The development of advanced multivariable, linear and nonlinear control design methods with applications to marine vehicles

This thesis primarily concerns control and identification of FPSO and Shuttle Tanker vessels, where nonlinear hydrodynamics raise the associated issue of nonlinear control. A 3-DOF model is presented for investigating Dynamic Positioning control, a problem where directional thrusters maintain ship position and heading against environmental disturbances. The coupled, multivariable dynamics are controlled using rapid tuning techniques to decouple the plant, yielding successful multivariable PI feedback designs. Identification of a coupled FPSO and Shuttle Tanker is achieved using an MLP neural network. Initially, the network is trained with simulation data for proof of concept, before employing real data from a Mitsubishi Heavy Industries scale model. Identification is successful, but performance degrades with increasing wave height. Two adaptive controllers are developed, based on polynomial LQG and LQG PC optimal control theory. The first uses a standard stochastic cost, approximated to produce a restricted structure controller that permits optimisation across several plant models at once, yielding a multiple model controller. Augmenting linearised ship models with online identification produces adaptive control giving interesting trade-offs between robustness and performance. The second adaptive controller is very similar, but based on a multi-step predictive cost function. Both controllers are applied to FPSO surge axis velocity control, where the LQGPC version produces better performance for a wave-induced reference. A multivariable nonlinear controller is examined for "sandwich" systems consisting of a linear transfer function "sandwiched" between input and output nonlinearities of a particular form. This system description is substituted into the solution of a time-varying polynomial optimal control problem, where the assumption of a frozen plant at each sampling instant requires slowly-varying plant signals in practice. The controller is successfully applied to a 2 x 2 plant with deadzone input and backlash output, with a demonstration that the performance is superior to a well-tuned linear controller.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral-pg

Grantor dc:publisher.institution

University of Strathclyde

Year dc:date.issued

2005

Author dc:creator

• Martin, Peter

DOI

10.48730/ptyt-2664

Identifier

uk.bl.ethos.417975
T11277

OAI identifier oai:identifier

oai:strathclyde:8w32r571f