Actuator Fault Reconstruction via Dynamic Neural Networks for an Autonomous Underwater Vehicle Model

This paper proposes the development of a scheme for fault diagnosis actuators simulated model accurately representing behaviour an autonomous underwater vehicle. The Fossen usually adopted to describe dynamics vehicle has been generalised in this take into account time-varying sea currents. proposed detection and isolation strategy uses data-driven approach relying on multi-layer perceptron neural networks that include auto-regressive exogenous prototypes provide reconstruction. These tools are thus exploited design bank dynamic residual generation trained basis input outputmeasurements acquired from simulator. In work, residuals designed represent reconstruction signals themselves. Moreover, network is also able perform task, case simultaneous concurrent faults affecting actuators. firstly describes steps performed deriving solution. Secondly, effectiveness demonstrated by means high-fidelity simulations realistic vehicle, presence marine current.