🎓 Lesson 24 D5

Calibrating UDEC Models Using Field Convergence Data

Calibrating UDEC models means adjusting the computer simulation’s rock properties until its predicted ground movement matches what was actually measured in the field.

🎯 Learning Objectives

  • Analyze field convergence time-series data to identify dominant deformation mechanisms
  • Calculate equivalent joint normal and shear stiffness values required to match measured displacement magnitudes in UDEC
  • Design a systematic calibration workflow using parameter sensitivity analysis and objective error metrics (e.g., RMSE)
  • Explain the limitations of UDEC calibration when rock mass heterogeneity or time-dependent behavior is poorly constrained

📖 Why This Matters

In underground mines and open-pit highwalls, inaccurate ground control predictions can lead to catastrophic failures, unplanned stopes, or excessive support costs. UDEC is widely used for discontinuous rock mass modeling—but a 'pretty' simulation is useless if it doesn’t match reality. Calibration bridges the gap between theory and practice: it transforms UDEC from a qualitative sketchpad into a decision-support tool engineers trust for stope design, pillar sizing, and blast-induced damage assessment.

📘 Core Principles

UDEC treats rock as an assembly of rigid or deformable blocks separated by interfaces (joints). Calibration hinges on three interdependent principles: (1) Observability — only parameters influencing measurable outputs (e.g., convergence at monitoring points) should be adjusted; (2) Identifiability — multiple parameters must not trade off against each other (e.g., reducing joint cohesion while increasing friction may yield similar displacements); and (3) Parsimony — use the minimum number of adjustable parameters needed to achieve acceptable fit (per Occam’s razor). Field convergence data provides boundary-condition constraints that anchor the model in physical reality, enabling discrimination between competing geomechanical interpretations (e.g., joint-dominated vs. intact-rock yielding).

📐 Root-Mean-Square Error (RMSE) for Calibration Quality

RMSE quantifies the average magnitude of mismatch between simulated and field-measured convergence. It is the primary objective function minimized during calibration. Lower RMSE indicates better fidelity — but values < 1.5 mm are typically targeted for high-precision instrumentation (e.g., vibrating-wire extensometers), while > 5 mm suggests fundamental model inadequacy.

RMSE

RMSE = √[ Σ(y_i,sim − y_i,field)² / n ]

Quantifies average absolute deviation between simulated and field-measured convergence values.

Variables:
SymbolNameUnitDescription
y_i,sim Simulated convergence at point i mm UDEC-predicted displacement (normal to excavation surface) at monitoring location i
y_i,field Field-measured convergence at point i mm Instrument-recorded displacement at identical location i
n Number of measurement points dimensionless Total count of paired simulation/field data points
Typical Ranges:
High-precision extensometer networks: 0.3 – 1.5 mm
LiDAR-based surface scan (open pit): 2 – 8 mm

💡 Worked Example

Problem: Given field convergence measurements at 4 locations: [2.1, 3.8, 1.9, 4.2] mm; corresponding UDEC-simulated values: [2.4, 3.3, 2.2, 4.0] mm.
1. Step 1: Compute residuals: (2.4−2.1)² = 0.09, (3.3−3.8)² = 0.25, (2.2−1.9)² = 0.09, (4.0−4.2)² = 0.04
2. Step 2: Sum residuals = 0.09 + 0.25 + 0.09 + 0.04 = 0.47
3. Step 3: Divide by n = 4 → 0.47 / 4 = 0.1175; take square root → √0.1175 ≈ 0.343 mm
Answer: The RMSE is 0.34 mm, which falls well within the target range of < 1.5 mm for high-fidelity calibration.

🏗️ Real-World Application

At the Cadia East block cave (NSW, Australia), UDEC was calibrated using 32 multi-point borehole extensometer (MPBX) datasets across a 40-m-deep extraction zone. Initial simulations overestimated crown convergence by 40% due to underestimating joint normal stiffness. Through iterative adjustment—guided by Morris screening sensitivity analysis—the joint normal stiffness was increased from 12 MPa/m to 28 MPa/m, reducing RMSE from 3.1 mm to 0.9 mm. This calibrated model subsequently predicted stable drawpoint spacing with ±12% accuracy in subsequent production rounds, directly informing draw control protocols.

✏️ Calibration Sensitivity Exercise

You are given convergence data from a 15-m-deep drift: measured displacements at 3 m, 6 m, 9 m, 12 m, and 15 m from face are [0.8, 1.7, 2.9, 4.1, 4.8] mm. A preliminary UDEC run yields [0.6, 1.4, 2.3, 3.5, 4.0] mm. Using RMSE, quantify the current misfit. Then, determine whether increasing joint cohesion (c_j) or joint friction angle (φ_j) would more effectively reduce the underprediction—based on UDEC’s known sensitivity ranking: φ_j contributes ~65% of convergence variance, c_j contributes ~22% (per Itasca 2022 benchmark study).

📋 Case Connection

📋 Underground Copper Mine Pillar Recovery Optimization

Post-extraction pillar instability threatening surface infrastructure

📚 References