Superimposable neural network for health monitoring of aircraft hydraulic system

To solve the aircraft health monitoring problem with temporal and phase characteristics, a superimposable neural network (SPNN) is proposed, considering a superimposable hidden layer for placing the handcrafted temporal neurons. First, the strong relationship within monitoring parameters is established by the configuration and fault tree model, for determining the mapping relationship between health and monitoring parameters. Next, the SPNN is used to solve this temporal modeling problem. The superimposed hidden layers change the topology of the original neural network and expand the input of information through handcrafted temporal features, mapping horizontal data features in a simple framework. Finally, random sampling is used to achieve instant reliability based on the machine learning model, and the design failure rate derived by the fault tree is compared to examine the degradation of aircraft functions and system health in the current condition. In the verification part, combined with quick access recorder (QAR) data of a complete flight of a certain civil aircraft, an aircraft hydraulic system temperature health monitoring model is established. The results show that SPNN models have better robustness in health monitoring problems. In the reliability analysis, the aircraft had a 37.14% decrease in the health of the hydraulic temperature module compared to the initial state.