Research On RBDO Methods Based On Improved RBF And Adaptive Sampling

Nowadays with the increasing of high reliability requirements, Reliability-Based Design Optimization(RBDO) methods that give full consideration to the uncertainty of product design is becoming the research hotspot. In engineering application, the probability constraints are always implicit, where time-consuming computer simulation technology is needed to get the corresponding output response after the design variables are determined. To improve this situation, the strategy of replacing the implicit probability constraints by surrogate models is introduced into the RBDO framework by more and more researchers.This dissertation mainly studies RBDO which based on the Radial Basis Function(RBF) meta-model and adaptive sampling method. In view of the poor robustness of existing RBF meta-model that is constructed only by a single kernel function, a new type of RBF model called MK-RBF model is put forward which can adaptively combine the commonly used kernel functions in RBF model and take full use of the advantages of them. For engineering optimization problems that simulation cost is expensive and samples are difficult to obtain, an adaptive sampling approach called Improved Maximin Distance Adaptive Sampling(IMMDAS) is proposed. The Sequential Optimization and Reliability Assessment(SORA) method combined with the MK-RBF meta-model technique and the IMMDAS approach is used to improve the accuracy and efficiency of RBDO. This method also is applied to optimize the drive shaft of the all-direction propeller. The result shows its excellent effect in engineering application.Firstly, for the defection that different kernel functions are suitable for different problems, a new RBF model called MK-RBF model is proposed which makes full use of the advantages of most of the common used kernel functions. With the help of heuristic weight calculation method, the reasonable combination of each kernel function is implemented in MK-RBF model, which is more accurate and robust than the original RBF model.Secondly, for engineering optimization problems that simulation cost is expensive and samples are difficult to obtain, an adaptive sampling approach called IMMDAS is proposed. Moreover, the proposed MK-RBF meta-model and the IMMDAS approach combined with the Sequential Optimization and Reliability Assessment(SORA) method is used to solve the RBDO problems. The accuracy of the meta-models of the probability constraint functions in the local area of the current design point has great influence on the accuracy of RBDO solution. The IMMDAS method takes this into account and can obtain high accuracy in that area with only a few samples. Therefore, it can make the best use of samples and greatly improved the solving accuracy and efficiency of RBDO problem.Finally, the RBDO model based on the stiffness and fatigue life constraints of the drive shaft of the all-direction propeller is established. The method that combined the above three strategies is applied to solve this RBDO engineering problem. Compared with other methods, the simulation costs of the drive shaft are decreased greatly, which also verifies the effectiveness of the proposed method.