This paper introduces a robust fault detection method for a modular multilevel converter power module. Taking into account the uncertainties of the power module model, the adaptive threshold interval of fault detection is established. The isotonicity of the power module dynamic is analyzed, and the response trajectory intervals of different uncertain factors are constructed by using a vertex algorithm. The adaptive threshold interval of fault detection is built by using the upper and lower outer envelopes of the response trajectory family of all uncertain factors. The proposed method is robust to the considered uncertainties and sensitive to faults. This can effectively reduce the false alarm rate and missing alarm rate of fault detection. Simulation results show the effectiveness of the method. This work is supported by the Project of Extra High Voltage Power Transmission Company of China Shouthern Power Grid Co., Ltd. (No. CGYKJXM20180354) “Research on Fault Diagnosis Technology of Flexible DC Power Module in Luxi Station”.