Abstract
The aerodynamic optimization of high-speed train heads is often hindered by the high computational cost of simulating complex geometries defined by numerous design variables. While Partial Differential Equation (PDE)-based modeling offers a way to describe smooth surfaces with fewer variables, it lacks compatibility with industrial CAD/CAM/CAE standards, which rely on Non-Uniform Rational B-Splines (NURBS). This paper presents a novel hybrid optimization method to convert PDE-represented train head surfaces into optimal NURBS surfaces. By integrating a unique weight deformation algorithm with Genetic Algorithms (GA) and Least Squares (LS) fitting, the proposed method achieves high-precision geometric conversion while significantly reducing the number of required control points.
Introduction
As high-speed rail technology advances, the aerodynamic performance of the train head has become a primary bottleneck for further speed increases. Optimizing these shapes typically involves heavy computational fluid dynamics (CFD) simulations. PDE-based geometric modeling is advantageous here because it can represent complex, high-order continuous surfaces using a minimal set of design variables. However, the industry standard for engineering software is NURBS. Bridging the gap between the efficiency of PDE modeling and the universality of NURBS is essential for practical industrial application.
Methodology: The Optimal Conversion Pipeline
The conversion process involves three main stages: PDE surface generation, NURBS formulation with weight deformation, and a hybrid optimization loop.
PDE Surface Generation
The train head is first modeled using 4th-order PDEs. This approach ensures that the resulting surface patches maintain high-order continuity at their boundaries, which is critical for minimizing aerodynamic drag and ensuring a smooth aesthetic profile.
Weight Deformation and NURBS Formulation
A standard NURBS surface is defined by control points and weights. Optimizing every weight individually creates an unmanageable number of variables for complex models. To solve this, the authors introduced a weight deformation algorithm. This technique uses a single parameter, $dw$, to adjust the overall concavity or convexity of a NURBS patch, effectively collapsing hundreds of potential variables into one.
Hybrid Optimization Framework
The conversion is treated as an optimization problem where the objective is to minimize the error between the NURBS surface and the original PDE surface.
- Genetic Algorithm (GA): The GA searches for the optimal number of control points and the best weight deformation value ($dw$).
- Least Squares (LS) Fitting: For each candidate structure proposed by the GA, the LS method calculates the precise spatial coordinates of the control points to minimize geometric deviation.
Results and Discussion
The method was validated using a symmetrical high-speed train head model decomposed into 22 distinct PDE surface patches. The optimization successfully reconstructed the full head geometry (44 patches total) with a maximum error of less than 1%.
Key findings include:
- Efficiency: The weight deformation approach reduced the number of control points by 5% to 6% compared to standard fixed-weight conversion methods for the same accuracy level.
- Data Compression: By allowing a 1% error threshold, the total number of control points for the half-model was reduced from 2,956 to 1,939—a reduction of over 34%. This significantly lowers the computational "weight" for downstream CAE simulations.
Conclusion
This research provides a robust framework for translating advanced mathematical surface models into industry-standard formats. By reducing the complexity of NURBS representations without sacrificing aerodynamic fidelity, this method facilitates more efficient design cycles for high-speed rail. Future work will explore the use of 6th-order PDEs for even greater geometric flexibility and the integration of this conversion process into real-time aerodynamic optimization loops.
Summary updated