Civil aircraft flight control systems adopt a multi-redundancy architecture to ensure safety. However, this design, while enhancing reliability, introduces coupling and nonlinear causal dependencies among components, leading to the propagation of single faults across components. Accurately identifying the causal relationships among components is crucial for analyzing the root causes of faults and identifying their propagation paths. Therefore, a method for identifying fault propagation paths in flight control systems based on causal graphs is proposed. Firstly, a dual attention-based multi-channel depthwise separable temporal convolutional network (Da-MDSTCN) is constructed. Through the attention fusion mechanism, the causal association features among components are analyzed, and a system structure causal graph is built in combination with graph theory. Secondly, a multi-channel weighted temporal convolutional network (MC-WTCN) is designed for feature fusion. Through a feature modulation strategy, the collaborative prediction of multi-dimensional state feature parameters is achieved, and a diagnostic framework based on prediction residuals and relative deviation analysis is established to realize the diagnosis of fault origins and the identification of propagation paths. Finally, experiments are conducted based on the A320 flight control system. The fault diagnosis accuracy reaches 94.8%, and the identification of fault propagation paths is more accurate, verifying the accuracy and effectiveness of this method in fault diagnosis and propagation path identification.