Neural Network Predictive Control of Propeller-Driven Underwater Glider with Controllable Wings and Rudder

Abstract

This paper presents a neural network (NN) motion control analysis of a propeller-driven underwater glider with independently controllable wings and a rudder. A conventional underwater glider is a fixed-wing buoyancy-driven underwater platform, which has already demonstrated highly energy efficient. However, buoyancy-driven underwater glider has limitations in terms of maneuverability and speed due to the limited external control surfaces. Due to that, we have designed the mathematical model of a propeller-driven underwater glider with independently controllable wings and a rudder by using the Newtonian and Slender-body theory. The neural network predictive controller, which based on multilayer perceptron, has been designed to control the glider motion. A three-layer network architecture, which has four control input nodes, four hidden layer nodes, and nine output nodes is designed as the forward model architecture. On the other hand, the inverse model architecture is used for the neural network controller. The feasibility and accuracy of the controller performance is studied through simulation. The simulation demonstrates that the control inputs and the target outputs are successfully predicted and achieved. The results show that the controller performance is satisfactory while maintaining the glider's stability.