Bond Graphs and Bayesian Networks Based Fault Diagnosis: Application to a Photovoltaic Power Supply

Model based fault diagnosis can be applied when sufficient information is available on the components and the structure of the process. The bond graphs, as multidisciplinary graphical modeling approach, are suitable since they help to deduce all the constraints and the causality within the model. The task of monitoring industrial processes does not only concern obtaining the desired performance of the control, but also performing a successful diagnosis of the fault in order to carry out the suitable reconfiguration. New generation of power converters should be fully automated and sustainable to ensure the continuity of service. The objective of this research work is to design an automatic supervisor of the power supply equipped with a reconfigurable interleaved boost converter by association of bond graphs and Bayesian networks approaches. The Bayesian Networks are considered as an artificial intelligence method which improves the decision procedure deduced from bond graph model by integrating reliability data of the components. The simulation results of some scenari proves the efficiency of the proposed method applied to a photvoltaic power supply.