New Hybrid Metamodel-based Global Optimization Methods And Applications In Vehicle Design

Metamodel techniques are widely used in solving computation-intensive black-box problems for its fast computation. Metamodel based global optimization methods are also used more and more in search of the optimum for a design. For the modelling techniques, metamodels are case-dependent and just can be used in a typical field. In addition, its accuracy is not good when few points are used for metamodelling. For the shortcomings mentioned above, single metamodel based global optimization methods are hard to be applied in real engineering. So, a widely used optimization method with high efficiency and accuracy is a very useful tool in Engineering.Several different metamodels are employed together in this work and the achievements are shown as follows:1. Hybrid and adaptive metamodel-based global optimization method (HAM) is developed. Three different metamodels with different characteristics are employed and integrated in this method, which enlarge its application fields compared with single metamodel based optimization methods. In addition, the appropriate metaodel can be selected automatically from the alternative metamodels in solving unknown problems, increasing the efficiency. The employed three metamodels can be adaptively updated in optimization process and the accuracy in the focused area will be gradually increased, which make it be able to solve high-dimensional problems. The proposed important region strategy further increases its accuracy and efficiency.2. Design space differentiation method is proposed, which can solve the problems with large number of variables and large design space. It further increase the performance based on previously introduced HAM method. In this method, the search process will be done in constructed key region similar in every iteration. Besides, the whole design is also searched to avoid being trapped in local minimum. Compared with HAM method which searched in key region every two iterations, the efficiency of DSD is noticeably increased. In addition, space expansion strategy can enlarge the key region when it is too small, which can further avoid being trapped in local minimum for the excess reduction of the key region.3. Pareto frontier pursuing method (PFP) is proposed in solving multi-objective optimization (MOO) problems, which is developed for computation-intensive, black-box problem with several objectives. PFP method is an alternative in solving engineering problems, which better integrates the concept of HAM method and the strategy for MOO problems and acceptable Pareto points can be provided. In this method, selection strategy successfully keeps the useful points in cheap points, which can make the selected points uniformly distributed in Pareto frontier. Grouping strategy successfully differentiates the points of different importance, and sorting strategy makes the most important point firstly selected. The proposed method has the ability in solving MOO problems with high efficiency inherited from HAM.4. The proposed HAM method is applied in solving real engineering problems. With HAM method, the constrained system of one car is optimized, and all the parameters meet the requirements in national regulation, which The HIC value decreased from1124to803。To ensure that the head of the dummy do not contact the hub of the steering wheel, retractor is added to the system. The real experiment demonstrates this design, which the HIC value of the dummy decreased from1124to712and the displacement of the chest is38.5mm.