Analytical and experimental comparison of probabilistic and deterministic optimization

Abstract

The objective of this paper is to evaluate experimentally the gain in reliability that can be achieved In the design of a system by using probabilistic optimization instead of deterministic optimization. The design of a truss structure, equipped with tuned dampers and subject to constraints on the dynamic response amplitudes, is used for the investigation. The locations of the dampers and of tuning masses added to the truss are used as design variables. The properties of the dampers are the main sources of uncertainties. The deterministic and probabilistic optimizations are performed with genetic algorithms. Probabilities of failure are measured in the laboratory by testing multiple realizations of deterministic and probabilistic optimum designs. Predicted and measured values of the reliabilities of the concurrent designs are compared. For a design problem selected to maximize the contrast in reliability between the two optimum designs, the probabilistic optimization provided a large improvement in reliability compared with the deterministic optimum design. This improvement was confirmed by experiments.