Currently, the direction of technological development is gradually turning to information technology, digitalization and intelligence. Large-scale machinery and equipment are becoming more and more delicate and complex. However, the majority of diagnoses in the field of mechanical faults are performed by experts or expert systems, which require domain experts to guide the completion. Therefore, it is difficult to realize fault diagnosis and predictive maintenance of mechanical equipment without rich knowledge in the field of mechanical faults. Considering that equipment vibration signals carry rich state information that can be used for fault diagnosis during the operation of mechanical equipment and are sampled at a fixed location according to a certain frequency, this provides a convenient way to establish an Event Knowledge Graph (EKG) with time and location attributes. Meanwhile, using the trained Convolution Neural Network-Long Short-Term Memory (CNN-LSTM) model, it is possible to realize fault classification and prediction functions, and then provide predictive maintenance solutions for users based on the established event knowledge graph. Taking rolling bearings, an important component of large mechanical equipment, as an example, firstly, a rolling bearing fault event knowledge graph was established based on Neo4j. Then, a CNN-LSTM model with fault diagnosis classification and prediction functions is trained and established based on the relevant data from the knowledge graph. Finally, specific fault solutions are obtained with the help of the established model and EKG.