To avoid the increase cost of acquisition of multi-source information and the impact of rapid diagnostic system problems in handling multi-source data, as well as to further improve the accuracy of fault diagnosis, we proposed the fault diagnosis methods that based on wavelet decomposition and direct current (DC) components values to fully extract output currents information of three-phase inverse. First, using the wavelet decomposition to extract fault features and normalization processing; at the same time, in order to further improve the positioning accuracy of the faulty power tube, then we extract the three-phase output current signal of the DC components values, and fusion these two kinds of information; and finally trained by adaptive gradient descent momentum BP neural network. Simulation results show that this method avoids acquiring and processing data from multiple sources, at the same time ,further improves the identification and location of the fault power tube, accuracy rate of 98.15%, achieves the inverter open circuit fault diagnosis.