@misc {pub4549,
	title = {Research on Crashworthiness and Cooperative Topology Optimization at HRI-EU},
	author = {Mariusz Bujny},
	year = {2020},
	month = {November},
	abstract = {Driven by rising competition on the market, strict emission regulations, and high customer expectations regarding vehicle safety, the complexity of the development process in automotive industry is constantly growing and leads to unintuitive design solutions accounting for a large number of load cases in multiple disciplines, different material types and manufacturing processes, and cost reduction by increasing commonality. In the conceptual design phase, Topology Optimization (TO) methods can support this complicated process by automatically generating optimal structures for a given set of loading conditions, within a predefined design space. At Honda Research Institute Europe (HRI-EU), we conduct research in the domain of TO based on two pillars:
1). Development of non-gradient TO methods for addressing crash scenarios, utilizing Evolutionary Algorithms (EAs) and Machine Learning (ML) techniques.
2). Building cooperative TO systems, where a human designer can influence the optimization process to a much greater extent than in standard methods.
 
In scope of the first research pillar, we address the fundamental problem associated with the unavailability of analytical sensitivities in crash TO. We propose methods relying on low-dimensional design representations, which allow for an efficient utilization of non-gradient optimization techniques, such as EAs or Bayesian Optimization. Furthermore, we employ ML techniques to reduce the computational costs via constructing models of sensitivities, predicting structural performance, and training models of favorable topological variations. The proposed methods successfully address highly non-linear crash problems that cannot be directly solved with the state-of-the-art methods. Finally, the generality of the proposed approaches allows for their application to other problems in structural mechanics, where analytical or numerical sensitivities are difficult to obtain.
 
The second research pillar focuses on developing TO methods using Cooperative Intelligence (CI) concepts, which allow for a much closer human-machine collaboration by taking into account additional input from the designer. Our contributions are threefold: Firstly, our proposed methods allow for an intuitive definition of relative importance of different load cases in multi-disciplinary optimization scenarios. Secondly, material preferences can be addressed using multi-material TO techniques, which come up not only with an optimal distribution of material, but also divide the structure into different material types. Thirdly, using novel, similarity-based TO techniques, the user can define a reference structure and flexibly adjust its similarity level to the topology being optimized, thus accounting for manufacturing and assembly limitations, as well as commonality aspects, or even search for creative solutions considerably different from the existing design concepts. Finally, the set of many complex design alternatives generated using the techniques described above can be conveniently analyzed using methods from Artificial Intelligence (AI), including clustering and dimensionality reduction techniques, which provide an indication for the further preference adaptations. Based on numerical experiments, we show the usefulness of the proposed methods for addressing large-scale industrial problems.},
	publisher = {No publisher just presentation},
	booktitle = {Meeting at TU Delft (virtual)}
}