Crashworthiness Optimization of Nested and Concentric Circular Tubes Using Response Surface Methodology and Genetic Algorithm

In this study crashworthiness optimization of nested and concentric circular tubes under impact loading is performed by coupling Finite Element model, Response Surface Models and Genetic Algorithm. Specific Energy Absorption SEA and Crash Force Efficiency CFE are used in crashworthiness optimization since these criteria are important indicators for evaluating crashworthiness performance. Length and thickness of three concentric tubes as well as radius of one tube are adopted as design variables which are effective parameters on SEA and CFE. To reduce the computational cost of the optimization procedure, simple and computationally cheap Response Surface Models are created to replace finite element analyses in further calculations. The Non‐dominated Sorting Genetic Algorithm –II NSGAII is applied to obtain the Pareto optimal solutions. Optimization results are presented for different selected designs that indicate relative importance of multi‐objective functions. Results show that the total weight of the vehicles can be reduced by using nested tubes comparing to single tubes with identical masses. These designs can be adopted for use in practice.


INTRODUCTION
Thin walled tubular structures are extensively used in automotive, aerospace and military industries due to their energy absorption capacity for the protection of the passengers, drivers and electronic devices of vehicle. Converting some amount of kinetic energy during collision to the plastic energy prevents passengers and devices from squeezing. Besides, inertial loads, which occur at the beginning of the collision, can be harmful for passengers during dynamic crash.
Axial crushing of thin walled circular tubes made of mild steel was introduced by Alexander 1960 as an excellent mechanism for energy absorption. Abramowicz and Jones 1984a, 1984b, and 1986 investigated the circular and square steel tubes under static and dynamic loading. These tubes absorb and convert large amounts of kinetic energy into plastic strain energy under severe loading conditions. In relation with continuous interest to energy absorption capability of thin walled tubes, Jones 1989Jones , 1996, and 2011 reviewed elasto-plastic behavior of structures under impact conditions. Kim 2002 firstly studied axial crushing behavior of multi-cell tubes. Zhang et al 2006 showed that the energy absorption of a single-cell tube can be increased by 50% when the section is divided into 3 3 cells and they investigated effect of different parts on the cross section. Najafi and Rais-Rohani 2011 made comparison between analytical and numerical results of axially loaded multi-cell tubes. Optimization procedure was successfully applied to increase the efficiency of many types of crush structures such as multi-cell tubes Kurtaran et al. 2002. For example, Hou et al. 2008 carried out multi-objective optimization of cell tubes by maximizing Specific Energy Absorption SEA and minimizing Peak Crush Force PCF values. They compared the result of 1-cell, 2-cell, 3-cell and 4-cell tubes with pareto curve. Yin et al. 2014a and 2014b investigated foam integration to multi-cell tube system. In different studies researchers suggested analytical solutions for mean crushing force curves of multi-cell tubes Zhang et al. 2006. Qiu et al. 2016 derived analytical solution for multi-cell tubes using Simplified Super Folding Element theory and optimized the cross-sectional dimensions of the hexagonal tubes using multi-objective optimization procedure.
Superior energy absorption capability of multi-cell tubes including nested multi-tubular structures are expressed in the literature. Usta et al. 2015 andTürkmen 2017 conducted parametric analyses to investigate the effects of the number of cross section and the number of nested tubes with gradually changing lengths on axial crushing performance of thin walled structures. An experimental work including quasi-static crush tests of these systems has been conducted by Nia and Chahardoli 2016a . Later on, they have applied optimization procedure to nested tubes through response surface approximation Nia and Chahardoli 2016b .
From the literature, it is seen that very limited study is available for crash optimization of nested tubes. Olabi et al. 2008 investigated the effects of the lateral impact on circular nested tube systems and show that optimized nested tubes gave desirable force deflection response compared to standard tube systems. Nia and Chahardoli 2016b optimized circular nested tube systems with different height and thickness subjected to quasi static crushing. In this study, multi objective optimization procedure is applied to optimize tri-tubular nested and concentric circular tube system by coupling Genetic Algorithm, Response Surface method and Finite Element FE model. Specific energy absorption SEA and Crush Force Efficiency CFE criteria are used in representing multiobjective objective functions and Peak Crush Force PCF in constraint function. Response surface models are developed for CFE and SEA of the nested tube structures. By using response surface models, nested tube structure is optimized for optimum crashworthiness performance. This study contributes to the literature an optimum design cluster of nested and concentric tubes regarding to SEA and CFE under dynamic impact loading.

FINITE ELEMENT MODEL
In this study, finite element model of nested and concentric circular tri-tubular system is used in obtaining SEA and CFE responses required in optimization. Nested and concentric circular tri-tubular system employed is shown in Figure 1. L 1 , L 2 and L 3 indicate length of tubes, t 1 , t 2 and t 3 denote thickness of tubes and r 2 indicates radius of middle tube.
Nested and concentric circular tri-tubular system is crushed in the axial direction by a 1132 kg rigid mass with a 4.22 m/s initial velocity. The bottom line nodes of all tri-tubes are clamped, while top nodes of the three tubes are constrained except axial direction due to provide deformation behavior. Tubes are modeled by using Belytschko-Lin-Tsay-4-node thin shell elements with three integration points through the thickness and one integration point in the element plane. The element size of 2 x 2 mm is found to be sufficient and suitable to simulate the crushing deformation of tubes 29160 elements, 29491 nodes . Automatic contact algorithms are activated to prevent penetration of tube surfaces after deformation. The static and dynamic friction coefficients are chosen as 0.2 and 0.3 respectively Bathe 2012 , Fang et al. 2014 . FE analysis is conducted using commercial code LS- DYNA Hallquist 2007 . In FE analysis, tube system is considered made of Aluminum 6063. Following material properties for Aluminum 6063 is used in this study. Density is ρ 2700 kg/m 3 , Young's modulus is E 68.2 GPa, initial yield stress is σ y 80 MPa, ultimate stress is σ u 173 MPa, and Poisson's ratio is ʋ 0.3. In order to validate the FE model, FE analysis result of the tri-tubular circular tubes under axial dynamical loading is compared with experimental results as shown in Figure 2. Experiment is conducted by using drop-tower test facility of the Scientific and Technological Research Council of Turkey TUBITAK . 1132 kg drop weight is released from a height that provides approximately 10 kJ. From comparison, it is seen that FE response is very close to the experimental result and therefore it can be concluded that FE model can be used in producing crush responses used in optimization.

OPTIMIZATION PROCEDURE
In order to evaluate the crashworthiness of the thin-walled structures, it is essential to define crashworthiness indicators. Energy Absorption EA , Specific Energy Absorption SEA , Mean Crushing Force MCF and Crash Force Efficiency CFE are usually used as the important indicators for evaluating crashworthiness performance Guler et al. 2010 . Energy absorption of a structure subjected to the axial loading can be expressed as: where δ is the axial crushing distance and P s denotes the axial crushing force. SEA is calculated as energy absorption divided by mass as below MCF is calculated as absorbed energy divided by axial crushing distance as CFE is calculated as mean crushing force P m divided by peak crush force PCF . CFE values close to one are desired for efficient designs.
Details of optimization steps are explained in the following sections.

Formulation of optimization problem
In this study, optimization of tri-tubular nested and concentric circular tube system is formulated as multiobjective constrained optimization problem. CFE and SEA criteria are used in expressing objective functions expressed as f 1 and f 2 successively. For optimum crashworthiness design, objective functions are desired to be maximized.
In multi-objective optimization procedure adopted in this study, multi objective particle swarm optimization MOPSO method is used. Considering multi-objective and constraint functions, optimization problem is formulated as below:

Response surface models for optimization criteria
In order to reduce the computational cost of the optimization problem with many design variables, simple approximate models are used to replace the costly original objective and constraint functions, which are calculated using expensive FE analysis. Simple approximate models are often referred to as Response Surface RS models and are obtained using Response Surface Method RSM . In this study, RS models are used for SEA and CFE criteria during optimization. RSM was originally developed for the model fitting of physical experiments by Box and Draper 1987 and later adopted in other fields.
In RSM, polynomial model is selected first to approximate the actual function. Often quadratic polynomial models are selected. Quadratic models can be written as Steps of RSM is summarized in Figure 3. Flowchart of optimization procedure adopted in this study is shown in Figure 4. In this study a subset of three-level Full Factorial DOE is selected using D-optimality criterion. Subset includes 45 experiments, which is sufficient enough to fit a quadratic function for 7 variables. To implement RSM, a MATLAB code is written.       Figures 5-6 it is seen that good correlation exists between RS model and FE results. Therefore, RS models can be used in the optimization for concentric circular tri-tubular system.

Solution of optimization problem
Optimization problem formulated in Equations 6 and 7 is solved using Genetic Algorithm GA . GA is coupled with the RS models to yield a global optimum as shown in Figure 7. The GA solves the optimization problem by simulating the biological evolution process based on Darwin's theory of survival of the fittest. First, a set of potential solutions referred to as population or chromosomes is selected. New and improved solutions are then generated using the previous solutions based on crossover and mutation technique. This process repeats until optimum values are reached. The critical parameters in GAs are the size of the population, cross over rate, mutation rate and number of generations. Detailed information about GA can be found in Ref. Goldberg 1989 . Figure 7: Coupling of FE Analyses, RS models and GA for crashworthiness optimization.
The Non-dominated Sorting Genetic Algorithm -II NSGAII is applied to obtain the Pareto optimal solutions Deb 2002 . Table 3 indicates the values of NSGAII parameters for considered problem which provides results with good repeatability.

Optimization results
In this section, multi-objective optimization results of concentric circular tri-tubular system are presented for different CFE and SEA values. Multi optimization method provides Pareto optimal sets for nested and concentric circular tubes. The Pareto optimal frontier provides a set of solutions according to CFE and SEA values, shown graphically in Figure 8. Solution of optimization problem expressed in Equations 6 and 7 is conducted with GA. Optimum values of design variables found are shown in Table 4. In Table 4, SEA and CFE values at optimum are compared for RS model and FE analysis. Undeformed and deformed shapes of optimum tube structures under impact for different SEA and CFE values, are demonstrated in Figure 9. In Table 4, it is seen that the values of optimum design lengths, thicknesses and radius change according to SEA and CFE values. The innermost tube tends to increase with the higher SEA and lower CFE from 150 to 200 mm. On the other hand, the length of outmost tube becomes lower for higher SEA and lower CFE. Besides, the thickness of each tube becomes higher values with the increase of SEA and reduction of CFE. The results prove that optimum designs show a tendency to reduce weight of the tube systems due to obtaining better SEA results. The radius of middle tube which is another design parameter becomes nearly at limit values 23 and 32 mm for each designs. Crash tubes are used traditionally as single tube structures in automotive industry. Therefore, results of optimum designs of nested tubes structures are compared with the results of two different types of single tube structures in Table 5. In Table 5, single tubes corresponding to a design number have identical mass values. In Single Tube Type 1, radius and length values correspond to average values in those of nested tubes shown in Figure  10. In Single Tube Type 2, radius and length values correspond to maximum values in those of nested tubes.
In comparison with the results, nested tubes give higher CFE values only for a few designs. The axial displacement and peak crash force values should be lower to obtain higher CFE values. It can be said that the stiffness of nested tubes should be increased to prevent more deflection in axial direction and the gaps between the tubes should be increased to reduce contact forces.
On the other hand, SEA values of each type are similar. SEA values of Nested Tubes are higher than Single Tube Type 1 and 2. The results show that nested tubes have more advantages in terms of SEA values. Therefore, total weight of the structure can be reduced by using nested tube structures.

CONCLUSION
In this study, optimization was carried out in order to improve the crashworthiness performance of nested and concentric circular tubes. SEA and CFE were considered as performance measure for crashworthiness. Specific Energy Absorption and Crash Force Efficiency were calculated using Explicit Finite Element analysis. For further calculations in optimization procedure, FE Analyses were replaced with RS models in order to reduce the computational cost. RS models for SEA and CFE were coupled with Genetic Algorithm to find optimum design variables and multi-objective optimization procedure was performed to maximize the objective function of SEA and CFE. NSGAII was applied to obtain the Pareto optimal solutions and the Pareto optimal frontier was plotted. Length and thickness of three concentric tubes as well as radius of one tube were considered as design variables.
25 different optimum crash tube models were obtained by using multi objective optimization method. SEA and CFE values of optimum designs of nested tube structures were compared with the results of two different types of circular tube structures having identical masses. From design point of view, it was observed that nested tubes had no advantages according to single tubes for CFE criterion. To increase the CFE values of nested tube structures, the stiffness of each tube and the gaps between nested tubes should be increased.
On the other hand, optimum designs gave better results than the single tube types in terms of SEA values. Therefore, nested tubes could provide lower weight to the vehicles.