# Comparison of Point Estimate and Monte Carlo probabilistic methods in stability analysis of a deep excavation

- 78 Downloads

## Abstract

Geotechnical engineers have long used the lumped factor of safety approach in design of retaining walls. In recent years, uncertainty inherent in soil properties has caught more attentions from researchers, thus reliability analyses are used widely. In short, combining the probabilistic and deterministic analysis is common practice nowadays and failure probability calculated by risk analysis is a kind of complement of safety factor. In this paper, reliability analysis of a deep excavation is presented using two probabilistic methods called Point Estimate Method (PEM) and Monte Carlo Simulation (MCS) method. Horizontal displacement of the excavated wall, as well as safety of factor, are adopted as a basis for assessing the performance stability of the system. The paper explains why probabilistic analysis with numerical methods is challenging, and how PEM and MCS can be used to calculate the statistical moments of output variables and to estimate probability of failure. The results are presented and compared in terms of statistical moments, probability of occurrence and most likely values at every stage of construction.

## Keywords

Excavation Reliability analysis Probabilistic methods Point Estimate Method Monte Carlo Simulation## Introduction

Due to the deposition nature of soil masses, their properties can vary significantly from place to place, even over short distances. As well, the measurement of geotechnical properties, especially in situ, is a very challenging undertaking. Even when soil properties can be readily determined, inaccuracies in measurement and differences between laboratory and field-scale behavior introduce significant error. As a result, the engineering of excavations in soil involves large uncertainties and predictions based on single evaluations have practically zero probability of ever being realized, and design decisions based on them are therefore open to question.

Various researchers have studied different aspects of deep excavations considering it as a deterministic problem in which the input parameters are assumed to be single valued (see, e.g. [6, 25]). They solved these problems by deterministic calculation using safety factors and adopting conservative assumptions in the process of engineering planning and design. In this way, an engineering system can be over-designed and extremely expensive. Furthermore, even advanced deterministic methods become useless when the uncertainty is very high. As a result, the reliability level of a geotechnical structure cannot be estimated quantitatively by deterministic analysis of problem.

For a long time, reliability based design approach has been used for slope stability and excavation problems (e.g., [2, 5, 7, 10, 17, 34]) employing various types of probabilistic methods such as First Order Reliability Method (FORM), Second Order Reliability Method (SORM) and Monte Carlo Simulation (MCS) method. The drawback of FORM and SORM methods is that the results depend on the mean value of input variables at which the partial derivatives of the safety margin are evaluated (invariance problem). Therefore, these methods are accurate only for linear functions, and the error will be quite large when the degree of non-linearity of performance function is higher. Thus, it is necessary to look for a more effective and accurate reliability calculation method. As a result, researchers appreciate Monte Carlo Simulation (MCS). MCS is usually employed as a reference to check the accuracy of other methods. The disadvantage of MCS is that the number of simulations increases substantially with the reduction of target failure probability (i.e., reduction of accepted level of risk), and the calculation time will be prolonged immensely.

Another alternative method to evaluate statistical moments of a performance function is the Point Estimate Method, or shortly PEM. The basic idea of this method is to replace the probability distributions of continuous random variables by discrete equivalent distributions having the same first three central moments, to calculate the mean value, standard deviation, and skewness of a performance function, which depends on the input variables. While the PEM does not provide a full distribution of the output variable, as Monte Carlo does, it requires little knowledge of probability concepts and could be applied for any probability distribution. In future, it might be widely used for reliability analysis and for the evaluation of failure probability of engineering systems. The common ground of probabilistic studies regarding slope and excavation engineering is that, the reliability index or failure probability of slopes is solved by the combination of the limit equilibrium slice method and some probabilistic methods (mainly FORM and MCS) (e.g., [22, 24, 30]) and there exists a wide gap in implementing these non-deterministic methods in complex analysis, such as finite different difference method of analysis.

In this paper, while trying to cover the existing gap in literature concerning probabilistic analysis of deep excavation problems, a comparison is made between Point Estimate Method (PEM) and Monte Carlo Simulation method (MCS) in stability analysis of a real deep excavation. A technique is employed for constructing Probability Distribution Functions (PDF) of basic uncertain variables, without any need to collect numerous data, which is a great concern in probabilistic analysis of geotechnical problems. Statistical moments of system response, the probability of failure and most likely values obtained in every construction stages and the sampling procedure in each of methods will be compared and discussed. Finally, a proposal is suggested for reliability analysis of deep excavation using available data in common geotechnical investigation schemes. The process of numerical modeling and generation of trial numbers in MCS is conducted in a finite difference code FLAC, which will be mentioned in the following sections.

## Soheil commercial complex case study

### Subsurface condition and construction details

Geometric configuration of Soheil commercial complex excavated wall

No. | Elevation | Nail/strand | Inclination wrt. horizontal (°) | Bonded length (m) | Unbonded length (m) | No. of strands | Prestressing force (kN) |
---|---|---|---|---|---|---|---|

1 | − 2.00 | Nail | 15 | 12 | – | – | – |

2 | − 5.00 | Strands | 15 | 8 | 16 | 4 | 600 |

3 | − 8.00 | Strands | 15 | 8 | 14 | 4 | 600 |

4 | − 11.0 | Strands | 15 | 8 | 12 | 4 | 600 |

5 | − 14.00 | Strands | 15 | 8 | 10 | 4 | 600 |

6 | − 17.00 | Strands | 15 | 8 | 9 | 4 | 600 |

7 | − 20.00 | Strands | 15 | 8 | 7 | 4 | 600 |

Some detailed information on Soheil complex excavated wall

Parameter | Value |
---|---|

Vertical height of the wall (m) | 21.0 |

Face batter (°) | 0.0 |

Backslope angle (°) | 0.0 |

Yield strength of reinforcement f | 400 |

Elasticity modulus of reinforcement E | 200 |

Elasticity modulus of grout (concrete) E | 21 |

Diameter of steel bar d (mm) | 28 |

Diameter of strands (mm) | 15 |

Drill hole diameter D | 110 |

Spacing S | 1.5 × 3.0 |

Spacing S | 3.0 × 3.0 |

Facing thickness (cm) | 15 |

### Numerical modeling

_{e}, B

_{e}, and D

_{b}as defined in Fig. 2, it was found in separate studies [3, 12, 20] that W

_{e}= 3D

_{b}and B

_{e}= 3(H + D

_{b}) were appropriate values for W

_{e}and B

_{e}; indeed, beyond these values, W

_{e}and B

_{e}have little influence on the horizontal deflection of the wall due to the excavation of the soil. For the instrumented wall to be simulated, H was 21 m, D

_{b}was 63 m, B

_{e}was 240 m, and W

_{e}was 30 m. FLAC2D V7.0 was used to carry out the finite difference based simulations of the excavated wall considering it as a plane strain problem and accounting for the long term behavior using drained conditions. For the studied wall section in current project located about 80 m from the excavation corner with the ratio of L/H equal to 8.0 (greater than 6 according to Finno et al. [14] and Ou et al. [26]) supported on a very dense cemented layer, the plane strain condition can be reasonably assumed. Thus, in order to compare the results of non-deterministic methods with the field measurements, the horizontal displacement measured at the top and center of the studied wall was selected in current study.

## Uncertainty model

In order to perform a probabilistic analysis, it is necessary to build the probabilistic distribution function (PDF) of input parameters, which is regarded as an uncertainty model. To do this, there is some methods including: (1) fitting distribution function to the collected data, (2) assuming a PDF for input parameters based on previous studies available in literature and (3) providing the p-box model of input parameters on the basis of Random Set (RS) theory. In short, in random set theory, set valued information with given probability measures are combined together leading to probability bounds in terms of discrete cumulative distribution functions (CDF). The bounds comprise any distribution compatible with the existing data including the actual distribution, if numerous data existed. Since in the common geotechnical surveys there is often limited data available, which is insufficient for fitting a suitable PDF to the samples, the random set procedure seems to be a suitable alternative. Moreover, the results of numerical analysis is very sensitive to variation of input distribution parameters; thus considering a PDF (i.e., normal, lognormal etc.) with assumptive statistical moments for the variables seems misleading. In the current research, in order to solve the mentioned obstacles, a pertinent technique is employed.

**x**= (x

_{1},…,x

_{N}) can be expressed as a random relation R, which is a random set \(\left( {{\Im },m} \right)\) on the Cartesian product \(X_{1} \times \cdots \times X_{N}\); where \({\Im } = \left\{ {A_{i} :i = 1, \ldots ,n} \right\}\) and m is a mapping, \({\Im } \to \left[ {0.1} \right]\), so that \(m\left( \emptyset \right) = 0\) and

_{i}are the focal elements (A

_{i}\(\subseteq\) X) and m is called the basic probability assignment. Each set, \(A \in {\Im }\), contains some possible values of the variable, x, and m(A) can be viewed as the probability that A is in the range of x. In current research, m is assumed equal to 0.5, which indicates that both sources of information provided for input parameters are equally reliable. Alternatively, the sets A

_{i}(i.e., each sets of spatially correlated data presented in Fig. 3) could be ranges of a variable obtained from source number i with relative credibility m

_{i}. Because of the imprecise nature of this formulation it is not possible to calculate the ‘precise’ probability, Pro, of a generic x ∈ X or of a generic subset E ⊂ X, but only lower and upper bounds on this probability (Fig. 3): Bel(E) ≤ Pro(E) ≤ Pl(E). In the limiting case, when i is composed of single values only (singletons), then Bel(E) = Pro(E) = Pl(E) and m is a probability distribution function.

^{2}, as discussed by Vanmarcke [39] through σ

_{Γ}= Γ·σ where σ is the standard deviation of field data (point statistics) and σ

_{Γ}is the standard deviation of the spatial average of the data over volume V. The variance reduction factor depends on the averaging volume and the type of correlation structure in the form:

_{i,Γ}can be calculated from the discrete cumulative probability distribution of the field data x

_{i}

_{ur}was select as K

_{ur}= 3 K, based on in situ Plate Load Tests, n constant was assumed to be equal to 0.5 and R

_{f}was 0.93 for all three layers.

Basic variables for material parameters (input values)

Soil, information source | c, kN/m | Φ | K |
---|---|---|---|

Fill materials | |||

Geotechnical report | 0–4 | 24–27 | 10–70 |

Expert knowledge | 2–5 | 26–29 | 30–100 |

Clayey sand with gravel (SC) | |||

Geotechnical report | 30–45 | 32–36 | 50–150 |

Expert knowledge | 40–55 | 34–37 | 100–250 |

Clayey gravel with sand (GC) | |||

Geotechnical report | 35–50 | 36–40 | 150–250 |

Expert knowledge | 40–60 | 37–42 | 200–300 |

Typical values of spatial correlation lengths, Θ, for soils as given e.g. by Li et al. [19] are in the range 0.1 to 5 m for Θ_{y} and from 2 to 30 m for Θ_{x}. Based on the subsurface condition in the project and some specific regulations in the region of interest (e.g., the minimum distance between the boreholes must be at least 15 m based on the uniformity level of subsurface soils), the average spatial correlation length for this study is assumed to be 10 m. The characteristic length, L, was taken as 30 m, which is based on analyses investigating potential failure mechanisms for this problem. Because of considering spatial correlation (Fig. 3), the discrete cumulative probability function becomes steeper.

## Sensitivity analysis

_{x−A}, and safety factor of stability at the final construction step are evaluated. Figure 4 shows the relative sensitivity obtained for each of variables for both criteria and indicates that C2-Φ2-C3-Φ3 and C2-K2-K3 are determined as uncertain basic variables in stability and deformation analysis, respectively; considering the threshold value equal to 5%.

## Probabilistic methods

### Point Estimate Method (PEM)

There are two main distinguishable approaches for the Point Estimate Method: first, the original approach proposed by Rosenblueth [31] and its modified method given by his followers (e.g. [8, 16, 18, 21]) that attempted to save computational cost by reducing the number of prefixed sampling points. Second, the approach proposed by Zhou and Nowak [40] including three different integration rules, namely n + 1, 2n, 2n^{2} + 1 (n is the number of basic variables involved in the analysis). Thurner [36] concluded that the 2n^{2} + 1 integration rule results in an optimum compromise between accuracy and computational effort. The reason for difference between these methods is the sampling and weighting technique employed in each of them. The weights assigned to all permutations in n + 1 and 2n are equal, which is considered clearly as a shortcoming, while in 2n^{2} + 1 different weights are assigned to provided permutations (i.e., greater weights to more probable permutations). Additionally, the number of sampling points in tales of PDFs in 2n^{2} + 1 is much greater than the other two, thus more precise values of probability of failure can be derived in this method. For these reasons, in this research the 2n^{2} + 1 rule is used for generating sampling points.

_{j}are typical predetermined points and w

_{j}are the respective weights given in Table 4. The sum of the weights is always 1.0 and the number of required calculations thus depends on the integration rule. Table 5 presents sampling points of basic variables in deformation analysis (i.e., C2, K2 and K3) along with assigned weights to each combination.

Evaluation points and respective weights in PEM [40]

Formulas m | Points (z | Weights factors w |
---|---|---|

2n | \(Z = \left( {0.0, \ldots ,0} \right)\) | \(w_{j} = \frac{2}{n + 2}\) |

\(Z = \left( { \pm \sqrt {n + 2} .0, \ldots ,0} \right)\) | \(w_{j} = \frac{4 - n }{{2\left( {n + 2} \right)^{2} }}\) | |

\(Z = \left( { \pm \sqrt {\frac{n + 2}{2}} \pm \sqrt {\frac{n + 2}{2}} , \ldots ,0} \right)\) | \(w_{j} = \frac{1}{{\left( {n + 2} \right)^{2} }}\) |

Sampling points and assigned weights to basic variables of deformation analysis (i.e., C2, K2 and K3) in PEM

Analysis no. | Input variables | Weights | ||
---|---|---|---|---|

C2 | K2 | K3 | ||

1 | 42.50 | 133.02 | 223.18 | 0.4 |

2 | 52.12 | 133.02 | 223.18 | 0.02 |

3 | 32.88 | 133.02 | 223.18 | 0.02 |

4 | 42.50 | 236.57 | 223.18 | 0.02 |

5 | 42.50 | 74.79 | 223.18 | 0.02 |

6 | 42.50 | 133.02 | 296.79 | 0.02 |

7 | 42.50 | 133.02 | 167.82 | 0.02 |

8 | 49.30 | 199.86 | 223.18 | 0.04 |

9 | 49.30 | 88.53 | 223.18 | 0.04 |

10 | 35.70 | 199.86 | 223.18 | 0.04 |

11 | 35.70 | 88.53 | 223.18 | 0.04 |

12 | 49.30 | 133.02 | 273.02 | 0.04 |

13 | 49.30 | 133.02 | 182.44 | 0.04 |

14 | 35.70 | 133.02 | 273.02 | 0.04 |

15 | 35.70 | 133.02 | 182.44 | 0.04 |

16 | 42.50 | 199.86 | 273.02 | 0.04 |

17 | 42.50 | 199.86 | 182.44 | 0.04 |

18 | 42.50 | 88.53 | 273.02 | 0.04 |

19 | 42.50 | 88.53 | 182.44 | 0.04 |

Sum | 1 |

### Monte Carlo Simulation (MCS)

The Monte Carlo Simulation (MCS) technique is considered as a very powerful tool for engineers and can be quite accurate if enough simulations are performed. The MCS consists of sampling a set of properties for the materials from their joint PDF and introducing them in the model; thus, a set of results (e.g., displacements) can then be obtained. This operation is repeated a large number of times and an empirical frequency-based probability distribution can be defined for each result. Information on the distribution and statistical moments of the response variable is then obtained from the resulting simulations. The accuracy of the MCS technique increases with the increase in the number of simulations. However, this can be disadvantageous as it becomes computationally expensive, and as such, the simulator’s task is to increase the efficiency of the simulation by expediting the execution and minimizing the computer storage requirements. In current study, the PDFs of basic input variables determined in sensitivity analysis are built and subsequently the random variables will be generated based on the aforementioned PDFs. Then, using the obtained results, the PDF of system response (i.e., horizontal deformation of point A in Fig. 2) for 10, 100, 200, 300 and 400 simulations will be sketched and compared in deformation analysis of the problem, while much greater trials will be performed in stability analysis section of the research.

### Constructing PDF of input variables

In order to apply PEM and MCS in Soheil project, it is necessary to have a probability distribution function for each random variable. To determine the distribution function for a particular soil property, more than 40 data points are needed [4]. In fact, due to the lack of test results and limited sampling in usual geotechnical projects, it is not usually feasible to build an exact PDF of the considered random variables. Thus, the following two alternatives are discussed to properly match the random set input variable given in Fig. 3 with a corresponding PDF used in the prospective probabilistic calculations. At the end, one alternative is selected for this purpose.

#### Alternative I, the “three-sigma” rule

_{l}is the estimated minimum value, x

_{r}the estimated maximum value, and n is the most likely value or mode. As an example, Fig. 5a shows an application of 3-σ rule to the random set of the modulus factor of layer two using the relationship given by Schneider. The most probable value is obtained by averaging all random set extreme values. The estimated extreme values (x

_{l}, x

_{r}) would be the most outer values of the random set bounds. This method can be used, but it could pose a drawback that the probability measure of the values around the random set bounds is very low. Therefore, this approach may underestimate some values around the edge of the random set. It might be more reasonable to use an approach that can allow for a more appropriate probability share for extreme values in the random set to provide a better match between the reference data and PEM/MCS input variable.

#### Alternative II, the selection of a best fit to a uniform distribution

Basic random variables and respective PDF details

Basic variable | Layer | Unit | Distribution type | Mean | Standard dev. | COV % |
---|---|---|---|---|---|---|

Cohesion, C | 2 | KPa | Normal | 42.5 | 4.3 | 10 |

Friction, Φ | 2 | ° | Normal | 34.7 | 1.0 | 3 |

Modulus factor, K | 2 | – | Lognormal | 137.5 | 36 | 26 |

Cohesion, C | 3 | KPa | Normal | 47.5 | 4.3 | 10 |

Friction, Φ | 3 | ° | Normal | 38.7 | 1.3 | 3 |

Modulus factor, K | 3 | – | Lognormal | 225 | 28.8 | 13 |

## Results and discussion

### Generating random numbers

^{2}+ 1) are evaluated (see Table 5). Figure 6 shows the sampling points of basic variables on their PDF graphs. It is clear that in this method, some points replace the PDF and the combination of these points determines the number of permutations. Moreover, the samples are distributed properly along the tales of PDF, which induces reasonable estimation of statistical moments of system response.

In order to generate the random numbers in MCS method, it was required to write a pertinent code in FISH language of FLAC2D software. The programming environment of FLAC2D has only the ability to generate uniform random numbers by its default commands. Thus, it seems necessary to write a code for generating normal and lognormal random numbers. For this purpose, the normal random numbers were generated using the existing commands and then by using existing formulas for transformation of normal to lognormal variables; the lognormal random numbers were created.

As it was mentioned previously, the MCS is the most precise probabilistic method, which due to building the correct tales of response PDF, there is a need to simulate a great number of realizations. Although this procedure is possible in simple geotechnical solutions, such as limit equilibrium analysis, but in complex problems such as Finite Difference (FD) analysis, this method needs great computational effort, which is practically impossible. Because of these limitations, the study of MCS results regarding deformation FD analysis, were concentrated on 10, 100, 200, 300 and 400 simulations and an evaluation is carried out between the results.

### Comparison of results

_{x,A}according to sensitivity analysis performed earlier. Accordingly, the discrepancy observed in estimation of PEM and MCS method can be attributed to different sample range derived for K2. Table 7 presents the sampling range provided for PEM and MCS. From this table, it can be deduced that this mismatch decreases with changing PDF of input variables from non-normal to normal (i.e., from asymmetric to symmetric or from highly skewed to low skewed distributions). The mean value obtained for U

_{x,A}in all numbers of simulations in MCS are presented against the mean value of system response in PEM in Fig. 8 for all construction stages. It is obvious that there is a significant difference between these two probabilistic methods in predicting the mean value. According to previous explanations, it can be concluded that if the probability distribution function of all variables were normal, then the difference between results would be minimized.

Sampling range of uncertain variables in PEM and MCS

Basic variable | Unit | Skewness | Sampling range | ||
---|---|---|---|---|---|

PEM | MCS (n = 400) | MCS (n = 10,000) | |||

C2 | KPa | 0.00 | 33–52 | 31–55 | 26–59 |

K2 | – | 0.81 | 74–236 | 50–272 | 47–355 |

K3 | – | 0.41 | 167–296 | 157–333 | 130–365 |

_{x,A}in both PEM and MCS methods are presented in Fig. 9 for all construction stages. It can be observed that the fluctuation trend of σ

_{Ux,A}for both MCS and PEM are in good agreement and the difference between obtained values is much lower than the mean values of U

_{x,A}. This indicates that despite the limited number of simulations employed in PEM; this method presents reasonable estimation for predicting the dispersion of results by employing much less number of simulations from MCS.

_{f}, in terms of serviceability, where H is the wall height. It can be seen from Table 8 that the probability of failure in all construction stages are very different in PEM and MCS methods, which can be referred to narrower range of sampling in PEM leading to larger values of μ in PEM than MCS, consequently shifting the PDF of system response (i.e., U

_{x,A}) to the right direction (i.e., increasing the p

_{f}). It is worth mentioning that the unstable trend in p

_{f}, both in progressing of construction stages and increasing the number of trials could be a clear indication that much more simulations are required in MCS analysis in order to achieve precise values of probability of failure in a deep excavation problem.

Probability of failure at different stages of construction

Stage | MCS (number of simulations) | PEM | ||||
---|---|---|---|---|---|---|

10 | 100 | 200 | 300 | 400 | ||

III | ~ 0 | ~ 0 | ~ 0 | ~ 0 | ~ 0 | ~ 0 |

IV | 3e | 5e | 1e | 3e | 1e | 0.003 |

V | 8e | 1e | 3e | 1e | 3e | 0.004 |

VI | 2e | 4e | 1e | 4e | 4e | 0.033 |

VII | 2e | 1e | 2e | 6e | 2e | 0.202 |

_{x,A}in final excavation stage for n = 10, 100 and 1000. It is obvious that a significant difference exists in scattering and convergence of results in these simulations, which clearly is the key factor in discrepancy between results presented in Table 9.

Statistical moments of MCS results in LEA analysis

Trials | Mean of FS | Reliability index, β | Probability of failure, p | Standard deviation of FS | Range of FS |
---|---|---|---|---|---|

10 | 1.524 | 11.66 | ~ 0 | 0.045 | 1.45–1.59 |

100 | 1.530 | 12.46 | ~ 0 | 0.043 | 1.44–1.62 |

1000 | 1.525 | 10.30 | ~ 0 | 0.051 | 1.30–1.68 |

10,000 | 1.523 | 10.70 | ~ 0 | 0.049 | 1.30–1.73 |

100,000 | 1.522 | 10.80 | ~ 0 | 0.048 | 1.25–1.74 |

1,000,000 | 1.522 | 10.80 | ~ 0 | 0.048 | 1.23–1.79 |

According to Table 9, it is obvious that standard deviation decreases with increasing the number of simulation, which is rational due to the growth of statistical population. In the other hand, this decrease of σ with increasing trial numbers is negligible which shows insignificant effect of number of simulations on FS evaluation from n = 1000. Comparing p_{f} obtained by MCS indicates that in every stage of construction the p_{f} (p_{FS < 1}) is equal to zero, which means that the original design was too conservative and significant savings could be made in the case of performing probabilistic analysis before and during the project.

## Conclusion

In this paper, two probabilistic methods of analysis (i.e., Monte Carlo Simulation and Point Estimate Methods) was employed in a real deep excavation problem. Comparing sampling range, statistical moments and most likely values of system response and probability of failure obtained in each method, it was deduced that the major factor in difference between these methods is the mean value determined, which induces significant difference in most likely values and p_{f}. This can be attributed to narrower range of samples obtained in PEM than MCS in lognormal distribution functions of input parameters and can be minimized, if normal (or nearly normal) PDFs were assigned to all variables. Although, the sampling procedure in PEM and MCS were very different, but both of these methods presented relatively similar values of standard deviation.

The MCS results in FD analysis presented an unstable condition in trials up to n = 400, which indicates that much more simulations are required in complex types of analysis; thus, PEM, despite differences mentioned earlier, can be preferable to MCS in predicting p_{f} values in serviceability reliability analysis of deep excavations with much less computational effort needed. The sampling procedure in PEM and MCS are quiet different, but it was observed that the statistical moments of input variables in both of these methods was in good agreement for normal distribution functions. From the parametric study performed in limit equilibrium analysis, it was found that the effect of varying the number of random trials was not significant beyond 1000 random trials in MCS and reasonable results regarding stability analysis can be achieved by this method in stability analysis of deep excavations. The paper showed that combining the probabilistic deformation analysis using PEM (under specific conditions) and MCS using finite difference methods with probabilistic stability analysis by MCS using limit equilibrium analysis could be a desirable procedure which is recommended in reliability analysis of deep excavations problems.

## Notes

### Authors’ contributions

AS collected the data, carried out numerical modeling, interpreted the results and wrote the manuscript in consultant with AJC. AJC supervised the project. Both authors read and approved the final manuscript.

### Competing interests

The authors declare that they have no competing interests.

### Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## References

- 1.Alonso EE, Krizek RJ (1975) Stochastic formulation of soil profiles. In: Proc Int Conf Appl Stat Probab (ICASP2), Aachen, Germany, pp 9–32Google Scholar
- 2.Babu GL, Singh VP (2009) Reliability analysis of soil nail walls. J Georisk 3(1):44–54Google Scholar
- 3.Briaud JL, Lim Y (1999) Tieback walls in sand: numerical simulations and design implications. J Geotech Geoenviron Eng 125:101–110CrossRefGoogle Scholar
- 4.Carter JP, Desai CS, Potts DM, Schweiger HF, Sloan SW (2000) Computing and computer modeling in geotechnical engineering. In: Proc Geoeng 2000, Melbourne, Australia, vol 1, pp 1157–1252Google Scholar
- 5.Chowdhury SS (2017) Reliability analysis of excavation induced basal heave. J Geotech Geol Eng 35(6):2705–2714CrossRefGoogle Scholar
- 6.Chowdhury SS, Deb K, Sengupta A (2016) Effect of fines on behavior of braced excavation in sand: experimental and numerical study. Int J Geomech ASCE. https://doi.org/10.1061/(ASCE)GM.1943-5622.0000487 CrossRefGoogle Scholar
- 7.Christian JT, Baecher GB (1999) Point estimate method as numerical quadrature. J Geotech Geoenviron Eng (ASCE) 125:779–786CrossRefGoogle Scholar
- 8.Christian JT, Baecher GB (2002) The point-estimate method with large numbers of variables. Int J Numer Anal Meth Geomech 26:1515–1529CrossRefGoogle Scholar
- 9.Cornell CA (1971) First-order uncertainty analysis of soils deformation and stability. In: Proc Int Conf Appl Stat Probab (ICASP1), pp 129–144Google Scholar
- 10.Duncan JM (2000) Factors of safety and reliability in geotechnical engineering. J Geotech Geoenviron Eng (ASCE) 126:307–316CrossRefGoogle Scholar
- 11.Duncan JM, Byrne PM, Wong KS, Marby P (1980) Strength, stress–strain and bulk modulus parameters for finite element analysis of stresses and movements in soil mass. Rep. No. UCB/GT/80-01, University of California, Berkeley, CaliforniaGoogle Scholar
- 12.Dunlop P, Duncan JM (1970) Development of failure around excavated slopes. J Soil Mech Found Div (ASCE) 96:471–493Google Scholar
- 13.FHWA (2015) Soil nail walls reference manual. U.S. Department of Transportation Federal Highway Administration, WashingtonGoogle Scholar
- 14.Finno RJ, Blackburn JT, Roboski JF (2007) Three dimensional effects for supported excavations in clay. J Geotech Geoenviron Eng (ASCE) 133(1):30. https://doi.org/10.1061/(asce)1090-0241(2007)133:1(30) CrossRefGoogle Scholar
- 15.Griffiths DV, Fenton GA (2000) Influence of soil strength spatial variability on the stability of an undrained clay slope by finite elements. In: Slope Stab 2000 (ASCE), pp 184–193Google Scholar
- 16.Harr ME (1989) Probabilistic estimates for multivariate analyses. Appl Math Model 13(5):313–318CrossRefGoogle Scholar
- 17.Hoeg K, Muruka RP (1974) Probabilistic analysis and design of a retaining wall. J Geotech Eng Div (ASCE) 100(3):349–365Google Scholar
- 18.Hong HP (1998) An efficient point estimate method for probabilistic analysis. Reliab Eng Syst Saf 59(3):261–267CrossRefGoogle Scholar
- 19.Li IK, White W, Ingles OG (1983) Geotechnical engineering. Pitman, BostonGoogle Scholar
- 20.Lim Y, Briaud JL (1996) Three dimensional nonlinear finite element analysis of tieback walls and of soil nailed walls under piled bridge abutment. Report to the Federal Highway Administration and the Texas Department of Transportation, Department of Civil Engineering, Texas A&M University, College Station, TexasGoogle Scholar
- 21.Lind NC (1983) Modelling uncertainty in discrete dynamical systems. Appl Math Model 7(3):146–152CrossRefGoogle Scholar
- 22.Low BK, Gilbert RB, Wright SG (1998) Slope reliability analysis using generalized method of slices. J Geotech Geoenviron Eng 124(4):350–362CrossRefGoogle Scholar
- 23.Lumb P (1975) Spatial variability of soil properties. In: Proc 2nd Int Conf Appl Stat Probab Soil Struct Eng, pp 397–421Google Scholar
- 24.Malkawi AIH, Hassan WF, Abdulla FA (2000) Uncertainty and reliability analysis applied to slope stability. J Struct Saf 22(2):161–187CrossRefGoogle Scholar
- 25.Nogueira CL, Azevedo RF, Zornberg JG (2009) Coupled analyses of excavations in saturated soil. Int J Geomech (ASCE). https://doi.org/10.1061/(ASCE)1532-3641(2009)9:2(73) CrossRefGoogle Scholar
- 26.Ou CY, Chiou DC, Wu TS (1996) Three-dimensional finite element analysis of deep excavations. J Geotech Eng (ASCE). https://doi.org/10.1061/(asce)0733-9410(1996)122:5(337) CrossRefGoogle Scholar
- 27.Peschl GM (2004) Reliability analyses in geotechnics with the random set finite element method. Graz University of Technology, GrazGoogle Scholar
- 28.Pukelsheim F (1994) The three sigma rule. Am Stat 48(2):88–91Google Scholar
- 29.Rackwitz R (2000) Reviewing probabilistic soils modeling. Comput Geotech 26:199–223CrossRefGoogle Scholar
- 30.Ramly HE, Morgenstern NR, Cruden DM (2002) Probabilistic slope stability analysis for practice. Can Geotech J 39:665–683CrossRefGoogle Scholar
- 31.Rosenblueth E (1975) Point estimates for probability moments. Proc Natl Acad Sci 72(10):3812–3814CrossRefGoogle Scholar
- 32.Seed RB, Duncan JM (1984) SSCOMP: a finite element analysis program for evaluation of soil structure interaction and compaction effects. Report No. UCB/GT/84-02, University of California, Berkeley, CaliforniaGoogle Scholar
- 33.Schneider HR (1999) Determination of characteristic soil properties. In: Barends et al., editors. Proc XII Int Conf Soil Mech Geotech Eng, vol 1, pp 273–281Google Scholar
- 34.Tobutt DC (1982) Monte Carlo simulation methods for slope stability. Comput Geosci 8(2):199–208CrossRefGoogle Scholar
- 35.Tonon F, Bernardini A, Mammino A (2000) Determination of parameters range in rock engineering by means of Random Ret Theory. Reliab Eng Syst Saf 70:241–261CrossRefGoogle Scholar
- 36.Thurner R (2000) Probabilistische Untersuchungen in der Geotechnik mittels Deterministischer Finite Elemente-Methode. Institut für Bodenmechanik und Grundbau, GrazGoogle Scholar
- 37.U.S. EPA: TRIM (1999) TRIM, Total Risk Integrated Methodology. TRIM FATE technical support document volume I: description of module. EPA/43/D-99/002A. Office of Air Quality Planning and StandardsGoogle Scholar
- 38.Vanmarcke EH (1977) Probabilistic modeling of soil profiles. J Geotech Eng Div (ASCE) 103:1227–1246Google Scholar
- 39.Vanmarcke EH (1983) Random fields—analysis and synthesis. MIT-Press, CambridgeGoogle Scholar
- 40.Zhou J, Nowak AS (1988) Integration formulas to evaluate functions of random variables. Struct Saf 5(4):267–284CrossRefGoogle Scholar

## Copyright information

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.