Photovoltaic inverters are characterized with the ability to convert DC voltage of solar panels into AC voltage to drive household appliances or boost voltage into the energy internet. As the core component of photovoltaic inverter, insulated gate bipolar transistor (IGBT) under an abnormal state will directly affect the normal operation of the system. In order to reduce the number of sensors, a fault diagnosis model based on convolutional neural network (CNN) is designed to monitor the open circuit state of IGBT with the current signal of DC side as the input signal. The simulation data generated by the designed Simulink module is used to train and test the model, thus achieving a good fault diagnosis performance. In addition, experimental tests under different level noises help to validate the effectiveness and robustness of the proposed method in noisy environments.