Real-Time Monitoring of Aircraft Engines Using a Feedforward Deep Neural Network

Real-time engine condition monitoring requires efficient performance parameter prediction models. Due to the highly nonlinear and dynamic characteristics of the engine, it is difficult to establish a high-precision dynamic engine model according to the laws of thermodynamics. To solve this problem, a method for modeling engine dynamic systems using feedforward deep neural networks with derived features is proposed, and an engine condition monitoring program is written. The method is verified using the full-flight stage data from the aircraft’s fast access recorder, and the results show that the introduction of time-derived features combined with univariate prediction through sliding window data can effectively improve the prediction accuracy of the feedforward deep neural network, even slightly higher than the long short-term memory network of the same structure, and the root mean square error of the prediction of engine performance parameters in the full-flight stage can reach less than 10%. The condition monitoring program enables real-time and offline condition monitoring of the engine through threshold alarms and historical data analysis.