Abstract : The work introduced in this thesis deals with the analysis of some observability properties for systems with unknown inputs using a graphical approach. This approach leads to quite simple conditions which can be easily implemented. Thus, it allows us to overcome some numeric difficulties that geometric and algebraic approaches present. State and input strong observability is one of the properties not treated yet on the basis of a graphical approach. This property consists on studying the observability of system's states variables for all the values of the input as well as the observability of both the state and input components. These properties which are stronger than classical observability and fault isolability are interesting to study. Indeed, the developed researches could be useful in the context of state observers or input estimator synthesis for robust control, supervision or fault tolerant control frameworks. Moreover, the studied properties also allow to verify whether the observability of the system is modified when the latter is subject to disturbances or faults. In the first part of this thesis, the graphical analysis of the total and partial state and input observability is done. The second part is dedicated to the sensor placement problem with the aim to recover the strong observability studied previously. The third part deals with the implementation of found results on a toolbox (LISA) dedicated to the structural analysis of linear and bilinear systems. LISA toolbox is made of basic algorithms implemented to verify properties related to the observability, state and input observability and fault detection and isolation. The algorithms implemented in LISA have polynomial complexity order and therefore they are suitable for large scale systems.