Neuromorphic ROV propulsion control based upon model-reference non-linear optimisation

Thruster propulsion is a significant component for making accurate motion control possible as well as achieving superior navigational performance of remotely operated vehicles (ROV), both for tethered and untethered classes. As a result, the contents of this paper mainly focus upon thruster functioning enhancement by forming an effective real-time closed-loop compensation. An electric-driven thruster, instrumented with drive and embedded electronics, is employed as a test-bed. Its electrical and mechanical properties are briefly described and the inherent existence of stiction (static friction) will be highlighted since such nonlinearity directly degrades the dynamic behaviour of the thruster. Therefore, a nonlinear optimisation technique, serving as a 1-step-ahead predictive controller, has been structured not only to eliminate this unwanted stiction effect, but also to have the following additional attributes: 1) to maintain commanded rotational direction, 2) to regulate angular velocity with respect to set-point, and 3) to produce a robust system with better noise immunity. The intelligent predictive algorithm is based upon a multilayer perceptron (MLP) neural network configured as a state-estimator, whereas a model-reference gives the desired response characteristic. The outcome of such combined modules will clearly outperform the classical methodologies as indicated by the comparison with: i) an ordinary predictive strategy, and ii) an industrial three-term PID regulation