Probabilistic Slope Stability Analysis with the “Response Surface Methodology” (Henry T. Chiwaye)
Scope of Presentation Overview of Response Surface Methodology (RSM) Implementation of RSM in probabilistic slope stability analysis Verification Examples General guidelines for use of RSM
Slope Design Approaches Deterministic Factor of Safety (FOS) Probabilistic Probability of Failure(POF) Risk Analysis Economic / Safety impact Uncertainty: Geology Strength Water
Probabilistic Analysis Monte Carlo Simulation Point Estimate Method (PEM)
Monte Carlo Simulation CC Friction angle Cohesion Frequency INPUT DATA MONTE CARLO ANALYSIS (SLIDE, Phase2, FLAC, UDEC ) Model OUTPUT RESULTS 1.0 Frequency FOS POF model = P ( FOS < 1.00 ) POF Highlights Large no. of runs (103). Reveals Sensitivities Very Flexible
Point Estimate Method (PEM) CC Friction angle Cohesion Frequency INPUT DATA 2n MODEL RUNS (SLIDE, Phase2, FLAC, UDEC ) Model OUTPUT RESULTS FOS Statistics Mean Variance POF Highlights Evaluate model at 2n points. Assume a form for the FOS probability distribution No sensitivity information
Response Surface Methodology Response Surface Techniques Monte Carlo Simulation Probability Of Failure (%)
Response Surface Techniques
Response Surface Techniques Concept Var 1 Var 2 FOS Evaluate model at selected points Use interpolation scheme to generate response surface
Response Surface Generation RSM Overview Response Surface Generation Model (SLIDE, UDEC etc) MONTE CARLO ANALYSIS EXCEL 1.0 POF OUTPUT RESULTS FOS Frequency
RSM Verification Approach RSM vs. Model (SLIDE) Cohesion & friction angle uncertain variables RSM using linear interpolation Models 90m
Model vs. RSM (POF %) Homogeneous Slope: (Normal)
Model vs. RSM (POF %) Homogeneous Slope : (Lognormal)
Model vs. RSM (POF %) 3 Material Slope : (Normal)
RSM vs. PEM (POF %) Requires 2n + 1 points vs. 2n for PEM
RSM Guidelines Piecewise Linear / Quadratic interpolation can be used. Grouping Variables Strength Cohesion Friction Angle Evaluation points must be in region of interest (+ / - 1 std dev).
Remarks Use of RSM with strongly correlated variables
Conclusions Good agreement between RSM and Monte Carlo Simulation Low computational times Practical way to incorporate numerical analysis in probabilistic slope design Reveals Sensitivities Very flexible
Questions?