@misc {pub3321,
	title = {Topology Optimization of Crash Structures with the Hybrid Evolutionary Level Set Method},
	author = {Mariusz Bujny AND Nikola Aulig AND Markus Olhofer AND Fabian Duddeck},
	year = {2017},
	month = {June},
	abstract = {Topology optimization plays an important role in many engineering fields, including crashworthiness. In most of the crash topology optimization methods, very strong simplifications of the underlying problem are made and often heuristic approaches are used. This makes the optimality of the obtained topologies arguable and limits the applicability of those methods just to selected use cases. Presented in previous works of the authors, Evolutionary Level Set Method, enables to solve the optimization problem directly, based exclusively on nonlinear, dynamic FE simulations of the crash problem. In spite of many advantages of the proposed method, including very good global search properties in highly nonlinear and noisy optimization landscapes and flexibility with respect to the use of different objectives and constraints, the computational costs are still very high. This can be to some extent mitigated through parallelization of the computations, which can be done very efficiently for Evolutionary Algorithms. Nevertheless, a further reduction of computational costs is inevitable. A hybridization of Evolutionary Algorithms with existing methods for crash topology optimization is one of the potential solutions of this problem. In this paper, a method taking advantage of the global search properties of Evolution Strategies enhanced with a local search based on a gradient information from the local Equivalent Static Loads Method is presented. To evaluate the proposed approach, a two-dimensional transverse bending crash case is considered. The performance of the proposed method for different objectives is compared with the standard Evolution Strategy. The results show that, an improvement in terms of convergence speed and performance of the obtained designs by the utilisation of predicted gradients in hybrid algorithms is possible. Thanks to the generality of the method, the approach can be further extended by introducing other estimated local gradients.},
	publisher = {International Society for Structural and Multidisciplinary Optimisation},
	booktitle = {12th World Congress of Structural and Multidisciplinary Optimisation}
}