Fault detection is very important to improve the reliability of power conversion devices. Faults of power semiconductors can be broadly divided into shorts and opens and are further classified into two types depending on whether there is an internal problem with the switch or anti-parallel diode. In this paper, fault-diagnosis methods for short-circuit and open-circuit states are proposed, respectively. A method of classifying and diagnosing faults by applying a gate signal to each switch is proposed to diagnose short-circuit conditions. This method uses only current magnitude information, which reduces the amount of required information and reduces diagnostic failures due to angle errors and current noise. A method is proposed to detect a faulty switch by applying a voltage vector and comparing the current angle with a lookup table to diagnose an open state. An iterative diagnostic algorithm is proposed to prevent diagnostic failure due to angle error and current noise. The effectiveness of the proposed diagnosis method is verified through experiments and simulations.