Abstract: In response to the problems of over-reliance on manual experience and low defect disposal efficiency of traditional transformer maintenance mode, the paper proposes a risk assessment and auxiliary maintenance decision method based on defect text recognition and knowledge graph for transformer. By establishing the defect text recognition model based on Bert-CNN, the method completes the dynamic word vector extraction and text local feature analysis of the defect records filled in by operation personnel, and automatically evaluates the severity and risk level of equipment defects. Then, based on industrial standards, operation specifications and expert experiences, knowledge graph is used to construct the operation and maintenance strategy library of transformer, which realizes the knowledge fusion and mapping of defect text recognition results and maintenance strategy library. The intelligent operation-inspection auxiliary function of the whole process from equipment defect record to operation and maintenance decision-making is improved. Finally, combined with algorithm comparison and case verification, the accuracy of the method to identify the defect severity, component and risk level is higher than 91%, and decision-making on differentiated maintenance and suggestion based on equipment defect conditions is realized, which helps to improve the intelligence and automation level of transformer operation and maintenance.