| In the last decade Evolutionary Algorithms have advanced to efficient and reliable problem solvers for single- and multiobjective nonlinear optimization tasks. Advanced features, like de-randomized self-adaptation of strategy variables, niching, the incorporation of online-learned landscape models (metamodels), and robust parallelization strategies contributed to this. However, there is still a gap between theory and practice that prevents the widespread use of such techniques in industry. Industry?s main needs are robust design optimization tools, i.e., tools that can deal with various uncertainties that are intrinsic to the production processes, such as vagueness of the problem definition (i.e. fuzzy constraints), and noise on the input and output variables of the system to be optimized. The goal of this proposal is to adapt advanced evolutionary algorithms for the task of robust design optimization. By adopting an integrated approach, the new methods will allow addressing these various problems in combination and simultaneously. The Natural Computing Group of LIACS has a long standing experience in algorithm design and performance assessment. Moreover it has many cooperation partners in academia, scientific institutes and industry, offering it the unique opportunity to combine theoretical and practical research. Hence the proposed research will include theoretical parts, such as the establishment of performance metrics and algorithm analysis based on stochastic process and order theory, as well as a practical part, where the new methods will be validated on real-world applications, such safety design for the automotive industry based on crash-test simulations and design of molecule alignment by pulsed lasers based on physical experimentation. |
