An Adaptive Hybrid Stress Transition Quadrilateral Finite Element Method for Linear Elasticity

Feiteng Huang, Xiaoping Xie, Chen-Song Zhang

TL;DR

This paper introduces a new way to perform adaptive mesh refinement on quadrilateral grids for linear elasticity problems. By designing specialized 5- to 7-node "transition elements" and a non-recursive refinement algorithm, the authors created a method that handles complex geometries efficiently while remaining completely immune to "Poisson-locking"—the numerical stiffening that usually plagues simulations of nearly incompressible materials.

What Problem Does This Paper Address?

In finite element analysis (FEA), quadrilateral meshes are often preferred over triangular ones for their superior accuracy in many engineering applications. However, they are a nightmare to refine adaptively. When you split one quad into four, you create "hanging nodes" where the new vertices meet the edges of unrefined neighbors. Standard practice is to either force neighboring quads to split (propagating refinement unnecessarily) or use complex constraints to "tie" the nodes together.

Furthermore, many standard elements suffer from Poisson-locking. As a material's Poisson’s ratio approaches 0.5 (becoming nearly incompressible, like rubber), the numerical model becomes artificially stiff, leading to massive errors. This paper seeks a unified solution: an adaptive framework that handles hanging nodes naturally via transition elements without succumbing to locking.

Methods

The authors utilize the Hybrid Stress Finite Element Method, based on the Hellinger-Reissner variational principle. Unlike standard displacement-based methods, hybrid methods treat both displacement and stress as independent variables, which provides a more flexible framework for avoiding locking.

Key technical contributions include:

  • New Transition Elements: They propose 5-node, 6-node, and 7-node quadrilateral elements. These elements "absorb" hanging nodes by incorporating them into the element's own geometry and interpolation functions.
  • Locking-Free Proof: The authors mathematically derive a priori error estimates for the 5-node element, proving that the error bound is independent of the Lamé constant $\lambda$. This confirms the method is "locking-free."
  • Non-Recursive AMR Algorithm: Most adaptive algorithms rely on a "refinement tree" (storing parent-child relationships between elements), which is memory-intensive and complex to code. The authors adapt a non-recursive approach (originally for triangles) to quadrilaterals, using a special data structure that only requires vertex coordinates and connectivity.

Datasets

The method was evaluated using several classic benchmarks in computational mechanics:

  1. L-shaped Domain: Used to test the adaptive Poisson solver against the well-known iFEM library.
  2. Moving Circle: A test to track a moving interface, evaluating the efficiency of the refinement and coarsening algorithms.
  3. Pure Bending Cantilever Beam: A standard benchmark for Poisson-locking.
  4. Cracked Plane Strain Panel: A high-stress singularity problem used to demonstrate the convergence rate of the adaptive hybrid stress elements.

Key Insights and Ideas

  • The "Hanging Node" as a Feature, Not a Bug: Instead of seeing hanging nodes as a constraint problem to be solved with "penalty methods" or "master-slave" eliminations, the authors treat them as additional nodes in a transition element. This maintains mesh conformity without over-refining the grid.
  • Stress Mode Optimization: The 5-node element uses a 7-parameter stress mode. The authors show that by carefully choosing these modes, they can satisfy the "inf-sup" stability condition, which is the mathematical "gold standard" for ensuring a mixed finite element method won't fail.
  • Data Structure Simplicity: By avoiding explicit refinement trees, the algorithm becomes much easier to implement in high-level languages like MATLAB, achieving linear computational complexity similar to state-of-the-art triangular mesh generators.

Key Results

  • Efficiency: In the L-shaped domain test, the new quadrilateral algorithm reached the target accuracy in roughly half the time (35 seconds vs. 79 seconds) and with fewer Degrees of Freedom (DOF) than the standard triangular bisection method in iFEM.
  • Locking Resistance: In the cantilever beam test, as the Poisson’s ratio was pushed to the extreme limit (0.49999999999), the relative error remained stable and small (approx. 1% for displacement), whereas standard elements would typically fail or show massive error spikes.
  • Optimal Convergence: For the crack problem, the adaptive method showed a much steeper error reduction curve compared to uniform refinement, proving it effectively "targets" the stress singularity at the crack tip.

Limitations and Future Work

  • 2D Focus: The current work is strictly limited to 2D plane elasticity. While the authors mention that the non-recursive algorithms have been extended to 3D for triangles, implementing these 5-7 node transition quads as 3D "transition hexahedra" would be significantly more complex.
  • Mesh Regularity: The proof for the 5-node element relies on a "shape-regularity" hypothesis. Extremely distorted or high-aspect-ratio quads might still pose stability challenges.
  • Implementation Complexity: While the data structure is "simpler" in terms of memory, the logic for categorizing five different types of edge refinement (Type 1 to Type 5) requires very careful implementation.

Summary updated

Abstract

In this paper, we discuss an adaptive hybrid stress finite element method on quadrilateral meshes for linear elasticity problems.To deal with hanging nodes arising in the adaptive mesh refinement, we propose new transition types of hybrid stress quadrilateral elements with 5 to 7 nodes.In particular, we derive a priori error estimation for the 5-node transition hybrid stress element to show that it is free from Poisson-locking, in the sense that the error bound in the a priori estimate is independent of the Lamé constant λ.We introduce, for quadrilateral meshes, refinement/coarsening algorithms, which do not require storing the refinement tree explicitly, and give an adaptive algorithm.Finally we provide some numerical results.
BibTeX
@article{huang2016adaptive,
  author = {Feiteng Huang and Xiaoping Xie and Chen-Song Zhang},
  title = {{An Adaptive Hybrid Stress Transition Quadrilateral Finite Element Method for Linear Elasticity}},
  journal = {Journal of Computational Mathematics},
  year = {2016},
  doi = {10.4208/jcm.1511-m4496},
}
Figure