ARTIFICIAL INTELLIGENCE TECHNIQUES APPLIED TO AUTOMATIC SHIP GUIDANCE

Abstract

It has been estimated that over eighty per cent of marine accidents have been caused by operator error. The skills of the operator in the handling of the vessel are variable and are subject to external influences. Over the past seventy years many advances have been made in the field of ship control. Early developments on proportional controllers have led to today's modern control systems which have interfacing capabilities with electronic navigation equipment. This research investigates traditional control methodologies and introduces the concept of applying artificial intelligence (AI) methods to the ship guidance problem. Research into AI techniques has been burgeoning over the last fifteen years and the main areas investigated are expert systems, fuzzy logic and neural networks. These areas are compared and the research proposes that it is feasible to design and develop a novel, advanced autopilot that is capable of learning the control functions of the operator as well as the manoeuvring characteristics of the vessel. An assessment is undertaken as to the feasibility of replicating a helmsperson's vessel handling functions with an intelligent neural network control system. This system has the capability of learning the course keeping and track keeping functions for a specific vessel. The research has been carried out under two specific task areas: neural course keeping control utilising simulation methods; and neural track keeping control exploiting the use of simulation and scale model techniques. The use of a scale model has allowed the collection of accurate training data through a integrated navigation and data collection system. The use of such a test bed has permitted the testing of the neural track keeping system. Alternative research has concentrated on the use of mathematical models of vessels and all the training data is created through the use of simulation techniques. Whilst this approach is suitable for the initial design of a neural control system can not fully replicate the disturbances acted upon and the responses of a real vessel. By utilising a scale model containing a navigation, data collection and control system it has been possible to expose the vessel to the real environmental data which is unobtainable when using simulation methods. The results of the neural control strategies implemented on the vessel guidance problem are evaluated against the teacher in terms of performance measures. The results indicate that the performance of the final track keeping system is of the manner desired in that it has learnt the control action of the operator. Areas for further research are presented including the application of alternative AI techniques and the use of more accurate navigation sensors.

Date of Award	1995
Original language	English
Awarding Institution	University of Plymouth