Effective equipment fault diagnosis plays a crucial role in ensuring the safe and stable industrial process. However, intelligent fault diagnosis of equipment remains challenging due to sparse fault data. This study proposes a novel knowledge-based railway equipment fault diagnosis framework (KEFD), which forms a diagnosis knowledge graph for efficient storage, retrieval and utilization of the domain knowledge. This framework could provide intuitive and efficient decision-making guidance for railway operators. Specifically, fault case records from Chinese railway Cab Integrated Radio communication (CIR) equipment are selected as the data source. KEFD first mines the standardized entities and relations from CIR diagnosis documents, transforming the textual data into a fault event-oriented diagnosis knowledge graph. Subsequently, KEFD utilizes a fault event detection method incorporating the extended structural embedding and distant supervision to accurately extract diagnosis information and add to the graph. Finally, the diagnosis task is transformed into a matching problem between nodes of fault phenomena and maintenance measures based on DistMult. Empirical validation on the CIR fault diagnosis dataset demonstrates that event extraction performance of KEFD reaches 54%, realizing 3% improvement over other baseline models. By fully leveraging rich structural and semantic information in the knowledge graph, KEFD significantly enhances the performance of fault event extraction and intelligent diagnosis, serving for the practical needs of railway departments.