The performance of an azimuth thruster in a dynamic positioning (DP) system can be adversely affected by thruster-thruster/current interactions. In order to avoid severe performance degradation caused by the interactions, forbidden zones or thrust efficiency functions must be explicitly considered in the design of thrust allocation algorithm, for the purpose of preventing azimuth thrusters from rotating towards certain directions. In this paper, we focus on studying the approach of using supervised learning algorithms to estimate thruster-thruster/current interactions between adjacent azimuth thrusters in tandem, based on scattered model test data. We first present the experimental setup and results of the model test. Then Gaussian radial basis function (RBF) network and feedforward neural network are applied to approximate the thrust efficiency function with respect to both thruster azimuth and current inflow velocity. The performance and reliability of the supervised learning algorithms are compared by analyzing the trained models.
Thruster-interaction effect must be taken into consideration when designing thrust allocation algorithms for dynamic positioning (DP) systems with azimuth thrusters. The complex thruster-thruster, thruster-hull, and thruster-current interactions can degrade thruster performance in a significant manner, and lead to gross inefficiency of power consumptions (Cozijn et al., 2017). As shown in Fig. 1, a semi-submersible platform is usually equipped with multiple azimuth thrusters, which can be turned to produce thrust in any direction in the horizontal plane. When two azimuth thrusters are operating within close proximity to each other, the thruster which is in the slipstream of the other one may suffer severe thrust losses, depending on their azimuth angles. Thruster-hull interaction also occurs due to frictions between thruster slipstream and vessel hull. The wake of an azimuth thruster can also produce a low-pressure region around hull bilge, and imping on the downstream pontoon, causing additional thrust loss.
Many model test and numerical simulation results have been presented in the literature, in order to establish estimation models for interaction-induced thrust loss. Nienhuis (1992) conducted a series of model tests to analyze the effects of thruster distance and azimuth angles on thrust loss in both open-water and under-plate conditions. The model test data was later used to establish the following formulas to calculate thrust reduction ratios of a downstream thruster under a flat bottom, as a function of thruster distance x and upstream azimuth angle ϕ (Dang and Laheij, 2004; ABS, 2013):
(equation)
(equation)
