Optimal design and performance prediction of stepped honeycomb labyrinth seal using CFD and ANN

Abstract

A stepped honeycomb labyrinth seal was optimized, and its leakage performance was predicted across various operating conditions using computational fluid dynamics (CFD) and artificial neural networks (ANNs). The process involved two stages: geometry optimization and performance prediction. In the first stage, incremental Latin hypercube sampling (i-LHS) was used to select geometric design points for training the ANN with CFD providing the leakage performance data. An ANN-based performance prediction metamodel was developed, and a genetic algorithm was applied to the metamodel to optimize seal geometry, achieving a 12.34% improvement in leakage performance over the reference seal. The second stage involved performance prediction across a wide range of operating conditions, including pressure ratios, rotational speeds, and clearances. Similar to geometry optimization, i-LHS was used to select the operating design points for training the ANN. A metamodel reflecting operating conditions was developed by evaluating the generalization and practicality of the ANN. The impact of pressure ratio, rotational speed, and clearance on the leakage performance was predicted. The leakage performance of the optimized seal was compared with the reference seal, showing improvements from 1.44% to 16.74%. This study revealed the effectiveness of ANN-based performance predictions for optimizing complex geometries, such as honeycomb seals, and developing models that account for various operating conditions.