Surrogate-based drag optimization of Autonomous Remotely Vehicle using an improved Sequentially Constrained Monte Carlo Method

For high-cost simulation-based optimization design problem, surrogate model is usually constructed to reduce computational cost and time. When there are complex constraints for actual engineering needs, the sampling method in an irregular design space should be further considered. In this paper, a Sequentially Constrained Monte Carlo (SCMC) method is first introduced, and the "maximization of minimum distance" criterion is applied to achieve uniform and progressive sampling within a limited sample size to construct the surrogate model in irregular design spaces. Four numerical cases are validated consisting of different types of constraints and dimensions. Results demonstrate that the proposed method has broad applicability in achieving uniform and progressive sampling in many kinds of irregular design spaces. A mathematical function defined in an irregular design space, and an Autonomous Remotely Vehicle (ARV) layout optimization case are then given. Compared with the traditional experimental design methods for regular design spaces, the surrogate model constructed using the proposed method with fewer sample points can achieve the same or higher fidelity level, thus making the accuracy of the constructed surrogate model high enough with limited sample points. The optimization result for the ARV also shows that, for the total drag, the typical optimal layout obtained based on the proposed sampling method and Kriging surrogate model has a 6.54% and 7.66% decrease at two speeds. In addition, the total drag predicted by the Kriging model is almost the same as that calculated by the viscous-flow CFD evaluation with an only 0.53% and 0.09% relative error, proving that the SCMC method has advantages and potential in the high-cost ship and offshore structure's optimization designs with complex constraints.