Abstract: The grid connected inverter is a fast-switching system and is in a long-term hard working environment. In this case, the nominal value of the controllable device will shift and gradually turn into a hard fault, which will cause great economic losses and even casualties. Therefore, we took the neutral point clamped（NPC） grid inverter as an example and construct BG model of NPC inverter according to the theory of bond graph（BG）. Then the diagnosis hybrid bond graph（DHBG） model was established after potential（De） and flow（Df） sensors were added to the BG model. Through the coverage path method we used node characteristic equation to deduce fault signature matrix（FSM） from the global analytical redundancy relationship（GARRs）. Based on this, we realize inverter incipient faults detection and localization（FDI）, and designed the adaptive threshold to reduce the false alarm. Finally, a simulation study using 20-sim platform was carried out to verify the effectiveness of the BG model-based GARRs method in incipient faults diagnosis of grid connected inverters.