Next Article in Journal
Robust Adaptive Super Twisting Algorithm Sliding Mode Control of a Wind System Based on the PMSG Generator
Next Article in Special Issue
Impact of Rock Fragment Shapes and Soil Cohesion on Runoff Generation and Sediment Yield of Steep Cut Slopes under Heavy Rainfall Conditions
Previous Article in Journal
Enhancing Water Management in Shale Gas Extraction through Rectangular Pulse Hydraulic Fracturing
Previous Article in Special Issue
The Characteristics and Seepage Stability Analysis of Toppling-Sliding Failure under Rainfall
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Surrogate Model Development for Slope Stability Analysis Using Machine Learning

1
Institute of Engineering Innovation, The University of Tokyo, Tokyo 113-8656, Japan
2
Department of Civil Engineering, The University of Tokyo, Tokyo 113-8656, Japan
3
Department of Engineering Mechanics and Energy, Faculty of Engineering, Information and Systems, University of Tsukuba, 1-1-1 Tennodai, Tsukuba 305-8573, Japan
4
Geoscience Research Laboratory, Co., Ltd., 2-3-25 Koraku, Bunkyo City, Tokyo 112-0004, Japan
*
Author to whom correspondence should be addressed.
Sustainability 2023, 15(14), 10793; https://doi.org/10.3390/su151410793
Submission received: 18 April 2023 / Revised: 4 July 2023 / Accepted: 5 July 2023 / Published: 10 July 2023
(This article belongs to the Special Issue Slope Stability Analysis and Landslide Disaster Prevention)

Abstract

:
In many countries, slope failure is a complex natural issue that can result in serious natural hazards, such as landslide dams. It is associated with the challenge of slope stability evaluation, which involves the classification problem of slopes and the regression problem of predicting the factor of safety (FOS) value. This study explored the implementation of machine learning to analyze slope stability using a comprehensive database of 880 homogenous slopes (266 unstable and 614 stable) based on a simulation model developed as a surrogate model. A classification model was developed to categorize slopes into three classes, including S (stable, FOS > 1.2), M (marginally stable, 1.0 ≤ FOS ≤ 1.2), and U (unstable, FOS < 1.0), and a regression model was used to predict the target FOS value. The results confirmed the efficiency of the developed classification model via testing, achieving an accuracy of 0.9222, with 96.2% accuracy for the U class, 55% for the M class, and 95.2% for the S class. When U and M are in the same class (i.e., the U + M class), the test accuracy is 0.9315, with 93.3% accuracy for the S class and 92.9% accuracy for the U + M class. The low accuracy level for class M led to minor inaccuracies, which can be attributed to a data imbalance. Additionally, the regression model was found to have a high correlation coefficient R-square value of 0.9989 and a low test mean squared error value of 5.03 × 10−4, which indicates a strong relationship between the FOS values and the selected slope parameters. The significant difference in the elapsed time between the traditional method and the developed surrogate model for slope stability analysis highlights the potential benefits of machine learning.

1. Introduction

Landslides are common geological hazards that cause significant social and economic damage. These natural hazards are influenced by various factors, including external triggers such as rainfall and earthquakes, as well as internal factors such as slope configuration and soil characteristics [1,2]. It is urgent for engineers and researchers to analyze the stability of slopes to prevent or mitigate the potential risks posed by landslides [3].
Traditionally, slope stability is evaluated by calculating the factor of safety (FOS) and determining an appropriate treatment design. If the FOS is greater than 1.0, the slope is considered stable, whereas if it is less than 1.0, the slope is considered unstable [4]. Common methods for calculating the safety factor in slope stability analyses include the limit equilibrium method (LEM) and numerical calculation methods based on theories of elasticity and plasticity [5,6]. However, the accuracy of the LEM is limited owing to the assumptions about slip surfaces and interslice forces, whereas numerical calculation methods require a precisely fitting constitutive model, which is challenging to achieve [7,8,9]. The complexity of the interactions among the factors affecting slope stability also makes it challenging to accurately evaluate real slope stability with an FOS.
Machine learning algorithms, such as artificial neural networks (ANNs) [10], support vector machines (SVM) [11], and gradient boosting machines (GBM) [12], have increasingly been used to evaluate slope stability as a nonlinear problem because of their ability to extract valuable information from actual slope case records. These methods evaluate the slope stability based on geotechnical parameters (cohesion and internal friction angle), slope geometry (slope height and slope angle), water conditions (pore water pressure), and dynamic conditions (earthquake effect) and have been proven to be promising in slope stability evaluations. However, even with an FOS > 1.0, the slopes may still fail. According to Mahmoodzadeh et al. [13], slopes with an FOS > 1.2 are considered safe, and the dataset used in previous studies is summarized in their study, which ranged from 10 to 699 cases. Despite the efforts by researchers to collect real on-site data, the dataset is often too small to establish its applicability and repeatability. Therefore, there is a need for surrogate model development [14] in slope stability analysis. The use of machine learning models allows for instantaneous results without the need for numerical simulations, making it an ideal approach.
This study investigates the use of machine learning models to analyze the complex nonlinear relationships in slope stability evaluations, utilizing a comprehensive database of 880 homogenous slopes (266 unstable and 614 stable) generated by FLAC 3D Version 7.0 software (Fast Lagrangian Analysis of Continua in 3 Dimensions, Itasca Company, Minneapolis, MN, USA) instead of traditional real on-site data. A classification model was developed to categorize slopes into three classes: S (stable, FOS > 1.2), M (marginally stable, 1.0 ≤ FOS ≤ 1.2), and U (unstable, FOS < 1.0). Furthermore, a regression model was used to predict the target FOS. This study used virtual slope data to develop the models and evaluate their performance.

2. Dataset

2.1. Modeling: A Simple and Homogeneous Soil Slope

This section models a simple and homogeneous soil slope, with a height of 6 m and an angle of 45°, as shown in Figure 1. The basic slope model had a width of 20 m and a height of 10 m. To maintain the rigid behavior of the model based on the boundary conditions, the slope must be embedded in the bedrock. In the analysis, the unit weight, elastic modulus, and Poisson’s ratio of the soil were set at 20 kN/m3, 14 MPa, and 0.3, respectively. To ensure the validity of the analysis, a range of shear strength properties are utilized, and both numerical simulations using FLAC 3D and analyses using the limit equilibrium method (LEM) were conducted for the parametric study. The FOS results are listed in Table 1.
Therefore, in most cases, the FOS obtained using the numerical simulation is comparable to that obtained using the traditional LEM method under different combinations of soil parameters (i.e., cohesion and internal friction angle). However, there are a few cases in which there is a larger difference in the FOS values between the numerical simulation and the LEM, particularly when the cohesion is 20 kPa and the friction angle is 0°, at which point the error can reach up to 25%. In addition, if the slope is unstable (i.e., FOS is less than 1.0), the numerical simulation analysis does not provide any calculated results. Conversely, if there is no FOS output, the slope is considered unstable.

2.2. An Established Dataset

In this study, several numerical simulations based on FLAC 3D were conducted to model numerous homogenous slopes and obtain their factors of safety (FOS). As aforementioned, all models had a density of 20 kN/m3, a Young’s modulus of 14 MPa, a Poisson’s ratio of 0.3, and a tensile strength of 0. Four key parameters were considered to represent the slope characteristics: slope height (H), slope angle (α), cohesion (c), and internal friction angle (φ). Values of 3 m, 6 m, and 9 m were considered for the slope height, and values of 26.57°, 45°, and 63.43° were used for the slope angle. For the soil parameters (shear strength), a range of 2–50 kPa was considered for soil cohesion, and a range of 0–45° was considered for the internal friction angle. This resulted in a dataset consisting of 880 homogenous slopes with FOS values as the output. The generated dataset is shown in Appendix A Table A1. The dataset was then used for the classification model for slopes and the regression model for FOS prediction, which are introduced in Section 3, model development. Note that the dataset used in this study was selected solely by the simple full factorial experiment method, without considering data balance. In addition, the decision to use homogeneous slopes in this study was a deliberate choice made to establish a clear understanding of the neural network’s behavior and performance under simplified conditions.
The frequency distribution of the FOS values used in this study is shown in Figure 2, with 266 unstable slopes and 614 stable slopes. To understand the relationship between the FOS values and other parameters, scatterplots of all key parameters against the FOS are plotted in Figure 3. Additionally, a typical relationship between the FOS and the internal friction angle is shown in Figure 4, with a slope height of 6 m, a slope angle of 45°, and various cohesion values (2–50 kPa). The results indicate that the calculated FOS increases with an increase in both the cohesion and friction angle, which is consistent with previous findings [4]. However, an increase in slope height and slope angle resulted in an initial increase in the FOS, followed by a decrease, which differs from previous research. This discrepancy may be attributed to the range of slope heights and slope angles selected in this study.

3. Model Development

Artificial neural networks (ANNs) are machine learning models inspired by the structure and function of the human brain. ANNs are used to model complex relationships between inputs and outputs and can improve their predictions over time [15,16]. In contrast, deep neural networks (DNNs), also known as deep learning models [17], are modifications of ANNs that differ primarily in the number of layers they contain [18]. While ANNs have a single layer of input neurons connected to a single layer of output neurons, DNNs consist of multiple layers of neurons, with each layer feeding into the next. The increased complexity of DNNs enables them to learn and identify complex features and patterns in data that may not be apparent in simpler models such as ANNs.
DNNs have a wide range of applications in civil engineering, including crack detection in concrete [19] and asphalt pavement [20], bridge damage identification [21,22], and the automatic recognition of soil desiccation cracks [23]. Overall, the greater complexity and improved feature recognition capabilities of DNNs make them powerful tools for a variety of machine learning tasks (e.g., classification and regression), particularly those involving large and complex datasets. Additionally, the use of a surrogate model with machine learning can help reduce the computation time and cost.

3.1. A Deep Neural Network Model for Slope Classification

A deep neural network for classification is a machine learning tool used to categorize input data into different classes [24]. This model facilitates data collection, network creation and training, and performance evaluation using cross-entropy and confusion matrices. A trained, feedforward, and fully connected DNN model designed for slope classification is shown in Figure 5. This network consisted of nine fully connected hidden layers, with each subsequent layer receiving input from the previous layer. The first hidden layer was connected to the network input. Each layer adjusts the input via a weight matrix and the addition of a bias vector, and the final layer, followed by the application of a softmax activation function, produces the network output in the form of classification labels. The hidden layer has a neuron size of {8 16 16 32 16 32 16 16 8}, determined by trial and error, which ensures excellent performance. The output has three classes, i.e., S (stable, FOS > 1.2), M (marginally stable, 1.0 ≤ FOS ≤ 1.2), and U (unstable, FOS < 1.0). When U and M are in the same class, the output has two classes: S and U + M. The process of building a deep neural network for classification typically includes the following steps:
(a)
Data preprocessing: The first step is to prepare the input data and target labels used to train the network. This typically involves dividing the data into training and test sets.
(b)
Network construction and training: The next step is to train the network using the training data. A DNN model is trained on the input data, which requires specifying the input data, target labels, and the type of network to be trained.
(c)
Prediction: Once the network is trained, the model can be used to predict the test set.
(d)
Performance evaluation: The performance of the network can be evaluated using a confusion matrix.

3.2. A Deep Neural Network Model for the Factor of Safety Regression

A DNN for regression is a machine learning technique that utilizes neural networks to predict numerical values from input data [17,25]. Furthermore, it involves training a network using input data and their corresponding target values and generating predicted values for new input data. In this study, a feedforward DNN model was designed for FOS regression. The network architecture was similar to that of the DNN model used for classification, with the main difference being the output value. In the regression network, the predicted target values (FOS) were generated by the final fully connected layer. Using trial and error, the DNN model for regression consisted of 11 layers with neuron sizes of {8 16 16 32 32 64 32 32 16 16 8}, as illustrated in Figure 6. The procedure for a regression DNN typically involves the following steps:
(a)
Data preprocessing: The dataset is divided into training and test sets, and necessary preprocessing steps such as feature normalization and the handling of missing values are performed.
(b)
Network construction and training: A DNN consisting of input, hidden, and output layers is constructed, with the architecture of the network determined by factors such as the number of hidden layers and nodes. The network is trained using a training set.
(c)
Prediction: The trained model is utilized to generate predicted values for new input data.
(d)
Performance evaluation: The performance of the model was evaluated using metrics such as the mean squared error (MSE) and correlation coefficient R-square value.

4. Results and Discussion

4.1. Slope Classification

In this study, 70% of the generated dataset was used to train the machine learning classification model, and the remaining 30% was used for testing. To eliminate subjectivity in data selection, both the training and test datasets were randomly selected. The rectified linear unit (ReLU) function was used as the action function for the classification model. Table 2 lists the training error and accuracy for three classes (S, M, and U) and two classes (S and U + M). For three classes, the training accuracy was 0.9919 with a training error of 0.0081, whereas for two classes, the training accuracy was 0.9692 with a training error of 0.0308. Both had high accuracies (>0.9000).
For the three classes, using the model to predict the test set, the test accuracy of the model was approximately 0.9222 with a test error of 0.0778. According to the confusion matrix in Figure 7a, the test accuracy for the U (unstable) and S (stable) classes was above 95%, whereas that for the M (marginally stable) class was approximately 55%, owing to a data imbalance. Therefore, class M had relatively fewer cases than the other two classes. For example, referring to the dataset in Appendix A Table A1, the M class has just 76 cases because the range of the FOS is relatively small, from 1.0 to 1.2, while the U class has 266 cases, and the S class has 538 cases. Thus, the generated class M was significantly smaller than the other two classes. However, class M is marginally stable; thus, it is probably stable and unstable. Such an M-class slope is also extremely dangerous and requires more attention to avoid slope failures. This classification model was developed to classify slopes into three classes, including S (stable), M (marginally stable), and U (unstable), with a higher test accuracy of over 90%.
Similarly, when U and M are in the same class, another classification model was developed to categorize the slopes into two classes: S (stable, FOS > 1.2) and U + M (unstable, FOS < 1.2). Using this training model to predict the test set, the test accuracy was approximately 0.9315, which was slightly higher than the accuracy of the three-class model, and the test error was 0.0685. Figure 7b presents the confusion matrix for the test set, showing that the test accuracy for the S (stable) class was 93.3%, and the test accuracy for the U + M class was approximately 92.9%, both exceeding 90.0%. Thus, this classification model was also developed to classify slopes into two classes, S (stable) and U + M (unstable), with a higher test accuracy of over 90%. Regardless of the class, the machine learning model had a high accuracy of more than 90%. If the dataset is sufficient, the three classes are considered superior.
However, the cross-validation misclassification error provides an estimate of how well a model performs on the new data. In this study, a 10-fold cross-validation was conducted, and the findings are presented in Table 2. Therefore, for the three classes, the cross-validation training error of 0.1104 was higher than the training error of 0.0081, which was much closer to the test error of 0.0778. In addition, the cross-validation accuracy of 0.8896 was lower than the training accuracy of 0.9919, which was much closer to the test accuracy of 0.9222. Therefore, relying solely on the misclassification error of the training data underestimates the misclassification rate of the new data. Consequently, the cross-validation error provided a more accurate estimate of the performance of the model on the new data than the training error. Similar findings were also observed for both classes.

4.2. Slope FOS Prediction

In this study, 80% of the generated dataset was allocated to training, and the training and test sets were randomly selected. The ReLU function was selected as the activation function for the regression DNN model, with linear activation used as the output layer. Thus, 704 data points were used to construct the machine learning regression model with four input parameters (slope height, slope angle, cohesion, and internal friction angle) and one output target value of the FOS. Figure 8 displays the training loss curve with iterations, and Figure 9 depicts the relationship between the “true” FOS and the predicted FOS for the test set on slopes based on the regression model. The mean squared error (MSE) for the test set was approximately 5.03 × 10−4, and the correlation coefficient R-square for the regression model was 0.9989, indicating a strong linear relationship between the predicted FOS value and the true FOS obtained by numerical simulation. This means that this regression model can be used to predict FOS accurately. Thus, the regression model accurately predicted the FOS value and effectively modeled the relationship between the FOS values and the selected slope parameters.
Actually, the FOS provides a quantitative assessment of slope stability. Calculating the FOS is essential in slope stability analysis to evaluate safety, optimize design, manage risks, support decision making, and comply with regulatory requirements. The proper assessment and control of the FOS are essential for ensuring the long-term stability and reliability of slopes. Therefore, the regression model used to predict the factor of safety (FOS) holds significant meaning for engineering practices.

4.3. Time Consumption

Table 3 presents a comparison of the times consumed by the various calculation methods for slope stability analysis. The traditional method, which uses FLAC 3D to determine the FOS, requires approximately 125 s for a typical case, as shown in Figure 1. In contrast, using the machine learning model for classification, the elapsed times for the training and test set classification of the 880 datasets for the three classes were only 1.549320 s and 0.001513 s, respectively. Similarly, for regression, the elapsed times for the training and test set regressions of the 880 datasets were only 0.045756 s and 0.003117 s, respectively. These results show that a surrogate model based on machine learning can be utilized to predict the FOS values in real time. The significant difference in the elapsed time between the traditional method and the developed surrogate model for slope stability analysis highlights the potential benefits of machine learning. By allowing quick and accurate predictions of the FOS, decision makers can take prompt action to prevent and mitigate hazards, thereby reducing the risk of accidents and damage. Consequently, such a surrogate model using machine learning models can complement traditional computational methods, accelerate the FOS prediction process, and contribute to the development of effective and efficient risk management strategies.

4.4. Discussion

Using the traditional limit equilibrium method (LEM), the factor of safety is defined by the following equation [26]:
F O S = s h e a r   s t r e n g t h   o f   s o i l s h e a r   s t r e s s   r e q u i r e d   f o r   e q u i l i b r i u m ,
It can also be expressed as follows:
F O S = τ f i τ i = c + σ i t a n φ τ i ,
where τ i and σ i are the shear stress and normal stress at the i-th slice of the slip surface, respectively, and c and φ are the cohesion and internal friction angle, respectively. The critical slip surface corresponds to the surface that yields the lowest factor of safety (FOS), with this minimal value representing the true FOS.
According to the defined FOS by LEM, the cohesion and internal friction angle play crucial roles as parameters. As the surrogate model can predict the FOS, a comparison between the typical FOS calculations (based on Table 1; data with the FOS less than 1.0 have been removed) obtained from the traditional LEM method and the surrogate model is listed in Table 4. The results indicate that the developed surrogate model can accurately predict the FOS value. When the FOS obtained from the LEM method exceeds 1.0, the relative error is mostly below 10%, except for Case No. 4, which reaches approximately 15%. It is believed that enhancing the accuracy of the surrogate model would require a larger and more diverse dataset. The traditional method primarily focuses on the cohesion and friction angle, whereas the surrogate model incorporates two additional factors related to slope shape: slope height and slope angle. Although further improvements are necessary to enhance the accuracy of the surrogate model, Section 4.3 demonstrates the effectiveness of this approach.
By incorporating the findings of the surrogate model into traditional studies, researchers can enhance the efficiency of slope stability assessments. This integration allows for a more comprehensive analysis and a deeper understanding of slope behavior, ultimately leading to improved engineering practices and decision making in geotechnical engineering.
The surrogate model can be used to calibrate and validate traditional LEM parameters. By comparing the FOS predictions of the surrogate model with the results obtained from the LEM, researchers may assess the accuracy and reliability of the traditional method. This helps in fine-tuning the LEM parameters and improving its predictive capabilities. Additionally, researchers can systematically vary the input parameters within a range and observe the corresponding changes in the FOS predicted by the surrogate model. This analysis provides insights into the relative importance and influence of different factors on slope stability, aiding researchers in identifying critical parameters and optimizing their analyses.

4.5. Contributions, Limitations, and Further Research

The main strength of this work lies in the proposal of a surrogate model using machine learning to evaluate slope stability and compare the FOS value with the traditional LEM method. This study contributes to slope analysis in the following ways: (a) expanding classifications: the concept of a marginally stable class for slopes is introduced, which challenges the traditional binary classification of stable and unstable slopes; (b) surrogate modeling: the study showcases the potential of surrogate models in slope stability analysis, offering a cost-effective and time-efficient alternative to traditional methods.
Thus, when conducting a slope stability analysis, the classification model can be used first. Table 5 summarizes the countermeasures based on the slope classification. In cases in which the result falls into class U, prompt treatment should be considered, such as modifying the slope shape (slope height and slope angle) or increasing the strength of the soil (cohesion and friction angle) using methods like water drainage. Additionally, the regression model can also be used to obtain the FOS for slope analysis if it exceeds 1.0. For slopes classified as class M, real-time monitoring and early warning are necessary to prevent sudden slope failure. When the slope is classified as class S, regular monitoring and maintenance can be implemented based on the FOS values. A higher FOS indicates a greater safety margin for the slope. The proposed countermeasures can be applied to the slope to prevent or mitigate the potential risks posed by landslides.
However, this work has several limitations that need to be acknowledged. Firstly, there is a lack of empirical validation for the surrogate model, specifically regarding its ability to accurately predict unstable slopes leading to landslide events. As a result, the utilization of this model for decision making purposes requires further clarification and study. Another limitation of the current study is its focus on homogeneous slopes with constant Mohr–Coulomb shear strength parameters. Future studies should aim to incorporate more realistic conditions that closely resemble actual field sites. For instance, researchers can consider incorporating complex soil and/or rock layers, as well as variations in other factors such as pore water pressure, rainfall, and seismic events. In doing so, the applicability and validity of the findings in practical scenarios can be enhanced. A final limitation worth noting is the issue of data imbalance, which requires greater attention.
In future studies, it is recommended to utilize newer and more diverse datasets with various input parameters to predict the factor of safety (FOS) in different slopes. This approach would help identify the most accurate algorithms and determine the most effective parameters that influence slope stability. In conclusion, while this work presents a valuable contribution to slope stability analysis, it is important to address the aforementioned limitations in future studies to further refine the models and enhance their applicability and reliability.

5. Conclusions

This study investigated the application of a surrogate model using machine learning to evaluate slope stability by capturing the intricate nonlinear and multidimensional relationships between the parameters. The slope analysis problem is divided into two parts: the classification of slopes into stable, marginally stable, and unstable classes and regression to predict the factor of safety (FOS) value. This study utilized a comprehensive database of 880 homogeneous slopes generated by the FLAC 3D Version 7.0 software for surrogate model development. The classification model was efficient, achieving a test accuracy of 0.9222, with a class accuracy of 96.2% for the U class (unstable), 55% for the M class (marginally stable), and 95.2% for the S class (stable). When U and M are in the same class (i.e., the U + M class), the test accuracy is 0.9315, with 93.3% accuracy for the S class and 92.9% accuracy for the U + M class. The regression model demonstrated a high correlation coefficient R-square value of 0.9989 and a low test MSE value of 5.03 × 10−4, indicating a strong relationship between the FOS values and the selected slope parameters. However, the generated dataset may not be representative of all the actual site conditions, and more complex geological conditions and other input factors must be considered. Moreover, such a surrogate model can complement traditional computational methods and accelerate the prediction of the FOS in slope stability analysis. This capability enables decision makers to promptly take the necessary actions to prevent and mitigate potential hazards, thus contributing to the development of effective and efficient risk management strategies. Incorporating surrogate models into slope stability problems can effectively achieve long-term sustainability while minimizing risks and preserving natural resources.

Author Contributions

Conceptualization, X.L., M.N. and P.-j.C.; methodology, X.L.; software, K.S., S.I. and X.L.; validation, X.L., M.N. and P.-j.C.; formal analysis, X.L.; writing—original draft preparation, X.L.; writing—review and editing, M.N. and P.-j.C.; supervision, M.N. and P.-j.C.; funding acquisition, M.N. and P.-j.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by JST (Moonshot Research and Development) (grant number JPMJMS2032).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are available upon reasonable request from the corresponding author.

Acknowledgments

The authors would like to extend their sincere gratitude to the reviewers for their invaluable contributions and insightful feedback, which have greatly improved the quality of this research.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A

Table A1. Dataset used in this study.
Table A1. Dataset used in this study.
Case No.Slope Height/mSlope Angle/°Cohesion/kPaFriction Angle/°FOSLabels
1326.5725-U
2326.57210-U
3326.57215-U
4326.57220-U
5326.57225-U
6326.57230-U
7326.57235-U
8326.57240-U
9326.57245-U
10326.5755-U
11326.57510-U
12326.57515-U
13326.57520-U
14326.57525-U
15326.57530-U
16326.57535-U
17326.57540-U
18326.57545-U
19326.57105-U
20326.571010-U
21326.571015-U
22326.571020-U
23326.571025-U
24326.571030-U
25326.5710351.04M
26326.5710401.12M
27326.5710451.2M
28326.57155-U
29326.571510-U
30326.5715151.04M
31326.5715201.12M
32326.5715251.2M
33326.5715301.28S
34326.5715351.36S
35326.5715401.46S
36326.5715451.54S
37326.572051.14M
38326.5720101.23S
39326.5720151.31S
40326.5720201.4S
41326.5720251.49S
42326.5720301.57S
43326.5720351.67S
44326.5720401.76S
45326.5720451.86S
46326.572551.4S
47326.5725101.49S
48326.5725151.58S
49326.5725201.68S
50326.5725251.76S
51326.5725301.85S
52326.5725351.95S
53326.5725402.05S
54326.5725452.17S
55326.573051.67S
56326.5730101.76S
57326.5730151.85S
58326.5730201.93S
59326.5730252.04S
60326.5730302.13S
61326.5730352.23S
62326.5730402.34S
63326.5730452.46S
64326.573551.91S
65326.5735102.02S
66326.5735152.12S
67326.5735202.21S
68326.5735252.3S
69326.5735302.4S
70326.5735352.51S
71326.5735402.62S
72326.5735452.74S
73326.574052.16S
74326.5740102.29S
75326.5740152.38S
76326.5740202.48S
77326.5740252.57S
78326.5740302.68S
79326.5740352.79S
80326.5740402.9S
81326.5740453.02S
82326.574552.41S
83326.5745102.55S
84326.5745152.64S
85326.5745202.74S
86326.5745252.83S
87326.5745302.93S
88326.5745353.07S
89326.5745403.18S
90326.5745453.3S
91326.575052.69S
92326.5750102.8S
93326.5750152.9S
94326.5750203S
95326.5750253.1S
96326.5750303.2S
97326.5750353.33S
98326.5750403.45S
99326.5750453.58S
100326.5720-U
101326.5750-U
102326.57100-U
103326.57150-U
104326.572001M
105326.572501.25S
106326.573001.5S
107326.573501.76S
108326.574002.02S
109326.574502.27S
110326.575002.51S
11134525-U
112345210-U
113345215-U
114345220-U
115345225-U
116345230-U
117345235-U
118345240-U
119345245-U
12034555-U
121345510-U
122345515-U
123345520-U
124345525-U
125345530-U
126345535-U
127345540-U
128345545-U
129345105-U
1303451010-U
1313451015-U
1323451020-U
1333451025-U
1343451030-U
13534510351.04M
13634510401.12M
13734510451.2M
138345155-U
1393451510-U
14034515151.04M
14134515201.13M
14234515251.21S
14334515301.28S
14434515351.36S
14534515401.44S
14634515451.54S
1473452051.13M
14834520101.23S
14934520151.31S
15034520201.41S
15134520251.49S
15234520301.57S
15334520351.67S
15434520401.76S
15534520451.87S
1563452551.38S
15734525101.49S
15834525151.59S
15934525201.68S
16034525251.76S
16134525301.85S
16234525351.95S
16334525402.06S
16434525452.16S
1653453051.63S
16634530101.75S
16734530151.85S
16834530201.94S
16934530252.04S
17034530302.14S
17134530352.23S
17234530402.34S
17334530452.46S
1743453551.85S
17534535102.01S
17634535152.12S
17734535202.2S
17834535252.31S
17934535302.42S
18034535352.5S
18134535402.61S
18234535452.74S
1833454052.12S
18434540102.25S
18534540152.39S
18634540202.47S
18734540252.58S
18834540302.69S
18934540352.77S
19034540402.91S
19134540453.02S
1923454552.36S
19334545102.5S
19434545152.63S
19534545202.74S
19634545252.85S
19734545302.94S
19834545353.04S
19934545403.18S
20034545453.32S
2013455052.58S
20234550102.74S
20334550152.88S
20434550202.99S
20534550253.1S
20634550303.21S
20734550353.32S
20834550403.46S
20934550453.59S
21034520-U
21134550-U
212345100-U
213345150-U
214345200-U
2153452501.17M
2163453001.4S
2173453501.64S
2183454001.87S
2193454502.13S
2203455002.36S
221363.4325-U
222363.43210-U
223363.43215-U
224363.43220-U
225363.43225-U
226363.43230-U
227363.43235-U
228363.43240-U
229363.43245-U
230363.4355-U
231363.43510-U
232363.43515-U
233363.43520-U
234363.43525-U
235363.43530-U
236363.43535-U
237363.43540-U
238363.43545-U
239363.43105-U
240363.431010-U
241363.431015-U
242363.431020-U
243363.431025-U
244363.431030-U
245363.4310351.04M
246363.4310401.12M
247363.4310451.2M
248363.43155-U
249363.431510-U
250363.4315151.05M
251363.4315201.12M
252363.4315251.21S
253363.4315301.29S
254363.4315351.36S
255363.4315401.45S
256363.4315451.53S
257363.432051.12M
258363.4320101.23S
259363.4320151.31S
260363.4320201.4S
261363.4320251.48S
262363.4320301.57S
263363.4320351.67S
264363.4320401.75S
265363.4320451.86S
266363.432551.34S
267363.4325101.5S
268363.4325151.59S
269363.4325201.68S
270363.4325251.75S
271363.4325301.85S
272363.4325351.94S
273363.4325402.05S
274363.4325452.16S
275363.433051.59S
276363.4330101.74S
277363.4330151.85S
278363.4330201.95S
279363.4330252.04S
280363.4330302.13S
281363.4330352.23S
282363.4330402.32S
283363.4330452.44S
284363.433551.8S
285363.4335101.99S
286363.4335152.1S
287363.4335202.2S
288363.4335252.3S
289363.4335302.4S
290363.4335352.51S
291363.4335402.63S
292363.4335452.75S
293363.434052.05S
294363.4340102.22S
295363.4340152.35S
296363.4340202.46S
297363.4340252.57S
298363.4340302.68S
299363.4340352.79S
300363.4340402.9S
301363.4340453.04S
302363.434552.27S
303363.4345102.46S
304363.4345152.63S
305363.4345202.74S
306363.4345252.85S
307363.4345302.96S
308363.4345353.06S
309363.4345403.17S
310363.4345453.28S
311363.435052.49S
312363.4350102.68S
313363.4350152.87S
314363.4350203.01S
315363.4350253.12S
316363.4350303.23S
317363.4350353.31S
318363.4350403.45S
319363.4350453.58S
320363.4320-U
321363.4350-U
322363.43100-U
323363.43150-U
324363.43200-U
325363.432501.15M
326363.433001.36S
327363.433501.56S
328363.434001.81S
329363.434502.05S
330363.435002.27S
331626.5725-U
332626.57210-U
333626.57215-U
334626.57220-U
335626.57225-U
336626.57230-U
337626.57235-U
338626.57240-U
339626.57245-U
340626.5755-U
341626.57510-U
342626.57515-U
343626.57520-U
344626.57525-U
345626.57530-U
346626.57535-U
347626.57540-U
348626.57545-U
349626.57105-U
350626.571010-U
351626.571015-U
352626.571020-U
353626.571025-U
354626.571030-U
355626.5710351.05M
356626.5710401.11M
357626.5710451.2M
358626.57155-U
359626.571510-U
360626.5715151.05M
361626.5715201.13M
362626.5715251.2M
363626.5715301.29S
364626.5715351.36S
365626.5715401.46S
366626.5715451.54S
367626.57205-U
368626.5720101.21S
369626.5720151.32S
370626.5720201.4S
371626.5720251.49S
372626.5720301.57S
373626.5720351.66S
374626.5720401.75S
375626.5720451.86S
376626.572551.18M
377626.5725101.43S
378626.5725151.58S
379626.5725201.67S
380626.5725251.76S
381626.5725301.85S
382626.5725351.94S
383626.5725402.05S
384626.5725452.17S
385626.573051.36S
386626.5730101.64S
387626.5730151.81S
388626.5730201.87S
389626.5730252.01S
390626.5730302.13S
391626.5730352.23S
392626.5730402.34S
393626.5730452.46S
394626.573551.51S
395626.5735101.81S
396626.5735152.03S
397626.5735202.2S
398626.5735252.27S
399626.5735302.41S
400626.5735352.5S
401626.5735402.62S
402626.5735452.74S
403626.574051.73S
404626.5740101.99S
405626.5740152.24S
406626.5740202.45S
407626.5740252.55S
408626.5740302.68S
409626.5740352.77S
410626.5740402.9S
411626.5740453.02S
412626.574551.88S
413626.5745102.18S
414626.5745152.45S
415626.5745202.68S
416626.5745252.84S
417626.5745302.95S
418626.5745353.06S
419626.5745403.17S
420626.5745453.31S
421626.575052.09S
422626.5750102.36S
423626.5750152.63S
424626.5750202.86S
425626.5750253.08S
426626.5750303.15S
427626.5750353.27S
428626.5750403.45S
429626.5750453.58S
430626.5720-U
431626.5750-U
432626.57100-U
433626.57150-U
434626.57200-U
435626.57250-U
436626.573001.08M
437626.573501.27S
438626.574001.44S
439626.574501.6S
440626.575001.81S
44164525-U
442645210-U
443645215-U
444645220-U
445645225-U
446645230-U
447645235-U
4486452401M
4496452451.15M
45064555-U
451645510-U
452645515-U
453645520-U
454645525-U
4556455301.1M
4566455351.25S
4576455401.39S
4586455451.57S
459645105-U
4606451010-U
46164510151.02M
46264510201.17M
46364510251.32S
46464510301.46S
46564510351.63S
46664510401.82S
46764510452.02S
468645155-U
46964515101.15M
47064515151.32S
47164515201.47S
47264515251.63S
47364515301.8S
47464515351.98S
47564515402.17S
47664515452.39S
4776452051.22S
47864520101.42S
47964520151.59S
48064520201.79S
48164520251.93S
48264520302.1S
48364520352.29S
48464520402.51S
48564520452.73S
4866452551.43S
48764525101.68S
48864525151.87S
48964525202.05S
49064525252.23S
49164525302.41S
49264525352.6S
49364525402.81S
49464525453.06S
4956453051.53S
49664530101.94S
49764530152.13S
49864530202.32S
49964530252.52S
50064530302.7S
50164530352.91S
50264530403.12S
50364530453.38S
5046453551.97S
50564535102.19S
50664535152.38S
50764535202.6S
50864535252.79S
50964535303.01S
51064535353.2S
51164535403.42S
51264535453.67S
5136454052.22S
51464540102.44S
51564540152.65S
51664540202.85S
51764540253.08S
51864540303.3S
51964540353.55S
52064540403.7S
52164540453.98S
5226454552.46S
52364545102.71S
52464545152.9S
52564545203.12S
52664545253.32S
52764545303.55S
52864545353.81S
52964545404.04S
53064545454.27S
5316455052.71S
53264550102.95S
53364550153.15S
53464550203.37S
53564550253.58S
53664550303.8S
53764550354.07S
53864550404.33S
53964550454.57S
54064520-U
54164550-U
542645100-U
543645150-U
5446452001M
5456452501.2M
5466453001.46S
5476453501.61S
5486454001.95S
5496454502.19S
5506455002.37S
551663.4325-U
552663.43210-U
553663.43215-U
554663.43220-U
555663.43225-U
556663.43230-U
557663.43235-U
558663.43240-U
559663.43245-U
560663.4355-U
561663.43510-U
562663.43515-U
563663.43520-U
564663.43525-U
565663.43530-U
566663.43535-U
567663.43540-U
568663.435451.01M
569663.43105-U
570663.431010-U
571663.431015-U
572663.431020-U
573663.431025-U
574663.4310301.07M
575663.4310351.16M
576663.4310401.31S
577663.4310451.39S
578663.43155-U
579663.431510-U
580663.4315151.04M
581663.4315201.14M
582663.4315251.25S
583663.4315301.35S
584663.4315351.46S
585663.4315401.59S
586663.4315451.75S
587663.432051.03M
588663.4320101.16M
589663.4320151.29S
590663.4320201.41S
591663.4320251.52S
592663.4320301.63S
593663.4320351.74S
594663.4320401.87S
595663.4320452.02S
596663.432551.26S
597663.4325101.39S
598663.4325151.53S
599663.4325201.65S
600663.4325251.77S
601663.4325301.9S
602663.4325352.02S
603663.4325402.15S
604663.4325452.28S
605663.433051.48S
606663.4330101.63S
607663.4330151.75S
608663.4330201.88S
609663.4330252.01S
610663.4330302.15S
611663.4330352.29S
612663.4330402.42S
613663.4330452.55S
614663.433551.71S
615663.4335101.84S
616663.4335151.99S
617663.4335202.12S
618663.4335252.25S
619663.4335302.39S
620663.4335352.54S
621663.4335402.69S
622663.4335452.84S
623663.434051.92S
624663.4340102.08S
625663.4340152.21S
626663.4340202.35S
627663.4340252.49S
628663.4340302.63S
629663.4340352.78S
630663.4340402.94S
631663.4340453.13S
632663.434552.14S
633663.4345102.29S
634663.4345152.43S
635663.4345202.57S
636663.4345252.71S
637663.4345302.86S
638663.4345353.02S
639663.4345403.18S
640663.4345453.37S
641663.435052.36S
642663.4350102.53S
643663.4350152.67S
644663.4350202.81S
645663.4350252.95S
646663.4350303.09S
647663.4350353.25S
648663.4350403.44S
649663.4350453.61S
650663.4320-U
651663.4350-U
652663.43100-U
653663.43150-U
654663.43200-U
655663.432501.07M
656663.433001.33S
657663.433501.55S
658663.434001.76S
659663.434501.97S
660663.435002.2S
661124525-U
6621245210-U
6631245215-U
6641245220-U
6651245225-U
6661245230-U
6671245235-U
6681245240-U
66912452451M
670124555-U
6711245510-U
6721245515-U
6731245520-U
6741245525-U
6751245530-U
6761245535-U
67712455401.04M
67812455451.19M
6791245105-U
68012451010-U
68112451015-U
68212451020-U
68312451025-U
684124510301.06M
685124510351.2M
686124510401.33S
687124510451.47S
6881245155-U
68912451510-U
69012451515-U
69112451520-U
692124515251.11M
693124515301.26S
694124515351.4S
695124515401.55S
696124515451.71S
6971245205-U
69812452010-U
699124520151M
700124520201.12M
701124520251.29S
702124520301.42S
703124520351.57S
704124520401.76S
705124520451.96S
7061245255-U
70712452510-U
708124525151.13M
709124525201.26S
710124525251.44S
711124525301.6S
712124525351.75S
713124525401.93S
714124525452.14S
7151245305-U
716124530101.07M
717124530151.22S
718124530201.43S
719124530251.55S
720124530301.74S
721124530351.93S
722124530402.11S
723124530452.3S
72412453551.01M
725124535101.19M
726124535151.37S
727124535201.53S
728124535251.67S
729124535301.89S
730124535352.07S
731124535402.27S
732124535452.47S
73312454051.1M
734124540101.29S
735124540151.45S
736124540201.64S
737124540251.86S
738124540301.97S
739124540352.22S
740124540402.44S
741124540452.65S
74212454551.23S
743124545101.42S
744124545151.59S
745124545201.75S
746124545251.97S
747124545302.11S
748124545352.35S
749124545402.57S
750124545452.79S
75112455051.34S
752124550101.53S
753124550151.72S
754124550201.89S
755124550252.07S
756124550302.31S
757124550352.52S
758124550402.73S
759124550452.96S
760124520-U
761124550-U
7621245100-U
7631245150-U
7641245200-U
7651245250-U
7661245300-U
7671245350-U
7681245400-U
76912454501.04M
77012455001.15M
7711263.4325-U
7721263.43210-U
7731263.43215-U
7741263.43220-U
7751263.43225-U
7761263.43230-U
7771263.43235-U
7781263.43240-U
7791263.43245-U
7801263.4355-U
7811263.43510-U
7821263.43515-U
7831263.43520-U
7841263.43525-U
7851263.43530-U
7861263.43535-U
7871263.43540-U
7881263.43545-U
7891263.43105-U
7901263.431010-U
7911263.431015-U
7921263.431020-U
7931263.431025-U
7941263.431030-U
7951263.431035-U
7961263.431040-U
7971263.4310451.02M
7981263.43155-U
7991263.431510-U
8001263.431515-U
8011263.431520-U
8021263.431525-U
8031263.431530-U
8041263.431535-U
8051263.4315401.04M
8061263.4315451.14M
8071263.43205-U
8081263.432010-U
8091263.432015-U
8101263.432020-U
8111263.432025-U
8121263.4320301.04M
8131263.4320351.14M
8141263.4320401.21S
8151263.4320451.31S
8161263.43255-U
8171263.432510-U
8181263.432515-U
8191263.432520-U
8201263.4325251.09M
8211263.4325301.19M
8221263.4325351.3S
8231263.4325401.39S
8241263.4325451.48S
8251263.43305-U
8261263.433010-U
8271263.4330151.01M
8281263.4330201.13M
8291263.4330251.23S
8301263.4330301.31S
8311263.4330351.42S
8321263.4330401.54S
8331263.4330451.66S
8341263.43355-U
8351263.4335101.03M
8361263.4335151.14M
8371263.4335201.27S
8381263.4335251.35S
8391263.4335301.46S
8401263.4335351.56S
8411263.4335401.62S
8421263.4335451.81S
8431263.434051.01M
8441263.4340101.13M
8451263.4340151.29S
8461263.4340201.24S
8471263.4340251.49S
8481263.4340301.58S
8491263.4340351.7S
8501263.4340401.81S
8511263.4340451.97S
8521263.434551.13M
8531263.4345101.25S
8541263.4345151.4S
8551263.4345201.51S
8561263.4345251.64S
8571263.4345301.73S
8581263.4345351.83S
8591263.4345401.96S
8601263.4345452.07S
8611263.435051.21S
8621263.4350101.31S
8631263.4350151.49S
8641263.4350201.64S
8651263.4350251.7S
8661263.4350301.83S
8671263.4350351.97S
8681263.4350402.06S
8691263.4350452.24S
8701263.4320-U
8711263.4350-U
8721263.43100-U
8731263.43150-U
8741263.43200-U
8751263.43250-U
8761263.43300-U
8771263.43350-U
8781263.43400-U
8791263.43450-U
8801263.435001.03M

References

  1. Massey, C.; Della Pasqua, F.; Holden, C.; Kaiser, A.; Richards, L.; Wartman, J.; McSaveney, M.J.; Archibald, G.; Yetton, M.; Janku, L. Rock slope response to strong earthquake shaking. Landslides 2017, 14, 249–268. [Google Scholar] [CrossRef]
  2. Li, Q.; Wang, Y.M.; Zhang, K.B.; Yu, H.; Tao, Z.Y. Field investigation and numerical study of a siltstone slope instability induced by excavation and rainfall. Landslides 2020, 17, 1485–1499. [Google Scholar] [CrossRef]
  3. Nagatani, K.; Abe, M.; Osuka, K.; Chun, P.-j.; Okatani, T.; Nishio, M.; Chikushi, S.; Matsubara, T.; Ikemoto, Y.; Asama, H. Innovative technologies for infrastructure construction and maintenance through collaborative robots based on an open design approach. Adv. Robot. 2021, 2021, 715–722. [Google Scholar] [CrossRef]
  4. Das, S.K.; Biswal, R.K.; Sivakugan, N.; Das, B. Classification of slopes and prediction of factor of safety using differential evolution neural networks. Environ. Earth Sci. 2011, 64, 201–210. [Google Scholar] [CrossRef]
  5. Sloan, S.W. Geotechnical stability analysis. Geotechnique 2013, 63, 531–572. [Google Scholar] [CrossRef] [Green Version]
  6. Cheng, Y.M.; Lansivaara, T.; Wei, W.B. Two-dimensional slope stability analysis by limit equilibrium and strength reduction methods. Comput. Geotech. 2007, 34, 137–150. [Google Scholar] [CrossRef]
  7. Reale, C.; Xue, J.; Gavin, K. System reliability of slopes using multimodal optimisation. Geotechnique 2016, 66, 413–423. [Google Scholar] [CrossRef] [Green Version]
  8. Tschuchnigg, F.; Schweiger, H.F.; Sloan, S.W.; Lyamin, A.V.; Raissakis, I. Comparison of finite-element limit analysis and strength reduction techniques. Geotechnique 2015, 65, 249–257. [Google Scholar] [CrossRef]
  9. Song, D.; Chen, Z.; Chao, H.; Ke, Y.; Nie, W. Numerical study on seismic response of a rock slope with discontinuities based on the time-frequency joint analysis method. Soil Dyn. Earthq. Eng. 2020, 133, 106112. [Google Scholar] [CrossRef]
  10. Sakellariou, M.G.; Ferentinou, M.D. A study of slope stability prediction using neural networks. Geotech. Geol. Eng. 2005, 23, 419–445. [Google Scholar] [CrossRef]
  11. Samui, P. Slope stability analysis: A support vector machine approach. Environ. Geol. 2008, 56, 255–267. [Google Scholar] [CrossRef]
  12. Zhou, J.; Li, E.; Yang, S.; Wang, M.; Shi, X.; Yao, S.; Mitri, H.S. Slope stability prediction for circular mode failure using gradient boosting machine approach based on an updated database of case histories. Saf. Sci. 2019, 118, 505–518. [Google Scholar] [CrossRef]
  13. Mahmoodzadeh, A.; Mohammadi, M.; Farid Hama Ali, H.; Hashim Ibrahim, H.; Nariman Abdulhamid, S.; Nejati, H.R. Prediction of safety factors for slope stability: Comparison of machine learning techniques. Nat. Hazards 2022, 111, 1771–1799. [Google Scholar] [CrossRef]
  14. Shi, N.; Xu, J.; Wurster, S.W.; Guo, H.; Woodring, J.; Van Roekel, L.P.; Shen, H.W. GNN-Surrogate: A Hierarchical and Adaptive Graph Neural Network for Parameter Space Exploration of Unstructured-Mesh Ocean Simulations. IEEE Trans. Vis. Comput. Graph. 2022, 28, 2301–2313. [Google Scholar] [CrossRef] [PubMed]
  15. Jain, A.K.; Mao, J.; Mohiuddin, K.M. Artificial neural networks: A tutorial. Computer 1996, 29, 31–44. [Google Scholar] [CrossRef] [Green Version]
  16. Dreiseitl, S.; Ohno-Machado, L. Logistic regression and artificial neural network classification models: A methodology review. J. Biomed. Inform. 2002, 35, 352–359. [Google Scholar] [CrossRef] [Green Version]
  17. Georgevici, A.I.; Terblanche, M. Neural networks and deep learning: A brief introduction. Intensive Care Med. 2019, 45, 712–714. [Google Scholar] [CrossRef] [Green Version]
  18. Dahiya, N.; Saini, B.; Chalak, H.D. Deep neural network-based storey drift modelling of precast concrete structures using RStudio. J. Soft Comput. Civ. Eng. 2021, 5, 88–100. [Google Scholar] [CrossRef]
  19. Yamane, T.; Chun, P.-J. Crack detection from a concrete surface image based on semantic segmentation using deep learning. J. Adv. Concr. Technol. 2020, 18, 493–504. [Google Scholar] [CrossRef]
  20. Chun, P.-j.; Yamane, T.; Tsuzuki, Y. Automatic detection of cracks in asphalt pavement using deep learning to overcome weaknesses in images and gis visualization. Appl. Sci. 2021, 11, 892. [Google Scholar] [CrossRef]
  21. Chun, P.-j.; Yamane, T.; Maemura, Y. A deep learning-based image captioning method to automatically generate comprehensive explanations of bridge damage. Comput. Civ. Infrastruct. Eng. 2022, 37, 1387–1401. [Google Scholar] [CrossRef]
  22. Yamane, T.; Chun, P.-j.; Dang, J.; Honda, R. Recording of bridge damage areas by 3D integration of multiple images and reduction of the variability in detected results. Comput. Civ. Infrastruct. Eng. 2023, 1–17. [Google Scholar] [CrossRef]
  23. Xu, J.J.; Zhang, H.; Tang, C.S.; Cheng, Q.; Liu, B.; Shi, B. Automatic soil desiccation crack recognition using deep learning. Geotechnique 2022, 72, 337–349. [Google Scholar] [CrossRef]
  24. Lozano-Diez, A.; Zazo, R.; Toledano, D.T.; Gonzalez-Rodriguez, J. An analysis of the influence of deep neural network (DNN) topology in bottleneck feature based language recognition. PLoS ONE 2017, 12, e0182580. [Google Scholar] [CrossRef] [Green Version]
  25. Xu, Y.; Du, J.; Dai, L.R.; Lee, C.H. A regression approach to speech enhancement based on deep neural networks. IEEE/ACM Trans. Audio Speech Lang. Process. 2015, 23, 7–19. [Google Scholar] [CrossRef]
  26. Liu, S.Y.; Shao, L.T.; Li, H.J. Slope stability analysis using the limit equilibrium method and two finite element methods. Comput. Geotech. 2015, 63, 291–298. [Google Scholar] [CrossRef]
Figure 1. A basic model for a simple and homogeneous soil slope with an FOS of 1.640 using numerical simulation.
Figure 1. A basic model for a simple and homogeneous soil slope with an FOS of 1.640 using numerical simulation.
Sustainability 15 10793 g001
Figure 2. Frequency distribution diagram of the generated FOS database.
Figure 2. Frequency distribution diagram of the generated FOS database.
Sustainability 15 10793 g002
Figure 3. Scatterplots of all input factors: (a) slope height, (b) slope angle, (c) cohesion, and (d) internal friction angle with the obtained FOS.
Figure 3. Scatterplots of all input factors: (a) slope height, (b) slope angle, (c) cohesion, and (d) internal friction angle with the obtained FOS.
Sustainability 15 10793 g003
Figure 4. A typical relationship between the FOS and the internal friction angle with slope height of 6 m, slope angle of 45°, and various cohesion values (2–50 kPa).
Figure 4. A typical relationship between the FOS and the internal friction angle with slope height of 6 m, slope angle of 45°, and various cohesion values (2–50 kPa).
Sustainability 15 10793 g004
Figure 5. A deep neural network model for slope classification.
Figure 5. A deep neural network model for slope classification.
Sustainability 15 10793 g005
Figure 6. A deep neural network model for regression.
Figure 6. A deep neural network model for regression.
Sustainability 15 10793 g006
Figure 7. Confusion matrix for the test set: (a) three classes and (b) two classes.
Figure 7. Confusion matrix for the test set: (a) three classes and (b) two classes.
Sustainability 15 10793 g007
Figure 8. Training loss curve.
Figure 8. Training loss curve.
Sustainability 15 10793 g008
Figure 9. The relationship between the predicted FOS and the “true” FOS using numerical simulation based on the test set.
Figure 9. The relationship between the predicted FOS and the “true” FOS using numerical simulation based on the test set.
Sustainability 15 10793 g009
Table 1. Factors of safety (FOS) using numerical simulation and LEM.
Table 1. Factors of safety (FOS) using numerical simulation and LEM.
No.Cohesion
/kPa
Friction Angle
FOS *FOS_LEM
[6]
FOS DifferenceRelative Error
/%
125-0.25--
2215-0.50--
3225-0.74--
42451.151.350.2014.81
555-0.41--
6515-0.70--
7525-0.98--
85351.251.280.032.34
95451.571.650.084.85
10105-0.65--
1110151.020.980.044.08
1210251.321.300.021.54
1310351.641.630.010.61
1410452.022.040.020.98
152051.221.060.1615.09
1620151.591.480.117.43
1720251.931.850.084.32
1820352.292.240.052.23
1920452.732.690.041.49
2050-0.20--
21100-0.40--
222001.000.800.2025.00
* ‘-’ means that the FOS value is less than 1.0 without a specified output value.
Table 2. Error and accuracy of training, cross-validation training, and test sets.
Table 2. Error and accuracy of training, cross-validation training, and test sets.
OutputTrainingCross-Validation TrainingTest
Three classesError0.00810.11040.0778
Accuracy0.99190.88960.9222
Two classesError0.03080.09080.0685
Accuracy0.96920.90920.9315
Table 3. Time consumption of various calculation methods for slope stability analysis.
Table 3. Time consumption of various calculation methods for slope stability analysis.
MethodsNumerical
Simulation
Classification
(Three Classes)
Regression
Set-TrainingTestTrainingTest
Case No.1616 (70%)264 (30%)704 (80%)176 (20%)
Time/s1251.5493200.0015130.0457560.003117
Table 4. Typical factors of safety (FOS) using LEM and the developed surrogate model.
Table 4. Typical factors of safety (FOS) using LEM and the developed surrogate model.
No.Cohesion
/kPa
Friction Angle
FOS_LEM
[6]
FOS by
Surrogate Model
The DifferenceRelative Error
/%
42451.351.16−0.19−14.37
85351.281.24−0.04−3.41
95451.651.54−0.11−6.72
1210251.301.330.031.97
1310351.631.630.000.04
1410452.042.02−0.02−1.05
152051.061.160.109.58
1620151.481.600.128.11
1720251.851.930.084.56
1820352.242.290.052.40
1920452.692.750.062.21
Table 5. Proposed countermeasures based on the surrogate model.
Table 5. Proposed countermeasures based on the surrogate model.
Slope ClassificationFOS Countermeasures
U<1.0Prompt treatment
M1.0 ≤ FOS ≤ 1.2Monitoring and early warning
S>1.2Regular monitoring and maintenance
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Li, X.; Nishio, M.; Sugawara, K.; Iwanaga, S.; Chun, P.-j. Surrogate Model Development for Slope Stability Analysis Using Machine Learning. Sustainability 2023, 15, 10793. https://doi.org/10.3390/su151410793

AMA Style

Li X, Nishio M, Sugawara K, Iwanaga S, Chun P-j. Surrogate Model Development for Slope Stability Analysis Using Machine Learning. Sustainability. 2023; 15(14):10793. https://doi.org/10.3390/su151410793

Chicago/Turabian Style

Li, Xianfeng, Mayuko Nishio, Kentaro Sugawara, Shoji Iwanaga, and Pang-jo Chun. 2023. "Surrogate Model Development for Slope Stability Analysis Using Machine Learning" Sustainability 15, no. 14: 10793. https://doi.org/10.3390/su151410793

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop