Numerical Validation in FEA: Essential Quality & Accuracy Checks

Document Type: Technical White Paper / Best Practices
Domain: Finite Element Analysis (FEA), Solver Verification, Numerical Conditioning
Keywords: Numerical Validation, FEA Quality Checks, EPSILON, MAXRATIO, Free-Free Modal Analysis, Strain Energy Check, CG CAE

1. Introduction: From Syntactic Integrity to Numerical Reliability

Modern graphical pre-processors are equipped with comprehensive mesh diagnostic tools designed to prevent the generation of distorted or numerically ill-conditioned elements. Once a finite element model passes these preliminary geometric checks, it is submitted to the solver. Achieving an error-free run is already a significant milestone: in complex assemblies, missing constraints or mistyped Multi-Point Constraint (MPC) definitions are common pitfalls. A clean run confirms that the model is at least syntactically sound.

Most commercial FEA solvers also provide a dedicated “check-only” mode to validate input decks without executing the full matrix factorization. Historically, when even modest models required hours of CPU time, this pre-check prevented costly computational waste caused by trivial syntax errors. While modern hardware has largely made direct execution faster and more convenient, inspecting the solver’s ASCII output files remains mandatory. Beyond error logs, these files report critical element quality warnings alongside total mass, center of gravity (centroid) coordinates, and moments of inertia—data that should always be cross-checked against independent CAD or pre-processor properties.

2. Global Equilibrium: Applied Loads vs. Reaction Forces

The first and most fundamental physical verification after a static run is comparing the resultant of the applied external loads against the resultant of the constraint reaction forces. Both resultants are typically output in the three translational directions and three rotational moments relative to the global coordinate system origin.

  • Load Application Integrity: Inspecting the external load resultant verifies that forces and pressures were properly transferred to the mesh. For instance, in a closed pressure vessel under uniform internal pressure, the net resultant force must be zero; a non-zero resultant immediately reveals missing element faces or misoriented shell normals.
  • Static Equilibrium Balance: Except in specialized analyses (such as inertia relief or unrestrained mechanisms), the reaction force resultant must match the applied load resultant in magnitude with opposite sign. Minor discrepancies on the order of machine round-off are normal, whereas noticeable imbalances indicate modeling errors, grounded MPCs, or improperly constrained rigid body motions.

3. Solver Conditioning & Accuracy Indices

To quantify numerical stability during matrix factorization and equation solving, industry-standard FEA codes compute specific diagnostic indicators.

The “EPSILON” Error Index (ε)

In linear static analysis, the finite element solver assembles and solves the global equilibrium system:

{F} = [K] · {u}

Because floating-point arithmetic introduces finite numerical approximations, the computed displacement vector {u} yields a residual force vector {δF} when substituted back into the stiffness matrix:

[K] · {u} − {F} = {δF}

The dimensionless error index EPSILON (ε) measures the energy ratio between the residual force error and the externally applied work:

ε = ( {u}T · {δF} ) / ( {u}T · {F} )

The smaller the magnitude of ε (typically on the order of 10−12 to 10−9 in double-precision runs), the higher the numerical quality of the solution. Large or growing values point directly to ill-conditioned stiffness matrices.

The “MAXRATIO” Stiffness Conditioning Index

Another vital indicator of numerical conditioning is MAXRATIO, defined as the maximum ratio between the diagonal terms of the original stiffness matrix [K] and the corresponding diagonal terms of the decomposed triangular factor (LU decomposition). High values indicate severe stiffness disparities across adjacent degrees of freedom, increasing susceptibility to round-off errors.

Solvers rely on predefined MAXRATIO thresholds to detect internal mechanisms or structural lability. Physically, elevated MAXRATIO values frequently arise when elements with vastly different stiffnesses converge at a single node—such as a very thin shell connected directly to a thick solid block or a rigid element. Major aerospace OEMs strictly enforce maximum allowable MAXRATIO limits on both internal and supplier-delivered FE models.

4. Kinematic & Connectivity Verification Checks

Even when a stiffness matrix is well-conditioned, hidden modeling flaws—such as unintended grounded constraints or disconnected sub-assemblies—can compromise structural behavior. Two classical verification tests isolate these issues.

Free-Free Rigid Mode Check

Performed via an unconstrained modal analysis (with all external boundary conditions removed), the Free-Free Check verifies that a free 3D structure exhibits exactly six rigid-body modes (three translations and three rotations) at near-zero frequency.

  • More than 6 rigid modes: Indicates disconnected components, uncoupled nodes, or unintended internal mechanisms. (If a physical mechanism is intentionally modeled, its degrees of freedom must be temporarily locked prior to running the check).
  • Fewer than 6 rigid modes: Reveals hidden “grounded” constraints—such as improperly formulated MPCs or RBE elements—that artificially restrain the structure in space.

In aerospace certification workflows, strict threshold ratios between the highest rigid-body frequency and the first elastic structural frequency must be satisfied before the model is approved for downstream dynamic or static analyses.

Unit Displacement & Strain Energy Check

Particularly common in aerospace qualification checklists, the Strain Energy Check evaluates the kinematic integrity of multi-component assemblies and boundary conditions. After removing all external loads, unit translational displacements are applied separately along the three global axes, followed by unit rotations around the three axes at the constraint points.

Because rigid-body motion must produce zero internal deformation, the resulting total strain energy across the entire structure must be virtually zero (below a strict numerical threshold). Simultaneously, inspecting the displacement field confirms that every connected part translates or rotates uniformly as a rigid body, immediately exposing poorly tied interfaces, missing contact definitions, or erroneous coordinate transformations.

5. Summary of Standard FEA Validation Checks

Validation Check What It Evaluates Typical Failure / Root Cause
Mass & Inertia Audit Total mass, center of gravity, and inertia tensor vs. CAD properties. Wrong material density units, missing non-structural mass, or duplicate meshes.
Load vs. Reaction Balance Global equilibrium between applied external forces/moments and reactions. Grounded MPCs, misoriented pressure normals, or missing load surfaces.
EPSILON (ε) Index Energy norm of the residual force vector {δF} relative to applied work. Severe matrix ill-conditioning or extreme stiffness jumps causing round-off loss.
MAXRATIO Index Ratio of original stiffness diagonal terms to LU decomposed factors. Internal lability, unconstrained DOFs, or abrupt thickness/material transitions.
Free-Free Modal Check Number and frequency of rigid-body modes in an unconstrained model. Disconnected parts (>6 zero modes) or unintended grounding (<6 zero modes).
Strain Energy Check Internal strain energy generated under enforced unit rigid translations/rotations. Over-constrained boundaries, defective MPC equations, or inconsistent local frames.
Table 1: Core numerical and kinematic validation checks for structural finite element models.

Historical Perspective & Modern Industrial Practice: Most of these diagnostic indices originated in the early days of computational mechanics, when solvers ran on 8-bit or 16-bit architectures and accumulated round-off errors could easily invalidate an entire simulation. Although modern 64-bit double-precision solvers are far less vulnerable to numerical truncation, rigorous model verification remains indispensable. In high-reliability sectors such as aerospace and defense, completing a formal numerical validation checklist is still a mandatory prerequisite before any FE model is cleared for structural certification.

Further Reading & Theoretical Reference: This article is adapted from Chapter 9 of the book “Computational Structural Engineering” by Claudio Gianini. To explore complete FEA verification workflows, solver theory, and industrial best practices, discover the full book on CG CAE or on Amazon.