Hybrid Thermodynamic and Machine Learning Approach for Performance Prediction and Optimisation of Gas Turbine Engines

Abstract

In this study, a data-driven approach for thermodynamic performance analysis of marine gas turbine engine using automated machine learning techniques is presented. A dataset of different operating conditions was generated by varying key input parameters, namely, turbine inlet temperature and compressor pressure ratio. The corresponding outputs include coolantto-gas mass ratio, compressor work, turbine exit temperature, net gas turbine work, and overall efficiency. Exploratory analysis using contour plots, response surface and pair plots was performed to identify correlations among variables. Subsequently, AutoGluon was employed to develop predictive models and compare 18 machine learning algorithms based on the coefficient of determination (R²). The results indicate that machine learning models provide superior predictive accuracy by effectively capturing the nonlinear thermodynamic relationships. The proposed methodology reduces the effort taken in computing along with maintaining high prediction accuracy, enabling rapid performance estimation. This approach is well-suited for design optimisation and real time performance prediction of marine gas turbine system.