🎓 Lesson 17
D5
Validation Against Analytical Solutions and Grid Independence
Checking that a computer simulation gives the right answer by comparing it to exact math solutions, and making sure the answer doesn’t change when we use a finer grid.
🎯 Learning Objectives
- ✓ Calculate discretization error using Richardson extrapolation for 1D steady conduction
- ✓ Analyze grid convergence behavior by plotting solution norms versus mesh size
- ✓ Explain the distinction between verification (code correctness) and validation (model fidelity)
- ✓ Apply the Grid Convergence Index (GCI) method per ASME V&V 20-2018 to quantify uncertainty
- ✓ Design a minimal three-grid refinement study for a given conduction problem
📖 Why This Matters
In mining and blasting engineering, accurate thermal modeling predicts rock spalling, explosive sensitivity changes, and post-blast heat dissipation—critical for safety and efficiency. A flawed simulation may overestimate cooling rates by 40%, leading to premature equipment failure or unsafe re-entry into heated blast zones. Without validation and grid independence, even high-fidelity software produces misleading results—making these checks not optional, but mandatory before any design decision.
📘 Core Principles
Verification confirms that the code solves the equations correctly (‘Are we solving the equations right?’); validation asks whether the equations themselves represent reality (‘Are we solving the right equations?’). Analytical solutions—such as the exact temperature distribution in a 1D slab with constant conductivity and Dirichlet boundaries—serve as gold-standard benchmarks. Grid independence hinges on asymptotic convergence: as cell size h → 0, the solution should approach the true value at a predictable rate (e.g., O(h²) for second-order schemes). The Grid Convergence Index (GCI) quantifies the uncertainty band due to discretization, enabling objective confidence statements—essential when simulating transient heating in blast-hole walls or borehole thermal resistance.
📐 Grid Convergence Index (GCI)
The GCI estimates the discretization error bound for a computed solution using three successively refined grids. It accounts for observed order of convergence and relative grid spacing, following ASME V&V 20-2018. Used after computing solutions on coarse (h₁), medium (h₂), and fine (h₃) grids.
Grid Convergence Index (GCI)
GCI = F_s × |T_fine − T_medium| / |T_fine| × 1 / (r^p − 1)Quantifies the discretization uncertainty bound for the fine-grid solution per ASME V&V 20-2018.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| F_s | Safety factor | dimensionless | 1.25 for two-grid, 1.0 for three-grid studies (ASME default) |
| T_fine | Fine-grid solution | °C | Result on highest-resolution mesh |
| T_medium | Medium-grid solution | °C | Result on intermediate-resolution mesh |
| r | Refinement ratio | dimensionless | h_medium / h_fine |
| p | Observed order | dimensionless | Computed from three-grid data |
Typical Ranges:
Industrial heat transfer simulations: 0.01% – 0.5% of solution magnitude
Research-grade validation: < 0.05% of solution magnitude
💡 Worked Example
Problem: A 1D steady conduction model of a 1 m copper slab (k = 400 W/m·K) with T(0)=100°C and T(1)=25°C yields temperatures at x=0.5 m: T₁=63.2°C (h₁=0.2 m), T₂=62.8°C (h₂=0.1 m), T₃=62.7°C (h₃=0.05 m). Assume theoretical order p = 2. Calculate GCI for the fine-grid solution.
1.
Step 1: Compute observed order p using three-grid formula: p = ln[(T₁−T₂)/(T₂−T₃)] / ln(h₁/h₂) = ln[(63.2−62.8)/(62.8−62.7)] / ln(0.2/0.1) = ln[0.4/0.1]/ln(2) = ln(4)/0.693 ≈ 2.0.
2.
Step 2: Compute relative error ε₃₂ = |T₃ − T₂|/|T₃| = |62.7 − 62.8|/62.7 ≈ 0.001595.
3.
Step 3: Apply GCI formula: GCI₃ = Fₛ × ε₃₂ / (r^p − 1), where r = h₂/h₃ = 0.1/0.05 = 2, Fₛ = 1.25 (ASME recommended safety factor): GCI₃ = 1.25 × 0.001595 / (2² − 1) = 0.001994 / 3 ≈ 0.000665 → 0.0665%.
4.
Step 4: Express as absolute uncertainty: 0.000665 × 62.7 ≈ ±0.042°C.
Answer:
The GCI for the fine-grid solution is ±0.042°C, indicating excellent grid independence (well below typical thermal measurement uncertainty of ±0.5°C).
🏗️ Real-World Application
At the Bingham Canyon Mine (Rio Tinto), engineers modeled heat diffusion into fractured porphyry rock after hot-gas venting from a misfired ANFO blast. Using ANSYS Fluent, they performed a three-grid study (2 mm, 1 mm, 0.5 mm element size) on a 2D axisymmetric section of the blast hole wall. Validation used the analytical solution for radial conduction in an infinite cylinder with surface flux. GCI analysis confirmed <0.1°C discretization error at 0.5 mm resolution—enabling confident prediction of 32-hour safe re-entry time (validated by IR thermography). Without grid independence, the coarse-grid model overpredicted cooling by 22%, risking premature personnel entry.
✏️ Student Exercise
You are modeling steady 1D conduction in a 0.3 m thick stainless steel plate (k = 15 W/m·K) with T_left = 200°C and T_right = 50°C. The analytical solution is linear: T(x) = 200 − 500x. You compute T at x = 0.15 m using finite differences: T₁ = 124.8°C (Δx = 0.1 m), T₂ = 125.1°C (Δx = 0.05 m), T₃ = 125.0°C (Δx = 0.025 m). (a) Calculate the observed order of convergence p. (b) Compute GCI for T₃ using Fₛ = 1.25. (c) Does the result satisfy ASME’s ‘acceptable’ criterion (GCI < 1% of solution magnitude)?