Application of Kriging Model with Sequential Infill Criterion on Multi-Objective Optimization of Nose Shape for High-Speed Train

Ziyuan Dai, Tianrui Li, S. Krajnović, W. ZhangCivil-comp 2024 · DOI

Abstract

This research presents a novel multi-objective optimization framework for high-speed train nose shapes using a Kriging surrogate model enhanced by a Sequential Infill Criterion (SIC). The study focuses on minimizing aerodynamic drag and lift while maintaining computational efficiency. By implementing a two-stage infill process that incorporates gradient information (PGEIC), the researchers achieved a surrogate model with less than 1% prediction error. The resulting optimized design demonstrates significant aerodynamic improvements, reducing head car drag by 4.15%.

Introduction

The aerodynamic performance of high-speed trains is critical for energy efficiency and stability. However, optimizing complex geometries like the train nose using high-fidelity Computational Fluid Dynamics (CFD) is computationally expensive. Surrogate models, such as Kriging, offer a solution by approximating the performance landscape. The challenge lies in selecting the most informative sample points to ensure model accuracy without excessive simulations.

Methodology: Sequential Infill Criterion (SIC)

To address the sampling challenge, a two-stage Sequential Infill Criterion (SIC) was developed to build a high-precision Kriging model efficiently.

The two-stage SIC process: Stage 1 focuses on global accuracy, while Stage 2 refines the Pareto front using gradient information.
  1. Stage 1: Global Exploration: The model initially uses the Expected Improvement Criterion (EIC) to explore the design space broadly, establishing a baseline understanding of the aerodynamic characteristics.
  2. Stage 2: Pareto Front Refinement: Once global convergence is reached, the model transitions to the PGEIC (Expected Improvement fused with Gradient Information). This stage focuses on the "Pareto front"—the set of optimal trade-offs between conflicting objectives like drag and lift—using gradient data to pinpoint the best designs with surgical precision.

The train nose was parameterized using five B-spline features, including nose height, cab window height, and contour lines, allowing for flexible geometric variations during the optimization process.

Results and Performance

The PGEIC-SIC method was validated against standard benchmarks and then applied to the train model. The integration of gradient information proved to be a significant advantage, allowing the model to converge to the optimal solution much faster than standard global search methods.

Prediction error convergence for head car drag (DH), tail car drag (DT), and tail car lift (LT) as infill samples are added.

As shown in the convergence data, the final surrogate model achieved a prediction error of less than 1% for all three optimization objectives: head car drag (DH), tail car drag (DT), and tail car lift (LT). This high fidelity ensures that the surrogate model's predictions are reliable substitutes for expensive CFD simulations.

Aerodynamic Improvements

The optimized nose shape, characterized by a more rounded and slender profile with a lower nose and cab window height, yielded the following improvements over the original design:

  • Head car drag (DH): Reduced by 4.15%
  • Tail car drag (DT): Reduced by 3.21%
  • Tail car lift (LT): Reduced by 3.56%

Conclusion

The study demonstrates that the PGEIC-SIC framework significantly enhances the efficiency and accuracy of multi-objective aerodynamic optimization. By strategically combining global exploration with gradient-informed local refinement, engineers can achieve high-performance designs with a fraction of the computational cost of traditional approaches. Future work will explore the application of this method to transient flow conditions and complex environments like tunnels and crosswinds.

Summary updated

Abstract

This study proposes a sequential infill criterion (SIC) appropriate for the Kriging surrogate to address this issue.Multi-objective functions are employed to test the feasibility of constructing a surrogate model based on SIC, and the SIC surrogate model then performs multi-objective aerodynamic optimizations on the high-speed train.The findings indicate that the improvement infill criterion (EIC) that fuses the gradient information (PGEIC) surrogate model achieves the lowest generational distance (GD) and prediction error.The performance of EIC for global search, EIC for Pareto front search (PEIC), and infill criterion for Pareto front search using only gradient information (PGIC) is poor.The final PGEIC-SIC surrogate model of train aerodynamics has less than 1% prediction error for the three optimization objectives.The optimal solution reduces the aerodynamic drag force of the head car and the aerodynamic drag and lift force of the tail car by 4.15%, 3.21%, and 3.56%, respectively, compared with the original model.Furthermore, the nose and cab window heights of the optimal model have been reduced, and the lower contour line is concave.Correspondingly, the streamlined shape appears more rounded and slender.
BibTeX
@inproceedings{dai2024application,
  author = {Ziyuan Dai and Tianrui Li and S. Krajnović and W. Zhang},
  title = {{Application of Kriging Model with Sequential Infill Criterion on Multi-Objective Optimization of Nose Shape for High-Speed Train}},
  booktitle = {Civil-comp},
  year = {2024},
  doi = {10.4203/ccc.7.3.11},
}
Figure