The Application Research Of Approximation Theory In Hull Form Optimization

Multidisciplinary design optimization (MDO) is the technology for optimizationdesign of complex engineering system, has shown strong vitality and broad prospectsfor development. hull form optimization which is the combination of MDO and CFDis a hot research, as well as an effective way of energy saving. And it is performedmainly by hydrodynamic performance optimization at present, which has theproblems of optimal time and the global optimal results. It can improve theoptimization efficiency and solve these problems by using the approximation theoryto construct the approximation model of hydrodynamic performance to reduce theamount of calculation and the complexity of optimization model.The essence of approximation theory is the use of statistical theory andoptimization theory for summarizing the characteristics of complex problems andconstructing the explicit model of complex problems as the approximate model.The hull form optimization with approximation model could come true based on thehigh accuracy of approximate model. And the accuracy of approximation model isdue to the samples selection and the approximate model. According to different kindof variables space form, this paper mainly focuses on the methods of selectingrepresentative samples in design space to improve the utilization efficiency of samplepoints and construct the approximate model with high accuracy. Finally the sampleselection method proposed in this paper has been successfully applied to the hull formoptimization. The main research contents are as follows:1. Sample selection methods in cubic spaceThe variable space for the hull surface modification method of radial basisfunctions for is the cube space, and there are no constraints between variables. Forthis type of variable space, based on the comparing of existing methods, UniformDesign with good robust is selected as sample selection method. From the conditionsof center-deviationL2-CD2to reach the lower and the large computation problem,the existing problems of uniform design are analyzed. And the method for largenumber of uniformly distribution sample points based on genetic algorithm and smallnumber of uniformly distribution sample points with big number of variables based on cutting method are proposed to solve the problems of uniform design in cubespace.2. Sample selection methods in simplex spaceThe variable space for the morphing method base on NURBS is simplex space,and the constraints between variables are∑n1Ci=1i=. The Simplex-Lattice Design andSimplex-Centroid Design has strict limits on the number of sample points and mostsample points on the boundary which are very unfavorable for approximation model.In this paper, the inverse transform method and the uniformly distribution samplepoints selection method in cube space are combined, the method for large number ofuniformly distribution sample points based on genetic algorithm in simplex space aredeveloped, and successfully applied to the wave resistance optimization.3. Dynamic sample points selection based on correlationThe current methods only consider the variable space, separate variable spaceand performance space, without considering the variable space’s needs for the samplepoints’ distribution. It is a static method that can not achieve "according to need".This paper presents dynamic sample points’ selection method, using feedbackinformation from performance space, to achieve the reverse guide sample points’distribution. The dynamic sample points’ selection method has more advantage thanthe static method in terms of the approximate model construction by test functions.