Predictions of AUV's Hydrodynamic Parameters Based on Variable-fidelity Modeling

Variable fidelity modeling (VFM) which is regarded as an effective surrogate model is widely used in engineering design. In VFM, low-fidelity modeling's (LF) role is to capture the trend of objectives while the high-fidelity modeling's (HF) role is to revise the accuracy of LF. LF and HF are fused through bridge function (BF) and then finish the approximation for true model. This paper adopts VFM to predict autonomous underwater vehicle's (AUV) hydrodynamic parameters (drag, lift and torque coefficients) on certain velocity and angle of attack, in which LF, HF are built using RBF and BF is built using Kriging. Crossover operator of genetic algorithm produces candidate sample set and the sample with maximum error between VFM and HF will be chosen as newly added sample. Finally, VFM is established with all samples. The results show that compared to HF, VFM can describe precisely the change of hydrodynamic parameters in design space. Under same computing condition and time, VFM can gain better precision than HF and with the increases of HF samples in VFM, the accuracy of VFM becomes higher.