A neural network-based modeling approach for transient performance prediction of gas turbine engines

Considering the difficulties in obtaining the specific component maps and highly iterative performance requirements when analyzing the transient performance, modeling the transient process is quite a complicated task. With few but sufficient experimental data, this study establishes the dataset-driven neural network models to predict thrust and exhaust gas temperature for the transient process of gas turbine engines. In addition, the transient parameters calculated from GasTurb13 common models are performed. Three neural network models, including convolutional neural network (CNN), long-short term memory neural network (LSTM), and CNN-LSTM, are built, trained, and tested. Compared with the numerical and experimental results, the LSTM model established in this research has a quite significant performance and an ability for forecasting the key parameters by inputting the other relevant parameters. According to the overall validation analysis, the recommended method could donate accurate results using few experimental data and small hardware resources, demonstrating awesome potential that completes the intelligent control of the aircraft and engine, and evaluates the dangerous influence of transient mechanical stress.