This chapter deals with robust fault detection and isolation using the bond graph (BG) approach and linear filters. The bond graph tool is used to model the dynamic system and the uncertainties on the sensors and actuators. The same model is used to generate systematically the analytical redundancy relations and the thresholds. A specific form of digital linear filter is used to evaluate the residuals and to ameliorate the robustness and the detectability of the faults. the developed procedure is applied to experimental data of an electromechanical system which is a subsystem of a mobile robot named Robotino.