š Lesson 7
D4
Monte Carlo Simulation for Probabilistic FoS
Monte Carlo simulation is a computer method that runs thousands of random trials to show how likely different outcomes are when dealing with uncertain inputs like rock strength or water pressure in slope stability.
šÆ Learning Objectives
- ā Calculate probability of failure (Pf) from Monte Carlo-simulated FoS distribution
- ā Analyze sensitivity of FoS to individual parameters using correlation coefficients and partial rank correlation coefficients (PRCC)
- ā Design a statistically robust simulation by selecting appropriate sample size (N ā„ 5000), distribution types (e.g., lognormal for cohesion), and correlation structures
- ā Explain the difference between aleatory and epistemic uncertainty in slope stability modeling
- ā Apply Latin Hypercube Sampling (LHS) to improve convergence efficiency compared to simple random sampling
š Why This Matters
Traditional slope stability analysis assumes single 'best-estimate' values for soil and rock propertiesāignoring natural variability and measurement uncertainty. In real mining operations, a 15% underestimation of cohesion or unanticipated pore pressure buildup can turn a 'safe' design into a catastrophic landslide. Monte Carlo simulation transforms qualitative risk judgments into quantitative metricsālike '95% confidence that Pf < 3%'āenabling engineers to justify slope angles, drainage investments, and monitoring strategies to regulators and mine management.
š Core Principles
Monte Carlo simulation rests on three pillars: (1) Probabilistic modeling of inputsāassigning realistic distributions (e.g., normal for unit weight, lognormal for cohesion, truncated normal for Ļ) based on site data and judgment; (2) Propagation through a deterministic modelārunning each sampled parameter set through a limit equilibrium (e.g., Bishop) or numerical (e.g., Phase2) solver; (3) Statistical inferenceāaggregating N outputs to compute Pf, reliability index (Ī²), and sensitivity measures. Critical refinements include handling parameter correlations (e.g., negative cāĻ correlation in clays), nonlinearity-induced skewness in FoS output, and convergence diagnostics (e.g., Gelman-Rubin statistic for MCMC variants).
š Probability of Failure & Reliability Index
The core output metric is Pf = proportion of simulations where FoS < 1.0. For normally distributed FoS, the reliability index Ī² = Ī¦ā»Ā¹(1 ā Pf) links probability to standard deviations from failure. While Ī² > 3.5 is often targeted (Pf < 0.00023), mining slopes commonly accept Ī² ā„ 2.3 (Pf ā¤ 0.01) under controlled monitoring.
š” Worked Example
Problem: A Monte Carlo simulation of 10,000 iterations yields FoS values with mean = 1.42, standard deviation = 0.28, and 217 simulations with FoS < 1.0. Assume FoS is approximately normal.
1.
Step 1: Compute empirical Pf = number of failures / total samples = 217 / 10,000 = 0.0217
2.
Step 2: Calculate reliability index: Ī² = Ī¦ā»Ā¹(1 ā 0.0217) ā Ī¦ā»Ā¹(0.9783) ā 2.02 (using standard normal table or inverse CDF)
3.
Step 3: Compare to industry benchmarks: Ī² = 2.02 corresponds to Pf ā 2.2%, typical for operational open-pit highwalls with active monitoring per SME Guidelines.
Answer:
The result is Pf = 2.17% (Ī² ā 2.02), which falls within the acceptable range for monitored working slopes (Pf ā¤ 5% per SME 2021).
šļø Real-World Application
At Newmontās Boddington Mine (Western Australia), Monte Carlo simulation was used to assess the long-term stability of the East Wall highwall after rainfall infiltration modeling. Input parametersāincluding saturated unit weight (Ī³_sat ~ N(20.1, 0.4) kN/mĀ³), effective cohesion (c' ~ LogN(12, 0.35) kPa), and effective friction angle (Ļ' ~ N(32Ā°, 2.5Ā°))āwere assigned based on 217 laboratory tests and 42 piezometer time-series. Using 8,000 LHS samples in Slide3 with Spencerās method, the analysis revealed Pf increased from 0.4% (dry) to 6.8% during extreme 100-year rainfall. This quantified justification led to installation of targeted subsurface drains and real-time inclinometer arraysāreducing Pf to <1.5% at a cost 40% lower than over-designing the entire wall.