This study is addressed to a problem of fault detection and isolation (FDI) of an inertial navigation system that is used by an unmanned surface vessel (USV). The inertial navigation system (INS) includes two sensors that measure angular and linear velocity. It is assumed that all parameters of the second order USV model are unknown. The developed algorithm is based on the parameter identification with a guaranteed finite time convergence, second order directional generators of residual signals for sensor faults detection and genetic algorithm for tuning design parameters. Simulation results demonstrate the effectiveness of the proposed approach providing accurate and timely fault identification of each sensor.