An Inverse Design Method for Caudal Fin of a Biomimetic Propulsion System for AUVs Using Artificial Neural Networks

Abstract

Biomimetic propulsion systems offer higher endurance and efficiency as compared to conventional rotating propellers. In recent advances, studies have been focusing on identifying caudal fin shapes which would give optimal performance. Many studies have done comparison among existing fish tail fins using experimental and numerical models. A simple forked rigid caudal fin with four geometrical parameters such as caudal peduncle depth, leading-edge angle, trailing edge angle and chord length is studied. The present study aims to develop an inverse design procedure for predicting the caudal fin shape for the required flow conditions and performance parameters for one-meter-long autonomous underwater vehicle (AUVs) in the range of 0.15 to 0.4 of Strouhal number. The AUV is to be propelled by a body caudal fin (BCF) propulsion system with a tail fin of length 100 mm. The design factors will be average thrust and input power per cycle at a particular tail-beat frequency, amplitude and swim velocity. Eight geometries are generated by varying caudal peduncle depth or leading-edge angle separately. The geometries are analyzed numerically in ANSYS Fluent for each flow condition. These datasets are then fed to a feed-forward backpropagation neural network designed in MATLAB. The neural networks are designed to interpolate single geometric parameter. The network designed to predict caudal peduncle is provided with 72 data sets, and it is able to predict the caudal peduncle value with 94% efficiency. The neural net which is trained with same number of data sets for leading-edge angle predicted to 97% accuracy. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.