A target cascading method using model based simulation in early stage of vehicle development 2019-01-0836
Vehicle specifications which satisfy various constraint conditions of target performance indices are required in the early stage of vehicle development. For this decision making, model based simulation plays important role to provide candidates of feasible specifications and can be used to find optimal specifications by applying optimization method. In order to enhance the efficiency of later development processes, those optimal specifications obtained from the optimization problem are required to be not only feasible but also flexible. To guarantee these requirements, it is important for decision maker to understand feasible region space and its mechanism where satisfies every constraint based on model based simulations. The purpose of this study is not to find best solution in design space but to understand feasible design space and its mechanism. A simple way to implement this study is to run many computer simulations based on the Design of Experiment (DoE) and extract feasible designs which satisfy target performances from the DoE results. However, it is not always straightforward to find details of feasible boundary with limited computer resources especially in design of complex and large system because many simulations are required to capture clear feasible regions. In this paper, an exploring method for feasible design space based on model based simulation is proposed by using global optimization method and active learning techniques. In this method, a surrogate model of feasible region is created using kriging model. Training data of the kriging model is chosen as a solution of inverse problem solved by global optimization method. The inverse problem to understand feasibility of specific point is sequentially defined based on the expected improvement index in kriging model.