The model based monitoring modules of energy processes present in risky industries are conceived using the Bond Graph model (BG). The coupling of the multi-domain energies (thermal, chemical…) involved in such processes yields to strongly non-linear dynamic models. Furthermore, the parameters of models (hydraulic coefficients, thermal exchange coefficients …) are not known. At the modeling level, the use of BG as multidisciplinary tool for systematic generation of non linear and linearized models of a complex process is considered; the BG of any installation is directly obtained starting from its Process and Instrumentation Diagram (PID). The determination of structural properties directly from BGs without need of calculation is then considered for the FDI (Fault Detection and Isolation) problem. Considering the non linear or the linearized model, two hypothesis are distinguished whether we treat the diagnosability of single or simultaneous faults. For the components, actuators and sensors set, direct monitoring methods (without the Analytical Redundancy Relations (ARR) generation) are proposed. Using a generic approach based on flatness, the monitorability of simultaneous faults on actuators is then considered directly from the non linear bond graph. This approach is direct because it needs only the flat outputs identification. The innovative interest is the use of only one representation « the BG » for modeling, flat outputs determination and then the structural monitorability using flatness. Finally, the research results were illustrated by an application to an industrial process of steam generation. The modular approach of generation of linearized models was automated by an informatics’ implementation.