Graphical approach for state reconstruction and monitoring analysis

This paper proposes a methodology to avoid the differentiation problem of the measured position information for velocity estimation when Bond Graph model-based fault detection and isolation is performed on mechatronic systems. The Bond Graph modeling tool is interesting to model mechatronic systems, which commonly contain position sensors in their measurement architecture. Indeed, the Bond Graph tool can not model these sensors, thus a measurement of velocity is considered, which corresponds to the power flow variable. The latter addresses two problems. Firstly, the measurement architecture presented on the Bond Graph model does not represent the real one. Secondly, the velocity variable is required for fault detection and isolation. Hence, a differentiation of the position measurement which amplifies the measurement noise, is required. To overcome the aforementioned problems, a representation of sensor function is presented, and a Luenberger observer that estimates the velocity variable is designed directly on the Bond Graph. Finally, this estimation is used for fault detection and isolation. For validation, the proposed method, is applied to an electromechanical system corresponding to a quarter of an intelligent autonomous vehicle (IAV), named RobuCar.