engleski

Neural Networks Based Combinatorial Identification Model for Increasing Redundancy of Sensors Information in Marine Control Systems

The model structure for dynamic system parameters monitoring, designed to increase the redundancy of sensor information and improvement of marine control system availability, is presented. The model consists of two subsystems, the first one for on-line monitoring and the second for off-line monitoring. The off-line monitoring subsystem is based on combinatorial identification of interrelationship between any two disjunctive subsets of analysed parameters primary set. The approximative regression neural networks were used for that purpose and probabilistic neural network was applied for efficiency evaluation of each simulation model. The procedure could be performed for all possible parameter combination within the steady process modes, defined previously. Afterwards, it is possible to choose the optimal model according to the obtained efficiency rank-list of all simulation models. The main on-line subsystem function is a faultlessness control of system sensors. If failure of one or more sensors occurs, the optimal simulation model for estimation of lost sensor information is activated. Moreover, residual monitoring between measured real-time sensor data and simulated data for each steady process mode of trained off-line subsystem can be used for diagnosing faults and/or significant deviations of conventional sensor control.

Marine Control Systems; Sensors; Neural Networks

nije evidentirano