Survey on traditional and AI based estimation techniques for hydrodynamic coefficients of autonomous underwater vehicle

Abstract

An accurately predicted dynamic model is essentially required to design a robust control system for an autonomous underwater vehicle (AUV) maneuvering in six degrees of freedom. The dynamic model consists of hydrodynamic derivatives which include inertial and damping coefficients. Traditionally, these coefficients can be estimated through Analytical and Semi-Empirical (ASE) methods, Computational Fluid Dynamics (CFD), and through experiments such as captive or free model testing. Additionally, System Identification (SI) estimators, e.g. extended Kalman filter, are also employed to predict the complete set of parameters. Recently, with the advent of Artificial Intelligence (AI) in almost every field of science, supervised machine learning algorithms such as Support Vector Machine (SVM) and neural network algorithms are also being applied to predict the coefficients with lesser computational cost and higher accuracy. Additionally, AI based parametric and non-parametric modeling of autonomous marine vehicles have also been discussed. Accordingly, the contributions of researchers and scientists, with respect to the evolution and application of these important techniques for marine vehicles, particularly torpedo-shaped AUVs, done over the past more than 75 years, have been thoroughly surveyed and sequentially consolidated in this article. At the end, based on the survey, merits and demerits of each technique over the others have been highlighted and published results have also been discussed to evaluate the effectiveness of these estimation techniques for autonomous underwater vehicles.