Partial discharge fault diagnosis is used to diagnose the defects in high voltage insulation in power system equipment. However, due to the limitation of experience and professional knowledge, it is of great difficulty to extract valuable fault information from the original monitoring data. In this paper, a partial discharge fault detection technology for transformer based on knowledge inference system is proposed and developed. The information collected by the partial discharge sensor is processed to obtain a three-dimensional map of phase analysis. The transformer fault is diagnosed and located by classifying the three-dimensional map and extracting the salient features. The proposed system can diagnose a variety of partial discharge behaviors, classifies defect sources, and supports online device status assessment and fault diagnosis. In addition, an unknown mixed discharge behavior was tested and results show that the diagnostic accuracy based on the proposed method is higher than traditional technologies.