Publication Date

2025

Document Type

Thesis

Committee Members

Mitch Wolff, Ph.D. (Advisor); Jared Kerestes, Ph.D. (Committee Member); Chris Marks, Ph.D. (Committee Member)

Degree Name

Master of Science in Mechanical Engineering (MSME)

Abstract

Laminar boundary layer separation can significantly degrade the efficiency of Low-Pressure Turbine (LPT) blades. While active flow control (AFC) methods can mitigate these losses, energy-efficient implementation requires activating the system only when performance decreases. This study validates an unsupervised machine learning framework that utilizes sparse, discrete surface-pressure measurements to distinguish between high and low aerodynamic loss states. A fuzzy c-means (FCM) clustering model was trained on limited pressure data obtained in a low-speed linear cascade across Reynolds numbers from 30,000 to 160,000 and used to categorize the flow regime in real time. At Re = 40,000, vortex generator jet (VGJ) blowing was increased until the model’s output shifted from a high-loss to a low-loss cluster. Simultaneously acquired total pressure loss data revealed a dramatic performance recovery that correlated directly with this shift. Furthermore, Particle Image Velocimetry (PIV) validated that the transition in the model output corresponded to the reattachment of the suction surface boundary layer. At Re = 60,000, the model successfully captured the inherent intermittency of the flow, accurately tracking the fluctuation between aerodynamic states as confirmed by PIV. These results indicate that global performance states can be reliably inferred from limited local data, offering a low-complexity pathway toward adaptive flow control in practical turbine environments.

Page Count

48

Department or Program

Department of Mechanical and Materials Engineering

Year Degree Awarded

2025


Share

COinS