Transient Gas Turbine Performance Calculations Based On Gaussian Process Regression

Date

2023-7-20

Author

Küçük, Semanur

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

211
views

502
downloads

Cite This

In this study, a transient simulation tool was developed to predict the performance of gas turbine engines using the Gaussian Process Regression (GPR) method, one of the Machine Learning methods. First, a code was written for predicting a turboshaft engine’s off-design and transient performances using the component-matching algorithm, which is commonly used in commercial gas turbine performance programs. This is an essential step since the outputs off-design code are inputs for the transient analysis. The same approach is used in the development of transient code using Gaussian Process Regression. Training data, which the off-design code has produced for the transient algorithm, cover all operating points of the engine at different speeds and power-drawn off-design points. Transient code developed using the component matching algorithm was then modified using the GPR algorithm. The new transient code takes each steady point of transient operation from the GPR algorithm and timedependent behavior is the same as in the component matching algorithm. The new transient code developed in this study using GPR was compared with the GasTurb13 and NPSS programs. Almost the same outputs have been obtained using the newly developed code and the programs. However, the newly developed code has reduced the convergence time from minutes to seconds. Another advantage of the new code developed in this study is that since the training data set for the current code using GPR can be taken from already existing test data of a gas turbine engine, there is no need to conduct calibration work, which may last for weeks, to obtain calibrated engine model.

Subject Keywords

Gas Turbine Performance Modelling, Thermodynamic Modelling, Transient Deck Generation, Component Matching Method, Machine Learning, Gaussian Process Regression

URI

https://hdl.handle.net/11511/104881

Collections

Graduate School of Natural and Applied Sciences, Thesis