Research On Modeling And Optimization Methods For Robust Design

Robust design which is used to deal with the uncertain factors has been gaining increasing attentions to many researchers and is applied to many fields.Recently with the development of science technology and engineering it is significant to make a farther research about robust design.Based on the existent work in literatures the dissertation further develops the non-probabilistic robust design theory.The main content of this paper include the following aspects:1)In most of practical engineering optimal problems,the relationships between quality characteristics and uncertain factors are unknown or complicated.A hybrid response surface method is proposed to solve this problem in robust design.By considering the simpleness of polynomial response surface method and compute precision and cost of artificial neural network method,a hybrid strategy of robust design is proposed,the strategy can balance the relationship between compute precision and cost;then use linear physical programming to adjust the values of objective function and variations.2)Combining the maximal variation analysis and linear physical programming a new mathematic model of non-probabilistic robust design is proposed.In view of robustness of objective function and constraints in robust design,the principle of variations which are generated in objective functions and constraints are particularly analyzed,then estimates the maximal variations of objective function and constraints through the use of maximal variation analysis.A new constraint is added to original optimization problem to ensure the variation of objective function is less than the value which designer sets;the constraints are divided into three types,robustness index is used to adjust robustness of each constraint,and then a bi-level mathematical optimal model is constructed.The top-level optimization is used to solve the original mathematical model;the lower-level optimization is used to judge the robustness of objective function and constraints.The solutions obtained by the approach are feasible and compared with other robust optimal methods our method has the advantages that it is straightforward and doesn't require presumed probability distribution of uncertain factors or gradient information of the original mathematical optimal model.3)The multiobjective robust design is studied.In view of multi-objective optimization problem in robust design,a method of multiobjective robust design based on improved genetic algorithms is given.The method consider mixed discrete variables in engineering optimization problems,the Pareto front can be more even by improving Pareto set filter in genetic algorithms.The method takes advantage of parallel computation and random search of genetic algorithms,make designers have more choices.A fuzzy compromise programming approach to determine the optimal solution of robust design is proposed.By using fuzzy preferences the proposed approach gives a global evaluation for conflicting objectives,takes into account the decision-maker's preference by his/her assigning weights to the objectives, and then gives a satisfied solution;4) Using the theories and methods of robust design in this paper a real engineering problem about speed increasing gearbox for wind generator is studied,and the mathematic model about gear-driven system is built.The results show that the theories and methods in this paper are valid and practicable.