Professional Documents
Culture Documents
Phase 2
Phase 2
Curran, J.H.
University of Toronto & Lassonde Insitute, Toronto, ON, Canada
Purpose
To investigate the effect of a purely random Gaussian field on slope stability using Shear
Strength Reduction (SSR). Comparison of purely random fields to interpolation based on samples from a random field. To investigate the effect of different interpolation methods, in particular Kriging, on the factor of safety distribution.
Kriging
Kriging is a statistically rigorous approach to interpolating unknown values
Can be applied to sparsely sampled rock data to estimate a spatially variable property
Requires a basis, i.e. a set of spatially situated
known values
Kriging
Estimates a property at an unknown point as a weighted linear sum of all known points
x x i n Z y = wi Z yi i=1 z z i
Kriging
Ordinary kriging is a weighted average method where the weights are determined by the solution
cij is the covariance between two points. wi are kriging weights and 1 is a Lagrangian optimization parameter.
Kriging Advantages
Helps compensate for data clustering Works well with correlated samples
Kriged points revert to the mean when no data points are close
Cohesion (c) (MPa) 0.025 0.05 0.10 COV 4.3% 8.7% 17.3%
Friction angle contour map for a representative spatially variable trial slope
lower than its homogeneous equivalent, an effect that increases with increasing soil COV.
0.4 0.35 0 = 1.34 0.3 Relative Frequency 0.25 0.2 0.15 0.1 0.05 0 1.05 1.1 1.15 1.2 1.25 1.3 Factor of Safety 1.35 1.4 1.45 0.3 0.25 0.2 0.15 0.1 0.05 0 Relative Frequency 0.4 0.35 0.4 0.4 0 = 1.34
0.35
0.3 0.25 0.2 0.15 0.1 0.05 0 1.05 1.1 1.15
0.35
0.3 0.25 0.2 0.15 0.1 0.05 0
1.35
1.4
1.45
Interpolation Methods
Ordinary kriging was tested and compared to other
interpolation methods for soil properties, namely inverse distance weighting (IDW) and the geometric Chughs Method (MCM).
kriging
IDW
MCM
Interpolation Methods
Sets of 10, 50, 100 and 200 input points were used as the
basis for interpolation. The smallest sets of input points create the most variable FS distributions.
3.5 3 Probability Density 2.5 2 200 Points 1.5 1 0.5 0 0 1 2 Factor of Safety 3 10 Points Probability Density 50 Points 100 Points 3.5 3 10 Points 50 Points 100 Points 200 Points 1.5 1 0.5 0
2.5
2
kriging
MCM
0 1 2 Factor of Safety 3
Factor of safety distributions for interpolated slopes for Type III rock
Interpolation Methods
Increasing the number of input points lowers the mean
factor of safety. This is due to the smoothing effect of interpolation, especially kriging. Larger sets of (randomized) input points produce a rougher field and greater possibility of localized weakness contributing to overall failure.
Interpolation Methods
1.10 MCM IDW Kriged FS Coefficient of Variation 1.05 Relative FS (*) 1.00 0.95 0.90 0.85 0.80 0 50 100 150 No. of Basis Points 200 250 40% 35% MCM IDW Kriged
30%
25% 20% 15% 10% 5% 0% 0 50 100 150 No. of Basis Points
200
250
Normalized mean FS and COV for each interpolation method and size of basis
of influence a value has on its neighbours, or the roughness of the field. The correlation length was kept constant. The effect of varying CL was left to future work.
Z
Z Y
Z Y Y X
0.58 0.56
0.58
0.54 0.52
0.58 0.56
0.56 0.54
0.5
0.52
0.52
0.5
0.5
0.54
Interpolation Methods
The FS predicted by inverse distance weighting and kriging
were found to be highly correlated, while MCM was poorly correlated with both.
2.5 2.3 FS - Inverse Distance Weighting 2.1 FS - MCM 1.9 1.7 1.5 1.3 1.1 0.9 0.9 1.1 1.3 1.5 1.7 1.9 FS - Kriging 2.1 2.3 2.5 2.5 2.3 2.1 1.9 1.7 1.5 1.3 1.1 0.9 0.9 1.1 1.3 1.5 1.7 1.9 FS - Kriging 2.1 2.3 2.5
Conclusions
This research supports the use of kriging as a valid tool for
interpolating rock properties when data are scarce. Because of the smoothing effect of kriging, it does not fully preserve the spatial variability of the original material.
It is thus a partial measure, a way to account for measured
deviations from mean rock properties without fully anticipating probabilistic weakness, or a way to test the resilience of a designed slope to material property variation.
Conclusions
Because SSR does not assume a mode of failure, it is a
good method for accommodating spatial variability Further research will study the effect of kriging parameters (e.g. correlation length, anisotropy, moving window kriging) and test sequential Gaussian simulation (SGS), a more computationally intensive method based on kriging.