Abstract
In modern day gas turbine hot sections, active cooling of liner walls is critical due to extremely high temperatures and long continuous operation. A secondary flow can be introduced that shields the metal surfaces by forming a film increasing the durability and life of the liner. Typically, this cooling flow is mixed with the mainstream by a series of holes or slots along the length of the liner wall. In the past, typically, these effusion holes were cylindrical. However, as the hot section temperature increases, more advanced hole designs are necessary for increased cooling effects.
Determining an optimal hole design is costly and time-consuming if approached with standard experimental trial-and-error techniques. Numerical tools can aide in optimization studies, but for complex transient flows requiring scale-resolving methods, simple optimization methodologies require thousands of design points and can take extended periods of time to complete.
In this study, a metamodel-based optimization of a shaped cooling hole is demonstrated in an automated workflow. The automated workflow consists of six geometrical hole parameters being investigated at flow conditions akin to the hot section at a constant blowing ratio. These six parameters completely define the shape of the hole leading to a complete optimized design for maximum cooling effectiveness. The metamodel is based on data from a stress-blended eddy simulation (SBES). To construct the meta-model, an adaptive technique is used that employs machine learning (ML) algorithms to minimize the number of required design points by dynamically modifying the design space sampling based on the gradients in the response. As part of the metamodel generation, a sensitivity analysis is performed to eliminate geometry parameters which do not influence the cooling effectiveness.
The optimization study is done in two parts: the first part involves an evolutionary algorithm-based optimizer which runs the fast-running metamodel through multiple design points to identify potential optimal solution candidates; the second part involves validating the candidate designs using actual solver calls. This two-pronged approach drastically reduces the time to achieve optimal effusion hole design while ensuring that the design space is explored thoroughly. Overall, this work demonstrates the enhanced optimization capabilities and timeline compression of a shaped cooling hole.
