Research On Multidisciplinary Feasible Architecture Based On Coupled Gradient And Kriging

Multidisciplinary design optimization(MDO)is an important method for the design problems of complex engineering systems.In this paper,the multidisciplinary feasible(MDF)architecture is studied from two aspects,including local optimization efficiency and global optimization performance.The coupled-gradient based Multidisciplinary Feasible architecture,gradient-enhanced-Kriging based Sequence Multidisciplinary Feasible architecture and two-level sequential Multidisciplinary Feasible architecture are developed.The multidisciplinary design optimization of an autonomous underwater vehicle(AUV)is also carried out.The main achievements and innovations are as follows:(1)Because of the high computational cost of the MDF architecture using traditional gradient method,a coupled-gradient based Multidisciplinary Feasible(CGMDF)architecture is developed.The analytic method is introduced into the multidisciplinary system,and then the multidisciplinary coupled gradient method is used to calculate the gradient of multidisciplinary system,so the process of calculating the partial derivative information of disciplinary coupling variables is used to replace the multidisciplinary analysis(MDA)with high computational cost.An adaptive selection method of coupled gradient equations is also presented to minimize the computational cost of system gradient and improve the efficiency of local optimization.Finally,three typical MDO problems are used to compare the CGMDF architecture with the finite-difference based Multidisciplinary Feasible(FDMDF)architecture.The results show that the computational cost of the CGMDF architecture is lower when the optimization result accuracy is the same,which verifies the local optimization efficiency of the presented CGMDF architecture.(2)Because of the poor global optimization performance of the existing MDF architectures,the idea of surrogate model based sequence optimization is introduced,and a gradient-enhanced-Kriging based Sequence Multidisciplinary Feasible(GSMDF)architecture is developed.The system-level gradient-enhanced Kriging(GEK)model is built for sequence optimization based on the multidisciplinary coupled gradient method,and the combination of minimizing the predictor(MP)and maximizing the expected improvement(EI)infill criteria is used to balance the local search capability and global exploration capability of the sequence optimization,which can simultaneously reduce the computational cost of optimization and improve the global optimization performance.Finally,the global optimization performance of the GSMDF architecture is verified by three MDO examples,and then a scalable global MDO problem is proposed for further testing.The results show that the global optimization performance of the GSMDF architecture remains the best when the scale of the scalable MDO problem changes.(3)Because of the problem that the GSMDF architecture does not further study the discipline-level computational cost,a discipline-level Kriging based sequence Multidisciplinary Feasible(DKSMDF)architecture is developed.By establishing the disciplinary Kriging models to assist sequence optimization,the number of expensive discipline analysis is effectively reduced and the computational cost is also reduced.Then,three MDO examples are used to compare the DKSMDF architecture with the GSMDF architecture.The results show that the DKSMDF architecture can significantly reduce the number of discipline analysis,while the multidisciplinary iterative analysis based on disciplinary Kriging model(called low-fidelity MDA)increases.Therefore,on the basis of the DKSMDF architecture and the idea of system level-sequence optimization,a two-level sequence Multidisciplinary Feasible(TSMDF)architecture is developed,which can reduce both the number of discipline analysis and the number of low-fidelity MDA,so as to minimize the computational cost of global optimization.Finally,three MDO problems are used to verify the efficiency of the TSMDF architecture.(4)Aiming at the multidisciplinary optimization design problems of AUVs,the TSMDF architecture is adopted to carry out the multidisciplinary optimization design of an AUV,and a satisfactory optimization result is obtained.Firstly,AUV is divided into four disciplines,including shape and hydrodynamic discipline,power and propulsion discipline,mass distribution discipline and manueverability discipline.An automatic simulation framework of shape and hydrodynamic discipline is built,and the analysis model of each discipline is established.Then,the data coupling relationship among these disciplines is analyzed,and the design variables,objective function and constraints of the AUV’s multidisciplinary optimization design problem are determined.On this basis,the MDO mathematical model of AUV is established.Finally,the multidisciplinary optimization design of AUV is realized using the TSMDF architecture,and the optimization result is also compared with the CGMDF architecture.The results show that the optimization result of the TSMDF architecture is great and the optimization efficiency is also high.