Regression Based Approach for Robust Finite Element Analysis on Arbitrary Grids. LDRD Final Report

Paul Kuberry, Pavel Bochev, Jacob Koester, Nathaniel Trask2020 · DOI

TL;DR

This paper introduces a novel finite element method that decouples shape function construction from mesh quality, enabling robust simulations on distorted grids. By using a meshless approach based on Generalized Moving Least Squares (GMLS), the method achieves optimal convergence rates and significant improvements in accuracy and stable timesteps for challenging problems like linear elasticity and Stokes flow, outperforming traditional methods on poor-quality meshes.

What Problem Does This Paper Address?

Traditional Finite Element Methods (FEM) rely on the mesh not only to define the geometry but also to construct the basis functions (shape functions). This creates a critical dependency: poor mesh quality (e.g., highly distorted or "sliver" elements) directly degrades the accuracy and stability of the simulation. While generating high-quality meshes is computationally expensive and often impossible for complex geometries or in scenarios like large-deformation mechanics, traditional FEM struggles with these "bad" elements. This paper addresses the challenge of performing accurate and robust finite element analysis on arbitrary, and often poor-quality, meshes.

Methods

The core innovation is a new approach to constructing finite element shape functions that are independent of the underlying mesh geometry. This is achieved by leveraging Generalized Moving Least Squares (GMLS) regression techniques, a meshless method. Instead of defining shape functions based on element vertices, GMLS constructs them locally using a cloud of "particles" (points) and a kernel function. This effectively separates the role of the mesh to solely representing the domain and facilitating integration, while the approximation quality is determined by the GMLS basis.

The resulting non-conforming, discontinuous piecewise polynomial finite element space is then naturally integrated within Discontinuous Galerkin (DG) or Interior Penalty (IP) frameworks, specifically using a Symmetric Interior Penalty Galerkin (SIPG) method. This allows the authors to benefit from the established stability and approximation theory of these methods. The GMLS problem is solved locally for each element, and the integration of polynomial products is efficient, often requiring only standard numerical quadrature.

Datasets

Not discussed in the paper. The paper focuses on numerical experiments using manufactured solutions on various mesh configurations, including highly distorted annuli and a rectangular domain for Stokes flow.

Key Insights and Ideas

  • Decoupling Mesh and Basis: The fundamental insight is that shape functions don't have to be tied to element geometry. By using a meshless approach (GMLS), the quality of the simulation becomes independent of mesh quality.
  • Robustness on Distorted Meshes: The method shows remarkable resilience to poor mesh quality, outperforming traditional Continuous Galerkin (CG-FEM) on highly distorted meshes, especially in elasticity.
  • Improved Timestep Stability: For linear elasticity, the new method allows for significantly larger stable timesteps, which is crucial for dynamic simulations.
  • Flexibility in Stabilization: The non-conforming nature and meshless basis allow for more relaxed definitions of stabilization parameters (like the penalty parameter in SIPG), contributing to improved stability.
  • General Applicability: The approach is demonstrated to be effective across various problem types, including scalar elliptic PDEs, linear elasticity, and incompressible Stokes flow.

Key Results

  • Accuracy Improvement: For linear elasticity on highly deformed meshes, the GMLS+SIPG approach showed a greater than 5x improvement in accuracy (L² norm) and 5.73x improvement in the H¹ seminorm compared to CG-FEM.
  • Timestep Enlargement: In linear elasticity, the method achieved an 8.38x larger stable timestep for a highly deformed mesh compared to CG-FEM when using a lumped mass matrix.
  • Optimal Convergence: Numerical results consistently demonstrated optimal asymptotic rates of convergence for all problem types studied, even on poor-quality meshes.
  • Reduced Mesh Dependence: While CG-FEM's error increases rapidly with mesh distortion, the GMLS+SIPG approach deteriorates much more slowly.

Limitations and Future Work

  • Absolute Error on Good Meshes: On very high-quality meshes, the GMLS+SIPG approach might not achieve as low an absolute error as well-established CG-FEM methods, though its convergence rates remain optimal.
  • Future Work: The authors plan to extend the approach to:
  • Transient hyperelasticity and plasticity.
  • Explore alternative stabilization techniques, such as Variational Multiscale (VMS).
  • Investigate further improvements to the GMLS kernel, search size, and particle set choices.

Summary updated

Abstract

This report summarizes the work performed under a one-year LDRD project aiming to enable accurate and robust numerical simulation of partial differential equations for meshes that are of poor quality. Traditional finite element methods use the mesh to both discretize the geometric domain and to define the finite element shape functions. The latter creates a dependence between the quality of the mesh and the properties of the finite element basis that may adversely affect the accuracy of the discretized problem. In this project, we propose a new approach for defining finite element shape functions that breaks this dependence and separates mesh quality from the discretization quality. At the core of the approach is a meshless definition of the shape functions, which limits the purpose of the mesh to representing the geometric domain and integrating the basis functions without having any role in their approximation quality. The resulting non-conforming space can be utilized within a standard discontinuous Galerkin framework providing a rigorous foundation for solving partial differential equations on low-quality meshes. We present a collection of numerical experiments demonstrating our approach in a wide range of settings: strongly coercive elliptic problems, linear elasticity in the compressible regime, and the stationary Stokes problem. We demonstrate convergence for all problems and stability for element pairs for problems which usually require inf-sup compatibility for conforming methods, also referring to a minor modification possible through the symmetric interior penalty Galerkin framework for stabilizing element pairs that would otherwise be traditionally unstable. Mesh robustness is particularly critical for elasticity, and we provide an example that our approach provides a greater than 5x improvement in accuracy and allows for taking an 8x larger stable timestep for a highly deformed mesh, compared to the continuous Galerkin finite element method. The report concludes with a brief summary of ongoing projects and collaborations that utilize or extend the products of this work.
BibTeX
@misc{kuberry2020regression,
  author = {Paul Kuberry and Pavel Bochev and Jacob Koester and Nathaniel Trask},
  title = {{Regression Based Approach for Robust Finite Element Analysis on Arbitrary Grids. LDRD Final Report}},
  year = {2020},
  doi = {10.2172/1669732},
}
Figure