ABSTRACT
This chapter presents a data-driven, digital twin of a complex laboratory gas turbine engine (GTE) using a deep learning framework. A deep neural network comprising long short-term memory (LSTM) cells is developed that maps the three inputs with seven output parameters of the laboratory GTE to model the dynamic response. The capability of the LSTM network to process the entire sequence of data is deployed in this work for predicting the dynamic response of the GTE parameter. The developed LSTM network is trained over a large-scale data set collected by performing numerous experiments. Finally, the model is validated against the experimental data set in real time. The prediction performance of LSTM networks is compared with a conventional non-linear autoregressive network and found to be efficient in predicting the dynamic response of GTE parameters. Lastly, the LSTM network–based digital twin is also compared against a mathematical model of the laboratory GTE. The prediction response shows that the LSTM network–based digital twin outperforms the mathematical model.
