Abstract
With the increasing need for volumetric B-spline representations and the lack of methodologies for creating semi-structured volumetric B-spline representations from B-spline Boundary Representations (B-Rep), hybrid approaches combining semi-structured volumetric B-splines and unstructured Bézier tetrahedra have been introduced. This includes a pipeline that transforms a trimmed B-spline B-Rep first to an untrimmed Hybrid B-Rep (HB-Rep) and then to a Hybrid Volume Representation (HV-Rep). Generally, the effect of $h$-refinement has not been considered over B-spline hybrid representations. Standard refinement approaches to tensor product B-splines and subdivision of Bézier triangles and tetrahedra must be adapted to this representation. In this paper, we analyze possible types of $h$-refinement of the HV-Rep. The revised and trim basis functions for HB- and HV-rep depend on a partition of knot intervals; therefore, a naive $h$-refinement can change basis functions in unexpected ways. This paper analyzes the effect of $h$-refinement in reducing error, discusses different $h$-refinement strategies, compares their respective trade-offs, and provides recommendations for different use cases.
Introduction: The CAD-to-Simulation Gap
In modern engineering, a significant bottleneck exists between design and analysis. Computer-Aided Design (CAD) models typically utilize Boundary Representations (B-reps) to define the "skin" of an object using B-splines or NURBS. However, simulation techniques like Isogeometric Analysis (IGA) require volumetric (solid) models that are perfectly watertight. Trimming—the process of using one surface to cut another to create complex shapes—is mathematically complex and often results in gaps or overlaps, making the conversion to a simulation-ready volume difficult.
The Challenge of Trimming in B-Reps
Trimming curves in CAD are often approximations, leading to models that are not inherently watertight. To address this, previous research introduced Hybrid Boundary Representations (HB-rep) and Hybrid Volume Representations (HV-rep). These representations use "revised basis functions" that fade to zero at trim lines and "trim basis functions" to specifically represent the cut edge, effectively sealing the gaps. However, as simulation requirements demand higher resolution, the process of $h$-refinement (adding knots to increase detail) becomes non-trivial. Naive refinement can alter the specialized basis functions in ways that introduce artifacts or break the watertight properties of the model.
Methodology: Refinement Strategies for Hybrid Models
This research investigates how to correctly apply $h$-refinement to these hybrid structures. The authors compare two primary strategies:
- Refine-then-Convert: Knots are added to the original CAD B-rep before it is transformed into the hybrid representation.
- Direct Hybrid Refinement: Knots are inserted directly into an existing HB-rep or HV-rep.
The process involves a "remainder partitioning" step, using a recursive divide-and-conquer approach to split regions near trim lines into smaller, manageable rectangular and triangular pieces. This ensures that the transition from the boundary to the interior volume remains consistent even as the mesh density increases.
Results and Analysis
The study utilized several synthetic mechanical models, including a tetrapod and an L-shaped trimmed surface, to evaluate the refinement strategies.
Error Reduction and Convergence
The results demonstrate that $h$-refinement effectively reduces surface error. For instance, in a test case involving an L-shaped hole, refinement reduced the maximum surface error from 0.035 to 0.010. As the model is refined, the "banded" region—where the basis functions are modified to handle the trim—narrows, and the hybrid model converges toward the original CAD geometry. However, a "trimming bottleneck" was identified: the ultimate accuracy is limited by the precision of the initial trimming curve.
The "Wrinkle" Effect
A counter-intuitive finding was the emergence of "wrinkles" or visual artifacts in highly refined meshes near sharp angles. This occurs because the adjustment algorithms designed to prevent surface intersections can "overshoot" when control points are too densely packed. This suggests that parameters like the $\epsilon$ separation factor must be dynamically tuned relative to the refinement level.
Conclusion and Future Directions
The proposed framework successfully maintains $C^0$ continuity across trimmed interfaces after multiple levels of knot insertion. While $h$-refinement is a powerful tool for increasing resolution in hybrid volumetric models, it requires careful management of parameter sensitivity and interior tetrahedral mesh complexity. Future work will focus on higher-order approximations of trimming curves and improving the efficiency of interior volume "stuffing" as the boundaries are refined.
Summary updated