Using Unsupervised Machine Learning to Experimentally Categorize Separation on Low-Pressure Turbine Blades

Document Type

Article

Publication Date

2026

Abstract

Laminar boundary layer separation in the low-pressure turbine section of gas turbine engines can reduce efficiency significantly. Vortex generator jets (VGJs) have been shown to prevent separation on turbine blade surfaces. The efficiency of the jets can be improved by activating them only when separation occurs. Prior related computational work has shown that unsupervised machine learning (ML) methods can be used to identify boundary layer separation characteristics, such as, conditions that generate higher and lower losses. In this study, sparse pressure data along the blade surface is used to develop a model to detect flow separation, enabling conditional jet activation. Low-pressure turbine blades equipped with VGJs will be tested in a low-speed, linear cascade wind tunnel. A cluster-based model has been trained to categorize the flow over a range of incoming turbulence levels and Reynolds number. Total pressure loss measurements across the midspan flow will quantify VGJ effectiveness. This proof-of-concept experimental study will demonstrate the potential to use clustering and sparse measurements to detect higher and lower loss flow conditions and improve the efficiency of active flow control.

Comments

Presented at the AIAA SCITECH 2026 Forum, Januray 12-16, 2026 in Orlando, FL.

DOI

10.2514/6.2026-1750

Link to Full Text


Share

COinS