Neural network predictive control of a deep submergence rescue vehicle

Abstract

Underwater vehicles consist of robotic vehicles that have been developed to reduce the risks of human life and to carry out tasks that would be impractical with a manned mission. The design of a depth control of an underwater vehicle is described in this thesis. For an underwater vehicle the most important one is that the vehicle must be stable throughout its entire operational range. In this project, the modelling and design of the depth control systems for a small underwater vehicle will be described. Three types of controllers are discussed here which include PD, Fuzzy Logic and Neural Network Predictive Control. These techniques have the purpose of ensuring zero steady state error and minimum error in response to step commands in the desired depth. Through this project, the interaction between motions in the vertical plane due to the stem plane deflections had been discussed. The neural network gives satisfactory results. However, the design is complex because there are a large number of parameters that needs to be tuned. Based on these criteria, a comparison study is performed and all of the controllers are found to give satisfactory results. The proposed Neural Network Predictive Control exactly gives the better result in terms of transient and steady state response which shows the effectiveness of the designed controller.