Research On Surrogate Modeling Techniques And Applied To Shape Design Of Autonomous Underwater Glider

Surrogate-based global optimization(SBGO)methods have gained popularity for their capability in handling complicated engineering design optimization problems.The fundamental issues that arise in SBGO from a practitioner's perspective,including performance of initial samples,accuracy of surrogate models,quality of adaptive sampling design methods and using design space reduction strategies or not,both play an important role on the performance of SBGO methods.The surrogate modeling techniques and surrogate-based global optimization are deeply studied in this paper.Moreover,the newly proposed SBGO method has been employed for shape design of autonomous underwater glider(AUG).The main research results are as follows:(1)A fast optimal Latin Hypercube design method is presented to overcome the timeconsuming and poor efficiency of the traditional optimal design of experiment methods.This method is based on the inspiration that a near optimal large-scale Latin Hypercube design can be built by a small-scale initial sample generated by Successive Local Enumeration(SLE)algorithm via Translational Propagation algorithm(TPSLE).TPSLE method can generate samples with arbitrary size by using sampling resizing strategy.The testing results show that TPSLE is more efficient in terms of the computation time,and have acceptable space-filling and projective properties.Thus,the more accurate surrogate models can be obtained.(2)It's hard to select an appropriate surrogate model without knowing the behavior of the real black-box problem a priori in most cases.To overcome this difficulty,an ensemble of surrogate-based adaptive global optimization(ESBA)method is proposed.The initial sample points are generated by TPSLE method and employed for constructing three classical single surrogate models including polynomial response surface,radial basis function and Kriging.The ensemble of surrogates is constructed by combining three surrogate models via relevant weight factors which are obtained by minimizing the generalized mean square crossvalidation error.The adaptive sampling strategy for ensemble of surrogates is proposed and the promising points are selected according to the known information during the search.The ensemble of surrogates and three individual surrogate model can be updated and rebuilt iteratively by adding new points sequentially until the convergence criterion is satisfied.The research results demonstrate that the proposed method is robust and efficient in dealing with different kinds of black-box problems in terms of accuracy,efficiency,robustness.It also broadens the range of engineering application.(3)For computation-intensive engineering design optimization problems,to construct an accurate surrogate model over a large design space remains a great challenge in many cases.A new global optimization method using an ensemble of surrogates and multi-layer design space reduction strategy(ESBA-MSR)is proposed to improve the optimization efficiency and ability of solving the large-scale engineering problems further.The design space is classified into three equal spaces containing: original global space,promising joint space,important local space using the multi-layer design space reduction strategy(MSR)during the iteration and the search is carried in one space.The accuracy of approximate model in important regions is improved gradually through identifying the region where the real global optimal point located most possibly and adding new sample points in the reduced design subspace.Tested using a large number of well-known benchmark problems and two engineering examples,the multi-layer design space reduction strategy can immensely relieve the computational burden and improve optimization efficiency.(4)The newly proposed multi-layer design space reduction strategy-based global optimization method ESBA-MSR is employed for AUG shape design.As a prototype,to obtain information of airfoils located in each key station of an available AUG with blended-wing-body.These airfoils are optimized to obtain high lift to drag ratio using the airfoil optimization system constructed.Then,the shape of the optimal AUG is rebuilt by all data of the optimal airfoils in the case of overall outline remains unchanged.Compared with the original AUG,the lift to drag ratio of the optimal AUG is developed by 0.29% to 24.16% at angle of attack 1°to 10°.Moreover,the maximum thickness of central cross-section airfoil and effective volume of AUG is also increased by 7.9% and 3.46%,separately.It proves that the newly proposed method ESBA-MSR can be a feasible and efficient approach in actual engineering application for locating the global optimum.