Enhancing Accuracy of Transition Models for Gas Turbine Applications Through Data-Driven Approaches

TL;DR

This paper employs machine learning and gene expression programming to significantly enhance the accuracy of separated flow transition predictions in gas turbines, outperforming traditional models across multiple parameters.

Contribution

It introduces a novel data-driven approach using gene expression programming and multi-objective optimization to improve transition prediction models for gas turbines.

Findings

Wall shear stress prediction improved by 40-70%.

Models outperform baseline in transition-related parameters.

Method shows potential across various geometries and conditions.

Abstract

Separated flow transition is a very popular phenomenon in gas turbines, especially low-pressure turbines (LPT). Low-fidelity simulations are often used for gas turbine design. However, they are unable to predict separated flow transition accurately. To improve the separated flow transition prediction for LPTs, the empirical relations that are derived for transition prediction need to be significantly modified. To achieve this, machine learning approaches are used to investigate a large number of functional forms using computational fluid dynamics-driven gene expression programming. These functional forms are investigated using a multi-expression multi-objective algorithm in terms of separation onset, transition onset, separation bubble length, wall shear stress, and pressure coefficient. The models generated after 177 generations show significant improvements over the baseline result in terms of the above parameters. All of the models developed improve the wall shear stress prediction by 40-70\% over the baseline laminar kinetic energy model. This method has immense potential to improve boundary layer transition prediction for gas turbine applications across several geometries and operating conditions.