This paper proposes a fault diagnosis method to diagnose multiple transistor open-circuit faults in a T-type three-level inverter. In this method, a finite-state machine (FSM) tracks state transitions caused by abnormal fault-linked current paths, and rough set theory (RST) is employed to optimize and obtain a minimum set of variables necessary to distinguish state transitions under various fault scenarios. After applying RST, voltage state variables expressed by Boolean logic relationships are adopted in the FSM to identify faults. This can also effectively reflect state transitions between single and multiple fault cases. The approach is immune to load disturbances and dead times. Through logic relationships, a circuit is designed for fast online fault location to minimize the impact of sampling frequency on diagnosis. Factors that affect diagnosis time and accuracy are considered and analyzed to ensure the reliability of the proposed method. Experimental results obtained under various conditions verify the effectiveness of this approach.