MECH60561 Chap.5 Mesh Design, Refinement and Quality
Mesh Design, Refinement and Quality
Mesh Design, Refinement and Quality treats discretisation as a controlled error study. h-refinement reduces element size, while p-refinement enriches interpolation. Element distortion, aspect ratio and mapping quality affect numerical accuracy, and local gradients determine where extra resolution is valuable.
The chapter distinguishes genuine convergence from a smooth-looking contour, separates global and local response measures and diagnoses locking, hourglass behaviour and singular peak stress through targeted comparisons. Mesh generation routes for designed, scanned and image-based geometry show why quality control begins before the solver and why the output definition must remain fixed across runs.
A stopping rule links remaining numerical change to the tolerance controlling the engineering decision.
What this chapter covers
- 01
H-Refinement
- 02
P-Refinement
- 03
Element distortion
- 04
Aspect ratio and orientation
- 05
Output-specific convergence
- 06
Locking and hourglass modes
- 07
Local stress gradients and singularities
Worked example · free
Plan a convergence study around a hole
- 3Calculate changes in the global displacement across successive meshes.
- 4Compare the local stress using the same physical location and extraction method.
- 3State separate conclusions for global and local response and propose targeted refinement.
Key terms
- h-refinement
- Mesh refinement that reduces characteristic element size.
- p-refinement
- Refinement that increases the polynomial richness of element interpolation.
- Hourglass mode
- A nonphysical low-energy deformation pattern that can occur with under-integrated elements.
Mesh Design, Refinement and Quality FAQ
Does a finer mesh always improve the result?
A suitable finer mesh can reduce discretisation error, but it cannot repair incorrect loads, supports, material data or formulation. Poorly shaped refinement can even worsen conditioning. Refine around the response and mechanism of interest, then compare the same output consistently.
How do I recognise a singular stress?
A stress at a point load, sharp re-entrant corner or idealised fixed edge may continue increasing as element size falls. Inspect the physical idealisation, compare values away from the singular point and report a meaningful averaged or path quantity when appropriate.
How is locking different from a coarse mesh?
Locking is a spurious stiffness mechanism caused by an interpolation constraint, not merely insufficient element count. Diagnose it with a benchmark and a controlled comparison using better element quality, higher order or an appropriate formulation while preserving the physical model.
Exam move
Choose a response quantity before creating a refinement sequence. Record element type, local size, order, quality measure and extraction location for every run. Compare displacement, reaction, energy or consistently recovered stress rather than relying on contour smoothness. Practise h and p refinement as separate changes so you can explain which approximation feature is being improved.
Inspect aspect ratio and orientation relative to the field gradient, and recognise that long thin elements are not automatically wrong when aligned with a smooth directional response. Build a diagnostic table for excessive stiffness, checkerboard motion, singular systems and rising local peaks. For each symptom, state a mechanism, a test that could refute it and a targeted correction.
Review mesh generation for clean CAD, reverse-engineered surfaces and image-based microstructures, including seed points, density bias and iterative quality improvement. Finish convergence answers by stating what stabilised, where it was measured and which modelling uncertainties remain. Never use one converged reaction to certify every response in the model. Plan a refinement study on paper before opening the mesher.
Draw the expected high-gradient regions around holes, notches, contacts and load introductions, then decide where transitions can remain gradual. Select a baseline mesh that is coarse enough to reveal change but credible enough to solve the physical problem. Keep geometry, materials, supports and loading identical across the sequence so the comparison isolates discretisation.
Track at least one global response and one local response when both affect the decision. Global displacement may stabilise well before a local stress, and a balanced reaction may change almost not at all. Use a table with mesh identifier, element count, characteristic size, worst quality, output value and percentage change. Percentage change is useful only when the same location and extraction rule are used.
If nodal averaging changes between runs, the apparent convergence may describe the display setting rather than the underlying integration-point field. Examine a suspected singularity with a path plot or values at fixed distances from the idealised point. Explain why the exact peak may lack a finite target and choose an engineering quantity that can support the decision.
For locking, compare against a bending or incompressibility benchmark while changing formulation or interpolation in a controlled way. For hourglass behaviour, inspect deformation pattern and stabilisation or artificial energy. Do not cure either by arbitrary stiffness. Review mesh creation routes as scenario answers: seed and fill a clean design, repair and mesh scanned geometry, or translate image voxels for microstructure.
Name the quality iteration and the user controls over size and bias. End the study with a stopping rule linked to decision sensitivity. A small numerical change is meaningful only when it is small compared with the tolerance, material variability and other modelling uncertainty. Report the unresolved limitations instead of declaring the entire model converged.