5. Conclusion

Building a model-based supervision system is a costly operation requiring several complex steps. The bond graph tool, thanks to its causal and structural properties, and its graphical and behavioral aspects via its functional architecture, is well suited to the integrated design of such systems. This ranges from modeling, off-line determination of monitoring conditions and means before industrial realization and, finally, to the generation of robust on-line monitoring algorithms.

In order to improve the alarm generation decision procedure, to avoid false alarms and non-detection, adaptive thresholds, taking into account parametric uncertainties (explicitly introduced on the bond graph model), are generated directly from the monitoring bond graph model.