To address the issues of insufficient diagnostic accuracy and poor interpretability in existing data-driven gas path fault diagnosis techniques for aircraft engines, a knowledge driven deep graph neural network (KD-DeepGNN) algorithm is proposed, which integrates expert knowledge of engines. The JT9D engine simulation platform was developed to generate operational data for the training and validation of the algorithm. Gas path analysis was employed to analyze the relevant symptoms involved in the degradation calculation of each component, thereby partitioning the sensor space. The graph structure was characterized and generated by comprehensively considering the gas path connections and transmission relationships between rotational components. On this basis, the algorithm uses stacked graph convolutional modules and hierarchical graph pooling to adaptively extract degradation feature information from the data, thereby enhancing the fault diagnosis accuracy and generalization capability of the algorithm under complex environments and operating conditions. The study demonstrates that the proposed model achieves a fault classification accuracy of over 97% on the test sets, with an improvement of at least 2.2% compared to existing methods.