Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 FLUID STRUCTURE INTERACTION OF FUEL TANK IMPACT Caroline M. SEDDON, Moji MOATAMEDI * L Stress Analysis Research Group, Institute for Materials Research, The University of Salford, Greater Manchester, M5 4WT [email protected] 1 Introduction The High speed projectile impact upon structures can lead to severe structural damage resulting in catastrophic failure, often causing huge loss of life. Aircraft structures are particularly vulnerable to high speed projectile impacts, resulting from uncontained engine debris following engine failure, objects encountered during flight such as bird strike or hail and impacts with runway debris during take off or landing. One particular area of concern is the impact of such high speed projectiles upon the aircraft fuel tank. Several catastrophic accidents have occurred in which high speed projectiles have directly impacted the fuel tank of the aircraft. In 1985 an accident occurred in which uncontained engine debris impacted the fuel tank of a Boeing 737 during take off at Manchester International Airport [1]. The impact caused rupture of the fuel tank causing a fire within the cabin. In 1995 uncontained engine debris impacted the fuselage of a McDonnell Douglas DC-9 causing failure of the fuel line [2]. This failure led to a cabin fire causing serious injuries to the passengers and crew. The most recent accident involving fuel tank impact was the incident involving the Air France Concorde at Paris Charles De Gaulle Airport in July 2000. This particular accident appears to have resulted from a fuel tank rupture caused by an impact from a piece of burst tyre. The subsequent accident investigations have suggested that the projectile did not penetrate the fuel tank upon impact causing rupture. It was however, suggested that the initial impact occurred between two adjacent wing ribs resulting in the fuel tank skin initially deflecting inwards and causing the adjacent panel to deflect outwards. At the same instant a strong hemispherical shock wave resulting from the impact was transmitted through the fuel, its movement combining with the outwardly deflecting panel inducing stresses sufficient enough to cause a localized failure of the tank. In the mid 1990’s research concerning the impact of aircraft fuel tanks was undertaken by the Federal Aviation Authority [3]. This study highlighted the poor recognition of the fuel tank failure modes in aircraft accidents for which fuel tank rupture had been identified. The subsequent investigations, however, assumed that the projectile impacts the fuel tank with sufficient kinetic energy to puncture and penetrate the tank causing hydrodynamic ram, an overpressure produced by the motion of the projectile within the fuel. In 1999 further work was undertaken by Anderson et al [4] which considered the numerical modelling of hydrodynamic ram, again assuming that the projectile fully penetrates the tank causing overpressure. Recent research undertaken specific to aircraft structures has mainly been concerned with the numerical modelling and validation of the failure of flat skin panels upon high speed impact [5-7]. It is understood that the probable failure mechanism discovered in the Concorde accident has yet to be fully confirmed both experimentally and numerically. The suggested failure mechanism of transient fluid structure interaction has relevance to many other industries such as rail, chemical, nuclear and offshore industries, in which risks are posed to safety critical structures. It is therefore becoming increasingly important to be able to model and predict such phenomenon. Analytical solutions can provide only a limited understanding of the nature of the behaviour; therefore it is necessary to investigate the phenomenon experimentally and to assess both the suitability and predictive capabilities of numerical modelling techniques such as finite element analysis for the analysis of transient fluid structure interaction. Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 The work reported here describes the development of an experimental test rig and test procedures loosely modelling the Concorde accident. Next a numerical modelling technique utilising full fluid structure interaction is described and is applied in the modelling of the experiment. Three cases were examined in which a full tank was impacted at 20 m/s and a partially full tank was impacted at 14 m/s and 21m/s. The experimental and numerical results are compared in order to assess the validity and accuracy of the numerical modelling methodology. 2 Expriment Test Facility The Health and Safety Laboratory experimental set-up consisted of a gas gun facility, the main parts of which can be seen in Fig. 1. The gas gun was 6m long with a smooth bore 75mm internal diameter barrel. The barrel was manufactured from stainless steel tube having a theoretical burst pressure in excess of 30 MPa, and was hydraulically pressure tested to 8 MPa. . FIG. 1 – 75mm Shock Tube Arrangement The gas gun was operated by charging the 2.5m long driver section with compressed air up-to maximum pressures of 3.8 MPa, when a commercially available bursting disc, situated at the end of the driver section, failed discharging the air into the barrel of the gas gun. To achieve lower pressures of approximately 1 MPa the bursting discs were replaced with a pneumatically actuated 50mm ball valve. In order to ensure a high degree of safety the projectile was fired through a trap consisting of a length of 20 cm diameter perforated steel pipe. This prevented the projectile from escaping into the room housing the gun, if the projectile should recoil from the test vessel. The perforations allowed for the use of high speed photography, providing a limited visualization of the projectile impact, and also provided some relief from the muzzle blast. The instrumented test vessel including the perforated section of the gas gun can be seen in Fig. 2. FIG. 2 – Instrumented Test Vessel and Gas Gun Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 Projectiles and Velocity Measurement The original test specification required projectiles weighing approximately 4.5 kg to provide continuity with the Concorde accident. The weight was, however, reduced to 2.5 kg as even at the lowest velocities the heavier projectiles caused failure of the tank. The projectiles were machined from cast nylon-6 rod for which a detailed list of material properties was available. Following machining the projectiles were 50cm long with a diameter of 74mm. The impact velocities of the projectiles were recorded using an optical system based upon a split laser light beam which was directed through two holes in the gun barrel near to the muzzle. The light beam was detected by two photodiodes for which the outputs were monitored by an electronic counter. The velocity of the projectile was then determined from the time taken for both light beams to be broken. For some of the test cases the velocity was also determined from high speed video recordings of the tests However, there were discrepancies between the two measurement techniques. The reasons for the discrepancies were not established. Test Vessel The test vessel comprised of a rectangular steel tank of internal dimensions 2 x 1 x 1 m. The vessel was constructed from 8mm mild steel and was designed to have rigid sides with flexible front and back panels in order to simplify the later numerical modelling. The flexible front and back panels were secured with bolts enabling interchanging of the panels, allowing a number of materials / thicknesses to be tested. The vessel was stood vertically a short distance from the gas gun (see Fig 2.) with one of its square faces secured to the ground, to enable the impact response of the flexible front and back surfaces to be determined. The close proximity of the test vessel to the gas gun ensured that a planar impact occurred. The vibrations arising during the impact were measured with Kistler piezoelectric accelerometers. The pressures within the fluid were measured using 0-250 bar Kistler piezo-resistive pressure transducers. P3 P2 Acc B Face C (rear) Face A (front) P1 Impact Point Acc A y z x (0, 0, 0) Face B Impact Point 597, 560, 0 (Face A) Accelerometer A 380, 560, 0 (Face A) Accelerometer B 560, 1440, 0 (Face A) Pressure Transducer 1 0, 1970, 500 (Face B) Pressure Transducer 2 200, 1593, 900 (suspended inside tank) Pressure Transducer 3 910,563,900 (suspended inside tank) FIG. 3 – Test Vessel Instrumentation In early tests all of the transducers were attached to the tank walls, however, in later tests two of the pressure transducers were suspended inside the tank in an attempt to de-couple the extreme vibrations. The Final test vessel instrumentation can be seen in Fig. 3. The test data was obtained using a 4000 series micro-link system configured to a logging rate of 200 kHz. The data collection commenced upon the breaking of the first light beam used in the velocity measurement. Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 3 Numerical Modelling Fluid Structure Interaction The numerical models were developed using LS DYNA explicit dynamics software [8]. This particular software package was selected due to the use of explicit formulation which is particularly suitable for problems with short duration such as impact or explosion. The programme also enables the modelling of fluid structure interactions using Arbitrary Lagrangian Eulerian ALE formulation with a fluid structure coupling algorithm. Typically structural finite element analysis uses Lagrangian formulation. In pure Lagrangian finite element analyses the computational mesh moves with the material, automatically following the material deformation. This method is only efficient and accurate for small to moderate deformations. In the case of large deformation problems the elements become highly distorted, reducing both accuracy and the stable time step size of the solution. In the case of the fuel tank impact problem if both the fluid and the structure are modelled using a Lagrangian formulation the fluid elements in the impact region become severely distorted, reducing the stable time step size, significantly increasing the CPU time. An alternative method is employed within LS-DYNA in which a single or multi material Eulerian formulation can be used to eliminate mesh distortion in the fluid. In an Eulerian formulation the finite element mesh is fixed in space and the material is able to flow through the fixed mesh hence eliminating the problems associated with mesh distortion. However, as the material is transported from one element to its neighbour nonlinear transport terms are introduced into the problem. The transport of mass, momentum, internal energy and state variables must be computed using advection algorithms such as the donor cell algorithm, a first order scheme or the Van Leer algorithm a second order advection algorithm. Within the LS DYNA programme the ALE formulation is implemented using an approach referred to as an Operator Split. Firstly a Lagrangian step is performed in which the finite element mesh deforms with the material. In the second phase, the advection step, the solution is mapped from the distorted Lagrangian mesh back onto the original mesh for a single or multi material Eulerian formulation. The solution can also be mapped onto an arbitrarily distorted mesh for an ALE formulation. In the case of the fuel tank impact the fluid inside the tank can be modelled using an Eulerian formulation for which the computational mesh is fixed. The tank structure can be modelled using a Lagrangian formulation as the mesh deformation is not too severe. Euler Lagrange coupling is then employed to enable the modelling of the interaction between the Lagrangian structure mesh and an Eulerian fluid mesh. The fluid structure coupling algorithm implemented within the LS DYNA programme allows the fluid material to flow around the structure but prevents penetration of the fluid material into the structure mesh. The flow of the fluid is prohibited from flowing through the structure using a penalty coupling algorithm. In this algorithm penalty forces are applied to both the fluid and the structure. As soon as a fluid particle penetrates into a Lagrangian structure a force of recall is applied to both fluid particle and structural node in order to return the fluid particle to the surface of the structure hence preventing penetration. The penalty forces are proportional to the penetration depth and penalty stiffness, behaving like a spring system. Application to Fuel Tank The previously described fluid structure coupling algorithm implemented within the LS DYNA programme was used to model the Fuel Tank Experiment. Figure 4 below shows a cut –away view of the finite element mesh for the case of the full tank. The fluid inside the tank is modelled using solid elements with an Eulerian element formulation. The material is modelled using a null material model with a Gruneisen Equation of State with the properties of water. Surrounding the fluid material is a region of air. This region acts as vacant cells into which the fluid material can flow during the impact. This region was also modelled using an Eulerian formulation, null material model with a linear Polynomial Equation of State and was assigned the properties of Air. In order to allow the water to flow into the air region during the impact the two meshes must share common Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 nodes at their interface. Next the tank structure mesh is superimposed upon the Eulerian fluid mesh. The Tank was modelled using Lagrangian shell elements and was assigned the material properties of steel. The projectile was modelled with rigid solid elements again with a Lagrangian formulation and was assigned the material properties of nylon. Eulerian Fluid Mesh Eulerian Air Mesh Lagrangian Structure Mesh superimposed within fluid FIG. 4 – Tank Finite Element Model – Cut Away View The nodes on the base of the tank were fixed in all degrees of freedom as in the experiment. The projectile was modelled initially touching the tank surface at the point of impact. The contact between the projectile and the tank was modelled using a surface to surface contact interface. The fluid structure interaction algorithm was employed, coupling the inside surface of the tank to the fluid particles inside the tank, preventing the fluid from penetrating through the tank walls. The analysis was run for 0.16 seconds with a time step size of 5 micro-seconds which corresponded to the data capture time and the 200 kHz sampling rate of the data logger used in the experiment 4 Results Figure 5 below shows the experimental and numerical acceleration time history at the point of impact for the partially full tank impacted at 14m/s 3.0E+04 2.5E+04 2.0E+04 Acceleration 1.5E+04 1.0E+04 5.0E+03 0.0E+00 -5.0E+03 -1.0E+04 -1.5E+04 -2.0E+04 Experim -2.5E+04 -3.0E+04 0.00E+00 4.00E-02 8.00E-02 Tim 1.20E-01 Numerica 1.60E-01 Time FIG. 5 – Acceleration Time History at the Point of Impact (ACC A) for Partially Full Tank 14m/s impact From Fig 5. it can be seen that the numerical model accurately predicts the peak acceleration at the point of impact. Good agreement can be seen between the experimental and numerical results; however, the numerical results exhibit more noise than the experimental results. The addition of damping to the numerical model may decrease the oscillations seen later in the solution. Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 Figure 6 below shows the experimental and numerical acceleration time history at the top of the impacted panel (Acc B) for the partially full tank impacted at 14m/s. 1.40E+04 1.20E+04 8.00E+03 Acceleration Acceleration 1.00E+04 6.00E+03 4.00E+03 2.00E+03 0.00E+00 -2.00E+03 -4.00E+03 -6.00E+03 -8.00E+03 -1.00E+04 0.00E+00 Numeric Experime 4.00E-02 8.00E-02 Time 1.20E-01 1.60E-01 Time FIG. 6 – Acceleration Time History at the top of the impacted panel (ACC B) for Partially Full Tank -14m/s impact Good agreement is again seen between the experimental and numerical results for the top of the impacted panel. As expected the peak accelerations are much less than at the point of impact. Again the numerical results display a more oscillatory nature particularly towards the end of the solution. Numerical damping would improve the overall decay of the curve. The peak acceleration measured in the experiment has been cut off above 5000 m/s2 due to the range of the accelerometer selected. However, the overall shape of the remainder of the curve shows good agreement with the numerical model. Figure 7 below shows the experimental and numerical acceleration time history at the point of impact for the partially full tank impacted at the higher speed of 21 m/s FIG. 7 – Acceleration Time History at the Point of Impact (ACC A) for Partially Full Tank 21m/s impact Again good agreement can be seen between the peak experimental and numerical acceleration results for this higher impact speed. Again the numerical results would benefit from some damping. Figure 8 shows the experimental and numerical results for the second accelerometer at the top of the impacted panel. As with the lower speed test the peak experimental acceleration is cut off at ±5000 m/s2. The actual peak acceleration was therefore not captured during the experiment. As with the previous results the numerical results appear to be much more oscillatory than the experiment results. Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 In comparison to the earlier case in which the projectile speed was lower there is not a great difference in the peak acceleration, however, the numerical results are slightly more oscillatory for the higher speed case than for the lower speed impact. 1.4E+04 1.2E+04 1.2E+04 1.0E+04 1.0E+04 Acceleration 1.4E+04 Acceleration 8.0E+03 6.0E+03 4.0E+03 2.0E+03 0.0E+00 8.0E+03 6.0E+03 4.0E+03 2.0E+03 0.0E+00 -2.0E+03 -2.0E+03 -4.0E+03 -4.0E+03 -6.0E+03 -6.0E+03 -8.0E+03 -1.0E+04 0.0E+00 4.0E-02 8.0E-02 1.2E-01 Numerica -8.0E+03 Experim 1.6E-01 -1.0E+04 0.0E+00 4.0E-02 8.0E-02 1.2E-01 1.6E-01 Time FIG. 8 – Acceleration Time Time History at the top of the impacted panel (ACC B) for Partially Full Tank 21m/s impact Figure 9 shows the acceleration time history at the point of impact (Acc A) for the final case in which the full tank is impacted at 21m/s. It can be seen that the measured peak acceleration has been cut off at ±20000 m/s2. The peak has been lost due to the incorrect selection of the accelerometer range. Despite the loss of peak it can be seen that the predicted numerical peak was lower than the measured peak acceleration. In this case the experimental results display more noise than the numerical results. Figure 10 shows the results for the top of the impacted panel from the final case in which the tank was completely full and impacted at a speed of 20m/s. Again from the experimental results it can be seen that the incorrect accelerometer range was selected for the test as the peak acceleration has been lost. However, the remainder of the curve shows good agreement with the numerical results. In this case the numerical results are slightly noisier than the experimental results. FIG. 9 – Acceleration Time History Acc A Full Tank 6.0E+03 Acceleration 5.0E+03 Acceleratio 4.0E+03 3.0E+03 2.0E+03 1.0E+03 0.0E+00 -1.0E+03 -2.0E+03 -3.0E+03 -4.0E+03 Experime -6.0E+03 0.0E+00 Time (s) Numeric -5.0E+03 4.0E-02 8.0E-02 FIG. 10 – Acceleration Time History Acc B – Full Tank 1.2E-01 Time (s) 1.6E-01 Résumé pour le 1ercolloque du GDR interactions fluide-structure – 26-27 sept. 2005 5. Conclusions In conclusion the projectile impact of a fluid filled tank has been investigated experimentally and numerically. The design of the large scale experiment was described and results were presented for three cases in which a partially full tank was impacted at two projectile speeds, 14m/s and 21m/s, and a full tank was impacted at 20m/s. A numerical modelling procedure was described in which full fluid structure interaction using a penalty coupling algorithm was employed to couple the fluid particles to the structural nodes. The numerical method was used to model the fuel tank experiment. Upon comparison of the experimental and numerical results it was shown that the peak acceleration and overall shape of the acceleration time history has been predicted fairly accurately by the numerical model. In some cases the experimental peak was not captured due to incorrect selection of the accelerometer range during the experiment. In such cases the remainder of the curves within the accelerometer range show good agreement with the numerical model. In most cases the numerical results were more oscillatory than the experimental results. It is thought that either using material damping or damping within the fluid structure coupling algorithm will improve the numerical response. Future work will include modelling the pressure time history and comparing the results with the experiment. ACKNOWLEDGMENTS The authors would like to thank Health and Safety Laboratory Buxton for the use of the experimental results. References [1] UK Aircraft Accident Investigation Board, 1985, “Report on the Accident to Boeing 737-236, G-BGJL at Manchester International Airport on 22 August 1985”, Aircraft Accident Report form the British Department of Transport No 8/88 (EW/C929). [2] National Transportation Safety Board, 1995, “Uncontained Engine Failure/Fire ValuJet Airlines Flight 597, McDonnell Douglas DC-9-32, Atlanta Georgia, Aircraft Accident Report NTSB ATL95MA106. [3] Moussa, N. A., Whade, M. D., Groznam D. E., et al., 1997, “The potential for Fuel Tank Fire and Hydrodynamic Ram from Uncontained Aircraft Engine Debris” DOT/FAA/AR-96/9, US Department of Transportation Federal Aviation Authority. [4] Anderson, C. E., Sharron, T. R., Walker, J. D., et al, 1999, “Simulation and analysis of a 23mm HEI Projectile Hydrodynamic Ram Experiment” International Journal of Impact Engineering, 36, pp. 981-997. [5] Knight, N. F., Jaunky, N., Lawson, R. E., et al, 2000, “Penetration Simulation for Uncontained Engine Debris Impact on Fuselage like Panels Using LS-DYNA.”, Finite Elements in Analysis and Design, 36, pp. 99-133. [6] Ambur, D. R., Jaunky, N., Lawson, R. E., et al, 2001, “Numerical Simulations for High Energy Impact of Thin Plates” International Journal of Impact Engineering, 25, pp. 683-702. [7] MacDonald, B. J., 2002, “A Computational and Experimental Analysis of High Energy Impact to Sheet Metal Aircraft Structures.” Journal of Materials Processing Technology, 124, pp. 92-98 [8] Hallquist, J. O., 1998, “LS-DYNA Theoretical Manual”, Livermore Software Technology Corporation, Livermore.
© Copyright 2026 Paperzz