Due to the influence of background noise and transmission path, fault signals are often submerged, and fault features are difficult to extract. Therefore, a fault diagnosis method combining SVMD and adaptive MOMEDA was proposed. The reconstructed signal was obtained by SVMD preprocessing, and then the optimal parameters of MOMEDA were adaptively selected by artificial fish swarm optimization algorithm with the average spectral negative entropy as the adaptation function. The obtained parameters were used to carry out MOMEDA filtering on the reconstructed signal. Finally, the envelope spectrum analysis was carried out to diagnose the fault type. By applying the proposed method to the simulation signal and experimental signal of the broken tooth fault of the gearbox active wheel, the pinion rotation frequency and its multiplication can be clearly distinguished in the envelope spectrum, and the proposed method has better performance effect than other methods.