Mesh Convergence vs. Adaptive Refinement: Finding the Accuracy Sweet Spot in Ansys

The Cost of Over-Meshing

In structural analysis, the most common source of error isn't the physics model—it's the mesh. If your elements are too large, you miss peak stresses; if they are too small, you waste hours of compute time and risk crashing your solver due to memory exhaustion. The goal is mesh independence: the point where adding more elements no longer significantly changes your results.

When seeking this point, engineers typically choose between two paths: Manual Mesh Convergence studies or Adaptive Mesh Refinement (AMR). The right choice depends on whether you need a predictable validation trail for a report or a fast path to a solution for a complex geometry.

Manual Mesh Convergence: The Gold Standard for Validation

Manual convergence is a systematic process of refining the global element size and tracking a specific Key Performance Indicator (KPI), such as maximum Von Mises stress or total displacement. This approach is essential for certification and high-stakes validation because it provides a clear, linear evidence trail.

• Process: You start with a coarse mesh, record the result, halve the element size, and repeat until the change in the result falls below a predefined threshold (e.g., <2%).
• Predictability: Because the refinement is global, the mesh remains structured and predictable, reducing the risk of unexpected numerical instability.
• The Downside: Global refinement increases the total element count exponentially. Doubling the resolution in a 3D model can increase the degrees of freedom (DOF) by a factor of eight, leading to massive RAM requirements.

Adaptive Mesh Refinement (AMR): The Efficiency Shortcut

Adaptive Mesh Refinement (AMR) automates the search for accuracy. Instead of shrinking every element in the model, Ansys uses error estimators to identify regions with high gradients—areas where the solution changes rapidly over a short distance.

AMR iteratively refines only the elements in these "hot spots." This allows you to maintain a coarse mesh in low-stress regions while achieving high precision at stress concentrations. While faster, AMR requires caution; if not constrained, it can create highly skewed elements (high aspect ratios) that may introduce artificial numerical errors.

Worked Example: Cantilever Beam with a Fillet

Consider a cantilever beam with a small fillet at the fixed support. The primary goal is to capture the peak stress at the fillet radius.

Scenario A: Manual Global Refinement

1. Set a global element size of 5mm. Result: Peak Stress = 210 MPa.
2. Set a global element size of 2.5mm. Result: Peak Stress = 245 MPa.
3. Set a global element size of 1.25mm. Result: Peak Stress = 252 MPa.
4. Set a global element size of 0.625mm. Result: Peak Stress = 254 MPa.

The result has converged. However, the final iteration may have 1,000,000 elements, most of which are in the middle of the beam where nothing is happening.

Scenario B: Adaptive Refinement

1. Start with the 5mm coarse mesh.
2. Enable Adaptive Mesh in the Ansys Mechanical Model tree.
3. The solver identifies the fillet as the high-gradient region and refines only those elements.
4. The solver reaches the 254 MPa result with only 150,000 elements.

Trade-offs and Technical Limitations

Feature	Manual Convergence	Adaptive Refinement (AMR)
Compute Cost	High (Exponential growth)	Low (Targeted growth)
Validation	Easy to document/plot	Black-box automation
Element Quality	Consistent/Predictable	Risk of skewness/distortion
Convergence Speed	Slow (Manual iterations)	Fast (Automated loops)

Critical Limitation: Convergence in displacement is almost always achieved faster than convergence in stress. If you only track displacement, you may prematurely stop your study and significantly under-predict the peak stress at a fillet or hole.

Verification and Action Plan

To ensure your results are mesh-independent, follow this workflow before finalizing any simulation report:

1. Define your KPI: Choose the most critical scalar value (e.g., Max Equivalent Stress).
2. Run a Convergence Plot: Plot the KPI against the number of elements. The curve should flatten out (asymptote).
3. Check Element Quality: If using AMR, check the Mesh Metric (e.g., Element Quality or Aspect Ratio) to ensure the adaptive process hasn't created distorted elements.
4. Cross-Reference: For simple geometries, compare your converged result against a known analytical solution (like the beam equation) to ensure the baseline physics are correct.