-element-model Failure Non-local criterion

a b s t r a c t
A failure criterion for laminated glass in case of impact is presented. The main idea of this criterion is that a critical energy threshold must be reached over a nite region before failure can occur. Afterwards crack initiation and growth is based on a local Rankine (maximum stress) criterion. The criterion was implemented in an explicit nite element solver. Different strategies for modeling laminated glass are also discussed. To calibrate the criterion and evaluate its accuracy, a wide range of experiments with plane and curved specimens of laminated glass were done. For all experiments nite element simulations were performed. The comparison between measured and simulated results shows that the criterion works very well. 2010 Elsevier Ltd. All rights reserved.

1. Introduction Impact against safety glass is of interest for example to civil engineers and car makers. The motivation for the work presented in this paper results from pedestrian protection in accidents involving a car. One scenario is that the pedestrian impacts the windshield with his head. In order to save the life of the pedestrian the acceleration of the head should not exceed a certain value. In addition it is important that the head does not touch any part inside the car (for example the steering wheel). Today the development process for cars is based on numerical simulations. For simulations regarding pedestrian protection explicit nite element solvers are used. The correct calculation of impactor acceleration and deformation of the glass is still a challenge. A simulation model must be able to predict the initial failure of the glass and the crack propagation. Several experimental and numerical studies have been carried out in the past to investigate the behaviour of laminated glass under dynamic loading conditions. One group of researchers dealt with laminated glass [1e3] and another group with pure glass [4]. Several modeling techniques for laminated glass are used. In Ref. [2]

* Corresponding author. E-mail address: thomas.pyttel@m.fh-friedberg.de (T. Pyttel). 0734-743X/$ e see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijimpeng.2010.10.035

a modeling technique with two coincident layers of shells was used, one for glass and one for PVB. Results with solid elements for glass and PVB have also been published [3]. Models based on shell elements are easier to use - especially when the modelled windshield is part of an entire car model. However, the two-dimensional stress state is a disadvantage for treating damage and failure. Local approaches for the simulation of damage and failure in laminated glass are presented in Ref. [2,3]. In Ref. [4] the failure of glass rods is investigated. Here a nonlocal approach was introduced with an a characteristic length for the pure glass of about 1 mm. In Ref. [5] for the failure of laminated glass ve phases were introduced. The rst phase is an elastic deformation for both glass plies. The second phase is the breaking of the rst glass ply. In order to simulate the duration of the rst phase we used a nonlocal approach in this study. Our characteristic length is much higher than in Ref. [4]. The reason is that our model describes the compound glass-PVB-glass. In Ref. [2] curved parts (real windshields) and in Ref. [1] plane parts (real car side glazing) were tested and simulated. We also used experiments with windshields but in addition we designed a test setup for plane parts in order to eliminate the inuence of the curvature and to study the inuence of the boundary conditions. The initial failure of the glass depends on the velocity of the impactor and the curvature of the glass. This is shown by the experiments presented in this paper. After initial failure the cracks

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

253

Fig. 1. Assembly of safety glass for windshields (values for thickness in mm).

grow with a velocity of about 2000 m/s [6]. A failure criterion must be able to describe these two effects. The safety glass studied in this paper consists of three layers shown in Fig. 1. There are two outer layers of glass with different thickness and one inner layer of polyvinyl butyral (PVB). The 2.10 mm glass layer lies on the car outer surface. The glass used for the test in this study was annealed. The idea for the failure model presented in this paper is based on experimental observations. Obviously the safety glass does not fail in any case immediately after impact. In many cases for a certain time after impact elastic deformations occur without any failure. In order to model the observed behaviour it was assumed that a critical energy value must be exceeded in a nite region before the safety glass fails. This is what we call Nonlocal failure criterion. The criterion was implemented in the commercial nite element solver PAM-CRASH [7]. The simulations were done with the explicit solution strategy, which is described in detail in Ref. [8]. In order to compare the capabilities of the failure model experiments were performed. These experiments were done with plane and curved safety glass specimens. The impactor weight and velocity varied from light and slow to heavy and fast. Based on these experiments the parameters for the criterion were found. Furthermore material properties for the constituents of safety glass are needed. 2. Experimental determination of failure 2.1. Experimental setup Two series of tests were done. In the rst, curved safety glass specimens were tested and in the second, plane ones. The tests with the plane parts were designed to provide information on the principal failure mechanism and effects during impact against laminated glass. In order to study the inuence of curvature real windshields were used. The impactor used for all tests was an head-impactor of the family of pedestrian protection models [9]. This impactor is available

Fig. 3. Boundary condition and sensor positions for displacement measurement (R 500 mm, Distance between sensors: a 50 mm).

in three different weights (3.5 kg, 4.8 kg and 6.3 kg). A three-axis accelerometer was mounted inside the head impactor. This accelerometer is used for all the acceleration measurements of this study. Fig. 2 shows the experimental setup for plane parts. On the left side a quadratic part is glued to a circular frame. The frame was designed to be as stiff as possible. The acceleration inside the impactor and the deformation of the glass during impact were measured. For this purpose a set of ten laser extensometers shown in Fig. 3 was used. For the test shown on the right side of Fig. 2 the plane part was hanging freely on two cables. The advantage of this conguration is the stress-free boundary condition. For all these cases the position of impact was the center of the part. Table 1 gives an overview of the tests performed with plane parts. The test number consists of three groups: rst impactor speed, second impactor weight and third information regarding boundary conditions. For example 05_3_b means 5.0 m/s, 3.5 kg, bounded. Tests against real windshields are standard tests done by car makers in order to protect pedestrians or passengers in case of an accident. A failure model for safety glass must be able to describe the different phenomena for different loadcases. The challenge is to cover the range from impacting the windshield outside the car, inside and at different positions. Fig. 4 shows the setup for the two tests performed. Table 2 gives an overview of the tests performed with curved parts.

Fig. 2. Experimental setup for plane parts (left bounded, right free).

254 Table 1 Loadcases for the impact tests with plane parts. Bounded 5.0 m/s 3.5 kg 4.8 kg 6.3 kg 05_3_b 05_4_b 05_6_b 10.0 m/s 10_3_b 10_4_b 10_6_b

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263 Table 2 Loadcases for the impact tests with curved parts. Free 12.0 m/s 12_3_b 12_4_b 12_6_b 10.0 m/s 4.8 kg 10_4_f Table 3 Time in ms between rst impact and failure for curved parts. Pos.1 10.0 m/s 10_4_pos1 Pos. 2 10.0 m/s 10_4_pos2

2.2. Selected results and discussion The high speed video of the tests gives the information regarding the time between rst impact and global failure of the glass. These results are collected in Table 3 for the impacts against the real windshield. Here the difference between impact from outside and from inside is exceptionally obvious. In case of the plane parts, deformation was measured in addition to acceleration. Typical results are shown in Figs. 5 and 6. Exact measurement of deformation during impact is very important for the development of a failure model. The simulation results of a successful failure model should not only match the measured acceleration, but also the measured deformation. The impact velocities were chosen to cover the range from no failure to totally damaged. The test with the highest impactor mass and the highest impactor speed nearly penetrates the safety glass and for the test with the lowest impactor mass and the lowest impactor speed no crack was observed. For the tests with the real windshields just the acceleration was measured. Fig. 7 shows the result for position 1 and 2. Obviously due to curvature the behaviours of case 10_4_pos1 and 10_4_pos2 are totally different. If the impactor comes from outside the car the rst response of the windshield is purely elastic. At 5.0 ms the glass fails. From 7 to 15 ms there is a free movement of the impactor. Finally the PVB supports the impactor with some inuence from the broken glass. If the impactor comes from inside the car the glass fails much earlier - at 1.2 ms. Compared with 10_4_pos1 the rst maximum of the impactor acceleration is then lower, the second maximum is higher and there is no free movement. 3. Nonlocal failure criteria 3.1. General description Glass fails without any localisation perpendicular to the mean loading direction. This is the motivation for the application of the Rankine (maximum stress) criterion

Pos.1 10.0 m/s 4.8 kg 5.0

Pos. 2 10.0 m/s 1.2

sfail maxjs1 j; js2 j

(1)

shown in Fig. 8. (Note that there is no failure when all stresses are compressive). Based on double ring tests the critical stress sc can be evaluated. The double ring test is like a four-point-bending test, but with circular loading surfaces in place of linear ones. This test was performed without any PVB under quasi-static conditions. In Ref. [2] the four-point-bending test was used. The results are similar to ours. The rst idea is to use this criterion within an explicit nite element solver as a local criterion. This means that for each integration point the criterion is checked for each time step (Local criteria are used for example in Ref. [2]). But important experimental phenomena cannot be described with such a procedure. An example is the impact against the windshield from outside the car (10_4_pos1). Directly after the impactor has touched the glass the calculated stresses of the nite elements in the contact zone are much higher than the critical value sc. Based on the Rankine criterion these elements would fail immediately and following this the crack pattern will grow by moving from element to element. In our simulation this crack grow takes about 10 ms until the crack pattern has reached the boundary. The reality looks different: Until 5.0 ms no failure can be observed and than, within an interval of less than 1 ms, the windshield fails completely. In order to describe the compound glass-PVB-glass the Rankine criterion was enhanced by a nonlocal approach. The basic idea is that a critical energy value (Ec) must be reached over a nite region before the glass can fail, independent of the calculated stress. The nite region is a circle of radius (Rc) on the rst element to reach the critical the point of rst failure stress sc. Fig. 9 shows the general idea of this approach.

Fig. 4. Experimental setup for curved parts (real windshields).

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

255

Fig. 5. Acceleration of the impactor (left) and deformation of the center (right) for test 10_4_b.

Fig. 6. Deformation for each sensor point for test 10_4_b (left at 5.0 ms and right at 20.0 ms).

The strain energy inside the nite region Rc (red circle in Fig. 9) is calculated for each time step i and for each element e. For the calculation of the time increment for one element

E > Ec

(4)

DEie

ZZ

_ dV e Dti s : d3

(2)

_ the strain rate tensor is used. In this equation s is the stress tensor, 3 and Dti the duration of the time step. The total strain energy for one element after n time steps follows then by

an orthotropic damage/failure criterion is activated for each element of the whole windshield in order to model crack propagation. An orthotropic damage tensor acts uniformly over the shell thickness and affects the undamaged stresses sij by

Ee tn

i n X i0

1 d1 s11 ~ij 4 1 dm s12 s 1 dm s13

1 dm s12 1 d2 s22 1 dm s23

3 1 dm s13 1 dm s23 5 0

(5)

DEie

(3)

The summation of Ee tn over all shell elements inside the nite region Rc delivers the strain energy E inside the nite region Rc at time tn. If

~ij are the damaged stresses in a coordinate system of principal axes s calculated based on stresses at the time the failure criterion is activated. This coordinate system is xed with respect to the glass material, when the rst damage occurs. d1 is the damage value in the direction of the rst principal axes, d2 is the damage value in the direction of the second principal axes (in the coordinate system

Fig. 7. Acceleration for test 10_4_pos1 (left) and 10_4_pos2 (right).

256

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263 Table 4 Parameter of the GSell model. k 9.0 MPa w 4.0 h 1.5 m 1.5

introduced above) and dm is the maximum of d1 and d2. The damage values d1 and d2 are increased linearly over 100 time steps after initial damage from 0 to 1. Initial damage means for d1 that the rst principal stress is greater than sc of the Rankine criterion and for d2 that stress in 2-direction is greater than sc. Elements are nally eliminated when

d1 d2 1
Fig. 8. Rankine failure criterion.

(6)

is reached. In order to avoid numerical instabilities an element based damping was introduced. The damping stress sd is calculated as

sd

pf0

_: E:3

(7)

Fig. 9. General idea of the nonlocal approach.

where E is the elasticity tensor, f0 is the elemental frequency and for the factor k the value 0:05 was used. Due to the nonlocal approach it is possible to model the rst phase - the pure elastic deformation without any failure. The duration of this phase is determined by the two parameters Ec and Rc. During this phase the elastic stresses of some elements may exceed sc in both principal directions. At the end of this phase E > Ec for these elements damage d1 and d2 starts to growth and after 100 s (time step of 1 s assumed) these elements will be

Fig. 10. Principal idea of the failure model.

Model A Shell Solid

Model B Shell Solid

Model C Shell Membrane

Tied connection
Fig. 11. Three possibilities of modeling safety glass.

Tied connection

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

257

Fig. 12. Determination of EC and RC values based on curved parts.

eliminated. That means within 100 s a crack pattern over a nite region of glass will develop. If we assume that this nite region is a circle with a radius ofr 200 mm (what is a good mean value with respect to the experiments we have done) the same region would fail in reality based on a constant physical crack propagation speed of v 2 km=s beginning from the center of the circle (200 mm 2 km=s$100 ms). 3.2. Determination of parameters In order to nd the critical energy Ec and the critical radius Rc, at least two tests are necessary. For the rst test the glass should fail

immediately after impact and for the second as late as possible. The idea is that the failure time (tf ) should be different between these two tests. tf can be evaluated for each test by using the a high speed video. In this study impacts from inside and outside of a car windshield were used. Fig. 10 shows principal curves energy over time for two impacts with three different radii (Rm, Rn and Ro). These curves are calculated from a nite-element model of the test cases, but without any failure criteria. Thus the calculation of many energy curves starting with a small radius up to a large radius is possible. The key for the determination of Ec and Rc are the failure time for each test tf1 and tf2 determined by the high speed video. Only for

Fig. 13. Cross section of the FE-model of head and windshield (green-aluminium skull, blue and brown-vinyl skin). (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

258

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

Fig. 14. 3.5 kg, 5.0 m/s (Testnumber: 05_3_b) acceleration of the impactor (left) and deformation normal to the undeformed specimen at the center (left).

the radius Rn (solid line in Fig. 10) one energy value E can be found which is the same at tf1 and tf2. This value E will be used as critical value Ec and the radius Rn will be used as Rc. In the simulation the laminated glass will fail for both tests at the right time only with this pair of parameters. 3.3. Implementation of the failure criteria in a nite element solver

Until E< Ec no failure is possible, even when the critical failure stress sc has been exceeded. As soon as E reaches the critical value Ec the Rankine criterion is activated. Due to the fact that at this time the stress of many elements of the part is greater than sc the cracks will develop immediately. Further crack growth is of course possible. 4. FE-model for laminated glass

The implementation was done in the commercial nite element solver PAM-CRASH [7]. During simulation the center of the nonlocal energy region must be determined. Once this point is known all elements within the radius Rc are assigned to the energy calculation. The sum of the internal strain energies for these elements gives the current value E.

4.1. Finite element discretisation In the rst impacting phase the behaviour of the glass is important for the response of the structure. After the glass has failed the PVB will carry the load. This is the reason why both

Fig. 15. 3.5 kg, 5.0 m/s (Testnumber: 05_3_b) deformation normal to the undeformed specimen (upper left at 2.0 ms, upper right at 5.0 ms, lower left at 10.0 ms, lower right at 20.0 m/s).

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

259

Fig. 16. 4.8 kg, 10.0 m/s (Testnumber: 10_3_b) acceleration of the impactor (left) and deformation normal to the undeformed specimen at the center (left).

materials must be modelled separately. For a discretisation with nite elements we tested three modeling techniques (shown in Fig. 11). Model A consists of shell elements for the glass and solid elements for the PVB. The connection between the nodes of the solids and the shells is described with tied elements. These tied elements are like small beams with a penalty based stiffness. The tied elements will be automatically eliminated as soon as all connected shell elements (glass) are failed. There is no certain failure criterion for the tieds. The distance between the nodes of the shell

elements and the nodes of the solids amounts to half the thickness of the glass layers. Model B works without the tied connection. For the mesh of solid elements the nodes of the shells are used. The disadvantage is that the thickness of the PVB is greater in the model than in reality and the material characters have to be compensated. The simplest approach consists of two layers of shells for the external glass and one membrane layer for the PVB in between (model C). As in model A the connection between the nodes of the shell and the membrane elements is described with tied elements.

Fig. 17. 4.8 kg, 10.0 m/s (Testnumber: 12_4_b) deformation normal to the undeformed specimen (upper left at 2.0 ms, upper right at 5.0 ms, lower left at 10.0 ms, lower right at 20.0 m/s).

260

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

Fig. 18. 6.3 kg, 12.0 m/s (Testnumber: 12_6_b) acceleration of the impactor (left) and deformation normal to the undeformed specimen at the center (left).

Our focus was to make the model as simple as possible. So we preferred model C. This modeling technique was used for all the simulations presented in this paper. Several tests with a solid based modeling (model A and model B) produced no advantage. 4.2. Material properties and parameters for the failure criterion The material model for the glass is isotropic and linearly elastic. The properties used for Youngs modulus E70 GPa, Poissons ratio v0:22 and the failure stress sC 0:06 GPa derived from the literature [6].

The material description of the PVB is based on a nonlinear viscoelastic model. The stressestrain relation for this model follows the GSell equation [10].

i 2 h _ k 1 ew3 eh3 s 3; 3

_ ref 3

_ 3

!m (8)

The properties for this model were determined from dynamic tensile and bulge tests and are given in Table 4. The GSell model offers the possibility to inuence the stressestrainecurve in three regions:

Fig. 19. 6.3 kg, 12.0 m/s (Testnumber: 12_6_b) deformation normal to the undeformed specimen (upper left at 2.0 ms, upper right at 5.0 ms, lower left at 10.0 ms, lower right at 20.0 m/s).

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

261

Fig. 20. 4.8 kg, 10.0 m/s (Testnumber: 10_4_f) deformation (left) and acceleration of the impactor (right).

Fig. 21. 4.8 kg, 10.0 m/s (Testnumber: 10_4_b) crack geometry after impact - left experiment, right simulation.

e parameter k inuences mainly the small strain region, e parameter w inuences mainly the medium strain region and e parameter h inuences mainly the large strain region. The description of the strain rate dependency of the material is based on the parameter m. The GSell model was successfully used for the modeling of impacts with plastics [11].

The determination of the parameters of the nonlocal failure criterion is based on the experiments for the curved parts. The difference between the failure times (given in Table 3) for the impact from outside the car and inside the car is sufciently large (4.3 ms). The failure times for the tests with the plane parts are all about 1.5 ms. Since the difference in failure time between two experiments is important for parameter calibration, it was not possible to

Fig. 22. 4.8 kg, 10.0 m/s (Test: 10_4_pos1) acceleration of the impactor (right).

262

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

Fig. 23. 4.8 kg, 10.0 m/s (Test: 10_4_pos2) acceleration of the impactor (right).

calculate Ec and Rc based on plane parts. Fig. 12 shows the calculated energy curves for three different radii for the two loadcases with the curved parts. The failure times are also marked with vertical lines. For a radius RC 210 mm an energy value EC 22:3 kNmm is found. These values are used for all the simulations in this study. 5. FE-simulation of impact tests and evaluation 5.1. FE-model of the impactor and contact to the windshield The impactor consists of a hollow aluminium sphere covered with an rubber skin. The modeling technique we used follows [12]. The aluminium sphere and the rubber skin are modelled with solid elements. For the sphere a linear elastic constitutive law and for the rubber skin a viscoelastic law were used. The contact denition used between the outer surface of the head-impactor and the windshield is a symmetric point-to-surface penalty formulation [13]. Fig. 13 shows the head and the windshield. The two outer layers of solids are the rubber skin. For the windshield the model C shown in Fig. 11 is used. 5.2. Boundary conditions There are three types of boundary conditions related to the three test setups. First, the glass is xed to a circular frame and the frame is assumed rigid. Second, the glass is attached to the car body (in the case of real windshields) and the car body is deformable. Third, the glass is hanging freely on two cables. The last type of boundary condition was chosen because it can easily be exactly realised in the simulation. For a xed boundary condition there is always the question how stiff this boundary condition is in reality. Without the nonlocal failure criterion simulation results depend strongly on the stiffness of the boundary condition. With the nonlocal criterion this inuence is signicantly reduced. Based on this experience the attachments glass-frame and glass-car are both realised with a penalty option. The stiffness of this numerical connection is chosen to be as large as possible, but without inuencing the time step of the model. 5.3. Comparison between experiment and simulation for plane parts For the congurations where the plane parts are xed at the frame the comparison between test and simulation is done for the acceleration of the impactor and the deformation of the safety glass. Regarding the deformation the movement of the center point over time is compared and in addition the shape of the deformed safety glass for four different time points (2 ms, 5 ms, 10 ms and 20 ms).

Experimentally the shape was measured with laser extensiometers at 10 points described in section 2.1. Thus the experiment is represented just by points in the diagrams. From the nine congurations tested (see Table 1) only three are shown. The tests with e lowest speed and lowest impactor weight (05_3_b) e medium speed and medium impactor weight (10_4_b) e highest speed and highest impactor weight (12_6_b) are representative. Figs.14,16 and 18 plot acceleration and deformation over time and Figs. 15, 17 and 19 show the shape of the part during deformation. In addition a comparison of the crack geometry for medium impactor speed and medium impactor weight (10_4_b) is shown in Fig. 21. Both, in simulation and experiment there are two main areas. An inner area with radial and circumferential cracks and an outer area with only a few radial cracks (Fig. 22). No deformation was measured for the test with the free hanging plane part. Thus there is just the comparison for the acceleration of the impactor shown in Fig. 20.

5.4. Comparison between experiment and simulation for curved parts For the impact against the windshields only the acceleration was measured. This curve is compared with the result from the simulation (Fig. 23). Finally a comparison between simulations with and without nonlocal failure criteria for test 10_4_pos1 was made. Fig. 24 shows that a simulation without nonlocal criterion is not able to describe

Fig. 24. 4.8 kg, 10.0 m/s (Test: 10_4_pos1) comparison of acceleration of the impactor with and without nonlocal failure criterion.

T. Pyttel et al. / International Journal of Impact Engineering 38 (2011) 252e263

263

the behaviour during the rst 5 ms. The glass fails immediately after impact. Due to this the acceleration never reaches the level what was measured. 6. Discussion and outlook The experiments cover the range from no failure to total damage and from convex to concave. For all these cases the simulation model is able to calculate impactor accelerations and deformations very well. The set of input parameters for the glass consists only of ve values: e e e e e Youngs modulus of glass Poissons ration of glass Critical stress for failure of glass sc Critical energy for failure of glass Ec Critical radius for failure of glass Rc

References
[1] Herndorn G, Allen K, Roberts A, Phillips D, Batzer S. Automotive side glazing failure due to simulated human interaction. Engineering Failure Analysis 2007;14:1701e10. [2] Timmel M, Kolling S, Osterrrieder P, Bois PD. A nite element model for impact simulation with laminated glass. International Journal of Impact Engineering 2007;34(3):1465e78. [3] Zhao S, Dharani L, Chai L, Barbat S. Analysis of damage in laminated automituve glazing subjected to simulated head impact. Engineering Failure Analysis 2006;13:582e97. [4] Repetto E, Radovitzky R, Ortiz M. Finite element simulation of dynamic fracture and fragmentation of glass rods. Computer Methods in Applied Mechanics and Engineering 2000;183:3e14. [5] Larcher M. Simulation of laminated glass loaded by air blast waves, DYMAT 2009-9th International Conferences on the Mechanical and physical behaviour of materials under dynamic loading. EDP Sciences 2009;2: 1553e9. [6] Kerkhof F. Bruchvorgaenge in Glaesern. Verlag der deutschen glastechnischen Gesellschaft; 1970. [7] PAM-CRASH. 100e102 Avenue de Suffren. 75015 Paris - France: ESI-Group; 2010. [8] Belytschko T, Liu WK, Moran B. Nonlinear nite elements for Continua and Structures. Wiley; 2001. [9] Laboratory Test Procedures for FMVSS 201U. Occupant protection in Interior impact, upper Interior head impact protection. U.S. Department of Transportation, National Highway Trafc Safety Administration; 1998. [10] GSell J. Rheology of polypropylene in the solid state. Journal of Materials and Sciences 1995;30(3):701e11. [11] Pyttel T, Weyer S. Crash simulation with glassy polymers e constitutive model and application. Internal Journal of Crashworthiness 2003;8:433e42. [12] Dharani L.R., Ji F.S., Dynamic analysis of normal impact of occupant head on laminated glass, SAE Paper 933114. [13] Belytschko T, Lin JI. A three-dimensional impact-penetration algorithm with erosion. Internal Journal of Impact Engineering 1987;5:111e27.

In addition the viscoelastic parameters of PVB are needed. The good agreement between experiment and simulation shown in this paper is only possible due to the enhancement of the Rankine criterion with the nonlocal approach described in section 3. The characteristic length we used is very large compared to known values for pure glass. This could be because that a compound structure with a strongly strain rate dependent layer (PVB) is considered. Further investigations are planned in order to develop cheap and easy methods for the determination of the two parameters EC and RC.

S-ar putea să vă placă și