Sunteți pe pagina 1din 6

Urban Habitat Constructions under Catastrophic Events (Proceedings) Mazzolani (Ed).

. 2010 Taylor & Francis Group, London, ISBN 978-0-415-60685-1

A feasibility study on modeling blast loading using ALE formulation


L. Kwaniewski, M. Balcerzak & J. Wojciechowski s
Warsaw University of Technology, Warsaw, Poland

ABSTRACT: The paper presents a comparison of numerical data for computer simulations of blast loading. Based on the published experimental data, a benchmark problem was selected, where the pressure loading subjected to a rigid steel plate and produced by near field hemispherical charges, is considered. The comparison is done in terms of peaks of reflected pressure (overpressure), reflected specific impulses, and time histories of reflected pressure. The numerical study was conducted using mainly three dimensional Arbitrary Lagrangian Eulerian (ALE) formulation implemented in the commercial code LS-DYNA . Several other modeling options and strategies were also considered and validated through comparison with the available experimental data. The sensitivity study on mesh resolution for ALE meshes was also considered. Although several discrepancies were indicated, the premature validation study shows big potential for this type of numerical modeling.

INTRODUCTION

direct empirical load with fluid-structure interaction using ALE 3D domain (Slavik 2009). 2 SELECTED BENCHMARK PROBLEM

The experimental and numerical studies on structures subjected to blast loads show complexity of the problem and many challenges facing this research area, Bulson (1997). Among the most fundamental objectives, there can be mentioned three: a reliable prediction of loads imposed on structures by explosives, correct representation of material behavior, and global analysis of large scale structures. This paper reports a preliminary feasibility study on approaches for numerical modeling of blast loads, implemented in the commercial program LS-DYNA (Hallquist 2009). Four approaches have been considered including: explicit blast wave representation using fluid-structure interaction with 3D and 2D multimaterial arbitrary Lagrangian-Eulerian (ALE) formulations, direct application of empirical explosive blast loads on structures, and the most recent, combined method, in which direct empirical loading is subjected to the ALE domain. Each of these approaches has its own advantages and limitations, although the last one seems to be the most universal. Numerical models of fluid-structure interaction with 3D ALE formulation are able to capture complex interaction of reflected and dispersed blast waves but are computationally very demanding and require very dense meshes. The ALE elements must be small enough to capture the nearly discontinuous shock front of the blast wave. The 2D approach is computationally much more efficient but is only applicable to simple axisymmetric problems. The application of direct empirical blast loads is the most effective method but it is limited to free-air cases where there are no interferencing objects which can cause reflecting or focusing of blast waves. To overcome this limitations a new method has been developed recently in LS-DYNA which combines

At this stage of the reported feasibility study the numerical results are compared with published experimental data for a selected problem investigated by Pope & Tyas (2002). In a series of experiments, 78 g PE4 hemispherical explosive charges were used to load a target plate at short stand-offs, varying from 400 to 1000 mm. With special care taken to produce repeatable testing conditions, Pope & Tyas obtained highly reliable pressure time histories recorded by pressure transducers placed in the target semi-rigid plate. A typical test setup is presented schematically in Figure 1. Comparison is done in terms of arrival time, peaks of reflected pressure (overpressure), reflected specific impulses for the time histories of reflected pressure at the central point of the target plate. Obtained experimental results are in good correlation with approximations given by well-known Friedlander function (Slavik 2009), with the maximum reflected pressure decreasing and arrival times increasing for increased stand-offs. It was also confirmed that at the near field stand-offs the charge shape is an important factor and can result in different loading magnitudes for the same charge masses. 3 FLUID-STRUCTURE INTERACTION USING ALE FORMULATION

3.1 Arbitrary Lagrangian-Eulerian formulation The ALE methods enable to conduct calculations for coexisting Lagrangian and Eulerian meshes and in this way to reproduce interaction between fluid and a

127

Figure 2. Definition of hemispherical charge using option initial volume fraction geometry for 2D mesh 2.5 2.5 mm.

Figure 1. Test setup for a series of experiments used by Pope & Tyas (2002) to produce time histories of reflected overpressure.

deformable structure.The pure Lagrangian description makes it easy to track interfaces and to apply boundary conditions. The ALE mesh, created to represent the fluid domain, is usually undistorted (can be automatically expanded) and might be fixed (as in typical Eulerian formulation) or can move according to the prescribed, arbitrary conditions, Hallquist (2009). An ALE mesh can be built of multimaterial elements or elements filled with a fluid and void. Interfaces and boundary conditions are difficult to track using this approach; however, mesh distortion is not a problem because the mesh is smoothed to its original shape and dimensions after each calculation cycle. In the model, fluid flows through the undistorted mesh and, due to coupling algorithms, interacts with a structure represented by a Lagrangian mesh or rigid boundary conditions. The Lagrangian mesh is independent and usually is immersed in the ALE regular mesh which defines the space where the fluid motion is traced. The ALE and Eulerian methods implemented in explicit FE codes such as LS-DYNA are based on the classical Lagrangian formulation (Hallquist 2009). The solution algorithm is incremental consisting of finite element explicit (Lagrangian) time steps followed by so called advection steps. For pure Lagrangian approach, where a mesh moves with the fluid, extensive element distortions are likely to occur with the mesh degeneration leading to small or negative volumes. To avoid numerical problems due to large distortions, the finite element mesh is smoothed when deformations become too large during the computation. The ALE formulation enables cyclic and automatic rezoning, based on the deformed boundaries between Lagrangian cycles. The advection step carries out an incremental remapping followed by the transportation (advection) of nodal and element variables, from the distorted to the new mesh. The computational cost of an advection step is typically two to five times the cost of the Lagrangian time step (Hallquist 2009). For blast analysis the ALE domain is filled with air modeled as ideal gas with the linear polynomial equation of state (EOS). The charge is modeled using special purpose material model named High Explosive Burn characterized by the Jones-Wilkins-Lee EOS

(Hallquist 2009). In this way the explosive as well as the air are explicitly modeled allowing for tracing the blast wave propagation through the ALE air domain and interaction with the Lagrangian structure through fluid-structure interaction. More detailed description of this novel formulation can be found in Donea (1980) and Hallquist (2009). 3.2 Axisymmetric 2D models An axisymmetric model is built of regular square elements with the symmetry axis set default along the y-axis of the model. The shape of the explosive material is cut out from the domain defined by multi-material ALE elements using initial volume fraction geometry option, as shown in Figure 2. The target plate is represented by boundary conditions constraining y-translation of the ALE nodes at the specified distance from the explosive material. All other boundaries have default, nonreflecting conditions. The calculations were carried out for meshes with different element sizes: 5 5 mm, 2.5 2.5 mm, and 1.25 1.25 mm to verify the effect of mesh density on the results. Figure 3 shows contours of pressure showing propagation of blast wave for 500 m stand-off. Figure 4 presents time histories of reflected overpressure for all considered stand-offs. It can be noticed that the arrival time of the blast waves is the same as for the experiments but the shape of the function differs from the typical Friedlander description. The peak reflected overpressures, calculated and from the experiments are compared in Figure 5. As expected the values are higher for smaller element size, but for the smallest considered element size of 1.25 mm the solver overestimates the results. Additionally, for the finer meshes some disturbances in reflected overpressure histories appear. Figure 6 presents comparison of calculated specific impulses with the experiment. The same as in the case of reflected overpressure, results are overestimated for the finest mesh in comparison to the experiment and other mesh densities indicating computational problems of the solver. 3.3 ALE 3D model Several modeling issues have to be considered for direct simulation of fluid-structure interaction with 3D ALE formulation. First, to reproduce properly the blast wave propagation, the air domain must be represented

128

Figure 3. Contours of pressure showing propagation of blast wave for 500 m stand-off and element size 2.5 mm.

Figure 7. Views of a quarter of the hemispherical charge defined using volume-filling command initial volume fraction geometry.

Figure 4. Time histories of reflected overpressure for different stand-offs and element size 2.5 mm.

Figure 8. Snapshots of iso-surfaces of pressure for 400 mm stand-off modeled with 3D ALE mesh of 5 mm.

Figure 5. Comparison of peak reflected overpressure for experiment and 2D ALE with element sizes 1.25, 2.5, and 5 mm. Figure 9. Time histories of reflected overpressure for element size 5 mm and different stand-offs.

Figure 6. Comparison of specific impulses for experiment and 3D ALE models with element sizes 1.25, 2.5, and 5 mm.

by a regular ALE mesh built of hexahedral elements preferably with the aspect ratio of unity. To include charges, two options can be applied. The material model for explosive charge can be associated with selected finite elements of the ALE mesh but with a rectilinear mesh this option is applicable only to cubical charges. For charges with complex shapes, an option called initial volume fraction geometry (Hallquist 2009) can be applied. This is a

volume-filling command used for defining the volume fractions of the multi-material ALE elements which can be occupied by a selected material. Figure 7 shows application of this option for generation of one quarter of the hemispherical charge cut out with the symmetry planes (compare Fig. 2). To reduce number of finite elements, only one quarter of the volume is represented for the cubical space where the blast wave is traced. The modeled space is bounded by two symmetry planes with boundary conditions constraining displacements in the perpendicular direction. The target plate is also represented by constrained perpendicular displacements of the selected nodes. On the other external surfaces the default nonreflecting boundary conditions are set. The 3D ALE model calculations were carried out for the finite element sizes 5 and 10 mm. Figure 8 shows snapshots of iso-surfaces of pressure for 400 mm stand-off modeled with 3D ALE mesh

129

Table 1. Comparison of peak reflected pressure and arrival time for experiment and empirical blast loads. Stand-off 400 mm 1000 mm

Figure 10. Comparison of peak reflected overpressure for experiment and 3D ALE models with element sizes 5 and 10 mm.

Peak pressure [MPa] experiment 12.4 calculation 78 g spherical charge: elem. size 25 25 mm 5.953 elem. size 10 10 mm 5.597 elem. size 5 5 mm 1.073 calculation 78 g hemi-spherical charge: elem. size 10 10 mm 9.684 calculation 1.37 78 g spherical charge: elem. size 5 5 mm 7.804 calculation 1.37 78 g hemi-spherical charge: elem. size 5 5 mm 12.512 Arrival times of shock wave [ms] experiment 1.52 calculation 78 g spherical charge: elem. size 25 25 mm 1.88 elem. size 10 10 mm 1.99 elem. size 5 5 mm 1.99 calculation 78 g hemi-spherical charge: elem. size 10 10 mm 1.72 calculation 1.37 78 g spherical charge: elem. size 5 5 mm 1.83 calculation 1.37 78 g hemi-spherical charge: elem. size 5 5 mm 1.61

0.7 0.132 0.408 0.413 0.648 0.552 0.900 8.58 10.03 10.03 10.03 9.66 10.14 8.85

Figure 11. Comparison of specific impulses for experiment and 3D ALE models with element sizes 5 and 10 mm.

of 5 mm. Figure 9 presents example time histories of the reflected overpressure for element size 5 mm. The same as in the case of the ALE 2D model arrival time of blast wave is practically the same as for the experiment. Figures 10 & 11 present comparisons of the peak reflected overpressures and of the specific impulses, respectively. Although the results are underestimated in the reference to the experimental data, it can be predicted that further refinement of the mesh would improve the results, because in contradiction to the 2D ALE model (section 3.2) the refinement itself does not cause disturbances of the time histories for reflected overpressure.

EMPIRICAL BLAST LOADS

In this approach, air blast pressure is computed with empirical blast equations and directly applied to Lagrangian elements of the structure. The method takes an advantage of well calibrated empirical equations derived using a compilation of results from large number of explosive air blast experiments (Slavik 2009). Without modeling explicitly the air between the explosive and the structure this approach is computationally very effective, especially for global analyses when large stand-off distances and large scale structures are considered. The disadvantage of this approach is that it cannot capture interaction of the blast wave with any objects in front of the analyzed structure.

To determine the blast loading a user needs to select among several cases and define the charge weight and its position relative to the structure. Two different stand-offs were considered 400 mm and 1000 mm. For both cases the target square steel plate shown in Figure 1, was modeled with the fully integrated shell elements with two different mesh densities corresponding to element sizes of 25 25 mm and 5 5 mm. All nodes were fully constrained to imitate the rigidity of the plate. The loading was applied using load blast enhanced option defining the charge mass of equivalent to TNT, its distance from the target and conversion factors (in this study millimeters, tons and seconds were used). This option allows modeling the action of shock wave for different charge shapes. Unfortunately, the considered benchmark configuration, with the hemispherical charge (shock wave without reflection and reinforcement from ground), is currently not available in the program LS-DYNA . For the load blast enhanced option one can choose either spherical charge without reflection of shock wave, or hemispherical charge situated near to the ground. Shape of the charge in both cases has some impact on the results, as it is shown further. Comparison of the experimental (Pope & Tyas 2002) and the calculated results for different cases considered is presented in Table 1. Example time histories of the reflected pressure calculated for the spherical charge, 400 mm stand-off and three mesh densities are presented in Figure 12. Figure 12 shows that for the shell element sizes 5 mm

130

Figure 12. Time histories of reflected pressure for spherical charge, 400 mm stand-off, and for three different mesh densities.

Figure 13. Time histories of pressure for spherical and hemispherical charges and 400 mm stand-off.

400 mm was calculated as equal to 5.95 MPa in the case of hemispherical charge and 9.68 MPa for the spherical charge. For the stand-off 1000 mm the values are accordingly 0.41 MPa and 0.65 MPa. The corresponding experimental data gives 12.4 MPa and 0.7 MPa for 400 mm and 1000 mm stand-offs, respectively. The numerical results are affected by the fact that the empirical formula Load blast enhanced is set for TNT, and in the experiment the PE4 charge was used. These materials are quite similar, however detonation velocity of PE4 is higher than for TNT (Bulson 1997). Therefore the time, when the shock wave reaches the plate, may be slightly different in the experiment for the 400 mm stand-off the time equals to about 1,54 ms, whereas in the calculations it is 2 ms (for 1000 mm stand-off it is accordingly 8,56 ms and 10,03 ms). Also the pressure on the plate for the same mass ofTNT and PE4 is smaller in the case of TNT, however it is also predictable and according to Weckert & Anderson (2006) for the peak pressure the equivalence ratio between TNT and PE4 is 1.37. If the mass of the charge is multiplied by 1.37 the peak pressure reaches 12.51 MPa for 400 mm standoff and 0.90 MPa (0.55 MPa for the spherical charge) for 1000 mm stand-off (see Table 1). The effective impulses obtained for the mass of charge multiplied by 1.37 are also presented in Table 1. For 400 mm stand-off the effective impulse is equal to 0.54 kPams for the hemispherical charge and 0.33 kPams for the spherical charge, while the experimental value equals to 0.38 kPams. For the 1000 mm stand-off there are accordingly 0.16 kPams and 0.10 kPams, with the experimental value equal to 0.13 kPams. These results seem to be quite reasonable comparing to the experiment. 5 COMBINED METHOD

Figure 14. Contours of pressure applied to the target palate and generated by option load blast enhanced.

and 10 mm the estimated blast loading applied to the target plate in terms of the overpressure is almost identical. The differences between the results calculated for the same shell element sizes 10 10 mm and for the hemispherical and the spherical charges and 400 mm stand-off, are presented in Figure 13. The distribution of the blast loading generated by the option load blast enhanced and applied to the target palate is shown in Figure 14 as contours of pressure. The peak reflected overpressure generated in the middle point of the target plate for the stand-off

In the new, combined method, only the air immediately surrounding the Lagrangian structure needs to be explicitly modeled with the ALE mesh. The blast pressure calculated using empirical blast equations (section 4) is directly applied to the most outer face of the ALE air domain. In this way effects such as focusing and shadowing, caused by interfering objects immersed in the ALE mesh, can be captured. The new method is supposed to accommodate a good balance between accuracy and computational efficiency. In the case described below the same test setup as presented in Figure 1 was applied with a 78 g spherical TNT charge at stand-off of 400 mm. The calculations were conducted for different ALE mesh densities: 20 20 20 mm, 10 10 10 mm and 5 5 5 mm. Figure 15 shows the FE model with the ambient layer positioned 200 mm from explosive and the ALE mesh for air. The rest of the air between the charge and the ALE mesh and the charge itself are not explicitly modeled. In this way the model has substantially reduced number of elements. The empirical blast load is applied directly to the outer face of the ALE mesh through the ambient layer.

131

Figure 15. FE model for combined method. Table 2. Comparison of peak reflected pressure and arrival time for experiment and combined method. Stand-off Peak pressure [MPa] experiment elem. size 20 mm elem. size 10 mm elem. size 5 mm Arrival time of shock wave [ms] experiment elem. size 20 mm elem. size 10 mm elem. size 5 mm 400 mm

Figure 17. Time histories of reflected pressure for spherical charge, 400 mm stand-off, and for three different mesh densities.

12.4 1.58 2.59 3.52 1.52 1.55 1.68 1.80

Although the increased mesh density improves qualitatively the results the peak values are still substantially underestimated. The early stage results suggest that the finer ALE meshes are required to get comparable peak pressures for short stand-offs (compare sections 3.2 & 3.3). 6 CONCLUSIONS

For all computational approaches with the ALE formulation the results appeared to be highly sensitive to mesh density especially for the peak values of the reflected overpressure. The numerical estimation is better for longer stand-offs and is more precise for the effective impulses than for the peak values of pressures. The effective impulse is considered as the most important damage causing factor of the blast.Although several discrepancies were indicated, the premature validation study shows big potential for this type of numerical modeling. REFERENCES
Bulson, P.S. (1997) Explosive Loading of Engineering Structures, E & FN SPON. Donea, J. 1980. Advanced Structural Dynamics, Applied Science Publishers LTD, London UK. Hallquist, J.O. 2009. LS-DYNA Keyword Manual,Version 971. Livermore: Livermore Software Technology Corporation. Pope, D.J. & Tyas, A. 2002. Use of hydrocode modelling techniques to predict loading parameters from free air hemispherical explosive charges. In 1st Asia-Pacific Conference on Protection of Structures Against Hazards, Singapore, November 2002. Slavik, T.P. 2009. A Coupling of Empirical Explosive Blast Loads to ALE Air Domains in LS-DYNA , 7th European LS-DYNA Conference 2009. Wecker, S. & Anderson, Ch. 2006. A Preliminary Comparison Between TNT and PE4 Landmines , Weapons System Divisoin, DSTO Defence Science and Technology Organisation

Figure 16. Iso-surfaces of pressure for the stand-off of 400 mm modeled with combined method.

The comparison of peak reflected pressure and arrival time for experiment and combined method is presented in Table 2. Figure 16 presents snapshots of pressure iso-surfaces showing propagation of the shock wave through air modeled as the ALE domain. The time histories of reflected pressure calculated for different mesh sizes are compared in Figure 17.

132

S-ar putea să vă placă și