Assessment of Shear Strength of Fine Grained and Coarse Grained Soil Using Actual EPB TBM Operating Data

Samadi et al.
International Journal of
International Journal of Geo-Engineering (2024) 15:20
https://doi.org/10.1186/s40703-024-00220-6 Geo-Engineering
ORIGINAL RESEARCH Open Access
Assessment of shear strength of fine‑grained

and coarse‑grained soil using actual EPB‑TBM
operating data
Hanan Samadi1 , Jafar Hassanpour1* and Jamal Rostami2
*Correspondence:
hassanpour@ut.ac.ir Abstract
1
School of Geology, College The necessity of predicting geotechnical parameters in soft ground tunnelling
of Science, University of Tehran, is crucial for selecting the appropriate tunnel boring machine (TBM), evaluating
Tehran, Iran the operating limit of earth pressure balance (EPB) machines’ parameters, and ensur-
2
Earth Mechanic Institute,
Department of Mining ing the safety and efficiency of TBMs during tunnel construction. In this research,
Engineering, Colorado School various EPB operating parameters such as cutterhead torque, thrust force, chamber
of Mines, Golden, CO, USA pressure, and cutterhead rotation speed (RPM) were utilized to estimate geotechni-
cal parameters like friction angle (φ) and shear strength (τ) for engineering geological
units ET1 to ET5 (fine-grained and coarse-grained soils) along the tunnels route, which
serve as indicative units for the entire tunnels path. Statistical methods and compu-
tational techniques, namely support vector regression (SVR) and feed-forward neural
network (FFNN), were trained using EPB operating parameters and geotechnical data
from Tehran metro line 6—southern extension sector (TML-SE6) and the East–west
section of line 7, Tehran metro project (TML-EW7). A comprehensive dataset compris-
ing borehole logging results along the tunnel path was gathered, with 85% of the data
randomly selected for training and the remaining 15% reserved for model testing.
Various loss functions and statistical metrics were employed to evaluate the accuracy
and precision of the method. The results of the proposed models demonstrate satisfac-
tory and reliable accuracy of the approaches.
Highlights
• Database collected, generated, and analyzed using daily operational records, TBM
data logger, in-situ analysis, laboratory tests, and statistical investigations;
• New formulas developed to predict geotechnical parameters of soil materials
by incorporating various operating parameters of the Tunnel boring machine;
• Investigation of predictive abilities of smart algorithms in estimating shear
strength and internal friction angle;
• Comprehensive evaluation conducted to compare and contrast empirical
and machine learning outcomes;
• Accuracy of models evaluated using multiple assessment metrics.
© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits
use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original
author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third
party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the mate-
rial. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or
exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://
creativecommons.org/licenses/by/4.0/.
Samadi et al. International Journal of Geo-Engineering (2024) 15:20 Page 2 of 18
Keywords: Soil strength parameters, EPB-TBM, Computing machinery, Sensitivity

analysis, Machine learning
Introduction
In recent decades, the earth pressure balance machine (EPB-TBM) has emerged as a
highly efficient mechanized tunneling technique for constructing underground projects
in urban areas. This method offers a high excavation rate and ensures extreme safety for
urban transportation [2, 13, 15, 16].
The EPB’s operational parameters and excavation safety are influenced by the geotech-
nical characteristics of the soil surrounding the machine during excavation. Therefore, a
dependable assessment of geotechnical parameters is crucial for achieving optimal EPB
performance and meeting safety standards in urban settings [11, 14, 17, 23].
Sapigni et al. [20] examined the empirical relationship between the rock mass rating
(RMR) and the cutter penetration rate in hard rock tunnel boring machines (TBMs).
Yagiz [24] conducted a statistical analysis to propose an empirical method for predicting
penetration rates based on rock mass parameters. Hassanpour et al. [5] investigated the
link between rock mass properties and TBM performance using data from various tun-
neling projects in Iran. Building upon previous research, Hassanpour et al. (2011, 2021)
developed analytical models to forecast the Field Penetration Index (FPI) based on geo-
logical parameters like Uniaxial Compressive Strength (UCS), Rock Quality Designation
(RQD), and joint spacing. Liu et al. [10] introduced the Support Vector Regression (SVR)
model to predict rock mass parameters, including UCS, Brittleness Index (BI), Distance
between Planes of Weakness (DPW), and orientation of discontinuities (α), aligning
with data from the tunnel boring machine operations in the Songhua River water supply
project.
For decades, soft computing techniques, especially artificial intelligence methods,
have been extensively used to show the relationship between rock mass properties and
TBM performance parameters [9, 18, 25]. Additionally, the deep learning algorithm has
been employed to predict EPB-TBM operational and performance parameters [4, 26].
Nevertheless, there has been no effort to predict geotechnical parameters of soil materi-
als based on EPB operational data through reverse analysis using artificial intelligence
techniques.
This study examines the potential of predicting geotechnical parameters based on EPB
operational data using statistical and soft computing methods. Support vector regres-
sion (SVR) and feed-forward neural network (FNN) were applied using input samples
collected by a field data logger.
To enhance the prediction reliability and develop a more accurate estimation model,
EPB operational parameters were chosen as input variables, while two geological
strength index parameters (internal friction angle and shear strength) were set as pre-
diction targets. The operational parameters, including cutterhead torque, thrust force,
and cutterhead rotation speed (RPM), provide crucial information about geological
conditions at the face, and the support pressure value is closely linked to geotechni-
cal characteristics. Mechanical properties of the machine, such as cutterhead diam-
eter, cutter arrangement, and characteristics of mixing bars and backup system, have
no impact on the supervised and unsupervised model. Thus, operating parameters
(torque, thrust, RPM, and chamber pressure) were selected as input variables. The
friction angle (φ) and shear strength (τ) are highlighted as key factors influencing EPB
performance, closely tied to operational parameters. Based on experimental observa-
tions, reducing operating parameters like torque and thrust is directly influenced by a
decrease in the soil’s friction angle and shear strength.
Site description
Data sets from two sections of Tehran metro lines 6 and 7, containing different engi-
neering geological units, were gathered to develop and validate the suggested pre-
dictive model. The location of the projects is illustrated in Fig. 1. Additionally, Fig. 2
displays the cutterhead of the TBM related to the reference projects.
The key details of the two EPB tunneling projects are outlined in Table 1. Addi-
tionally, Table 2 presents the geological characteristics and soil properties of the geo-
logical engineering units along the tunnel path. According to the boreholes logs, the
terrain surrounding these tunnels primarily comprises sand with clay, clayey silt, and
silty soil. Figure 3 depicts the soil stratifications along the tunnels’ alignment based on
the findings from the eight exploratory boreholes.
Fig. 1 A geographical situation of the projects in the current study, a Tehran metro line 6 19], b Tehran metro
line 7
Fig. 2 a DSU.EPB.0916.127-TBM employed in Tehran metro line 7, b Herrenknecht S-523 EPB employed in
Tehran metro line 6
Table 1 General characteristics of tunneling projects

No Project Tunnel Available TBM type and manufacturer TBM Opening ratio
length data (km) diameter
(km) (m)
1 Line 6 metro 6.8 5 EPB (Herrenknecht) 9.19 0.35

tunnel, southern
extension sector,
(Iran)
2 Line 7 metro tun- 14 10 EPB (Lovat) 9.16 0.28
nel, east–west lot,
(Iran)
Table 2 Summary of geological features and soil properties of ET units along the tunnels
Engineering ET-1 ET-2 ET-3 ET-4 ET-5
geological
units
Soil type Sandy gravel & Very gravely Very silty clayey Clayey silty Clayey silt & silty
gravely sand sand with silt sand with gravel, sand with clay with sand, very
& clay very sandy clay (or gravel sandy clay (or silt)
silt) with gravel
USCS GW, GW-GM, SC, SC-SM & GC SC, SM & CL SC, SM CL, ML & CL-ML
GP-GC, SW & SP (rarely CH)
Results and discussion

Database development
To develop the study’s database, geotechnical soil parameters and EPB operational
data are gathered. Input variables consist of RPM, thrust force, cutterhead torque, and
chamber pressure. Notably, fixed machine properties like cutter spacing, diameter, cut-
ter number, and tip width were excluded from the analysis. Based on the collected and
developed database, this study focuses on predicting shear strength (τ) and friction angle
(φ) of soil materials along the tunnelling projects.
After completing data collection in the TML-SE6 and TML-EW7 projects, a database
was established to investigate and analyze the geotechnical parameters. It includes EPB
operating parameters and the engineering geological properties of clayey and sandy
soils (18 data points at different chainage points along the tunnel). Detailed borehole
logging results were obtained from the Tehran metro line 6 southern extension sector
Fig. 3 Changes in the values of soil parameters in the tunnel routes [19]
(TML-SE6) and the East–west section of line 7 Tehran metro project (TML-EW7). 85%
of the data was used for training, with the remaining 15% reserved for model testing.
This approach allows for a comprehensive analysis of potential correlations and depend-
encies among the factors involved. The dataset’s size and diversity ensure the study has
adequate statistical power to draw meaningful and reliable conclusions about the rela-
tionships being investigated. Therefore, the number of defined zones and collected data
in this study is sufficient to explore the relationships between input parameters and geo-
technical parameters in these two projects.
The collected and generated dataset is shown in Table 3. It is important to note that all
chosen data includes dependable geotechnical and machine performance data. Hence,
the fundamental descriptive statistical distribution of variables in the database is out-
lined in Table 4, encompassing all selected input variables and prediction targets.
Figure 4 shows the scatterplot matrix depicting the connections between shear
strength (τ) and friction angle (φ) parameters in the dataset, along with their asso-
ciations with the independent variables representing EPB’s operational parameters. To
delve deeper into the influence of individual factors (EPB operational parameters) on
geotechnical aspects, Fig. 5 presents the Pearson correlation outcomes for internal fric-
tion angle and shear strength. These visuals provide an in-depth analysis of the relation-
ships among the designated parameters.
Developing new empirical equations

Principal component analysis (PCA) is a multivariable statistical method that calcu-
lates important components, commonly employed to emphasize variation and decrease
dimensionality [3, 8].
In the present study, PCA was conducted on output and input parameters to
determine the variance ratio related to the relationship between EPB parameters
Table 3 The developed database generated by EPB operational parameters and analysis of
geotechnical field investigation
Chain. (m) Borehole USCS Phi Tau (kN/m2) Tq (kN m) Th (kN) RPM S (bar) H (m)
497 BH-1 CH 21.2 110.4 2100 8700 1.1 0.8 15 TML-SE6

680 BH-2 CL 18.3 122.4 5100 18,900 1.4 1.1 16
894 BH-3a SC 35.1 217.4 2100 17,400 1.4 1.2 18
2098 BH-7 SC 23 173.3 4900 26,000 1.3 2.4 20
2745 BH-7a CL 20.2 110.8 3500 22,900 1.5 1.9 15
2628 BH-7b CL 18.8 170.7 3500 22,900 1.5 1.9 15
2277 BH-7c SC 32 152.4 3000 25,100 1.5 1.8 13
3015 BH-8a CL 18 174.4 1900 27,080 1.6 2.1 20
2013 BH2-7 GC 28.7 105.8 6900 23,000 1.1 0.4 9 TML-EW7
2948 BH3-7 GC 31.7 177.3 5900 2800 0.9 0.7 14
4607 BH4-7 GM 36.3 196.7 7500 3300 1 0.02 15
7504 BH7-7 SM 34.9 277.3 10,060 1300 0.8 0.1 21
6482 BH6-7 SC 31 292.9 10,300 3100 0.9 0.4 25
7352 BH7-7–2 SC 32.8 272.01 10,700 2700 1 0.2 20
9371 BH9-7 SC 29.3 195.8 7200 1300 1.02 0.1 20
10,061 BH10-7 SC 26.7 186.7 4300 1500 1.08 0.1 21
5426 TP3-7 SC 28.4 149.5 3600 3010 1 0 14
7503 TP5-7 SC 29.3 201.1 1400 1300 0.9 0.1 21
Table 4 Summary of basic information of the model parameters

Parameters Categ Min Max Std. D Var Ave Med Skew Kurt
Torque (kN m) Input 1400 10,700 2915.9 850,028 5241.0 4600 0.64 − 0.71
Thrust (kN) Input 1300 27,080 10,108.2 1.02E + 08 10,694.5 3249 0.56 − 1.64
Friction angle (degree) Output 18 36.3 5.96 35.63 27.53 29 − 0.34 − 1.25
Shear strength (kN/m2) Output 105.8 292.91 54.4 2959.6 182.65 175.8 0.55 − 0.21
RPM Input 0.8 1.6 0.24 0.062 1.166 1.09 0.37 − 1.38
Earth pressure (bar) Input 0 2.42 0.814 0.662 0.885 0.595 0.60 − 1.21
Fig. 4 The scatterplot matrix and curve of considered parameters in the estimation of a internal friction
angle, b shear strength
Fig. 5 The Pearson correlation matrices between each single factor, a internal friction angle, b shear strength
Table 5 Results of principal components analysis (PCA) for internal friction angle
KMO and Bartlett’s Test
Kaiser–Meyer–Olkin Measure of Sampling Adequacy 0.727

Bartlett’s Test of Sphericity Approx. Chi-Square 86.53
df 10
Sig 0.000
Anti-image matrices
Torque Thrust RPM Ch. P φ
Anti-image Covariance Torque 0.648 0.000 0.043 − 0.005 − 0.132

Thrust 0.000 0.028 − 0.038 − 0.034 − 0.025
RPM 0.043 − 0.038 0.088 0.029 0.049
Ch. P − 0.005 − 0.034 0.029 0.055 0.049
φ − 0.132 − 0.025 0.049 0.049 0.546
Anti-image Correlation Torque 0.929 0.002 0.179 − 0.027 − 0.221
Thrust 0.002 0.638 − 0.766 − 0.871 − 0.203
RPM 0.179 − 0.766 0.733 0.423 0.223
Ch. P − 0.027 − 0.871 0.423 0.694 0.287
φ − 0.221 − 0.203 0.223 0.287 0.858
Total variance explained
Component Initial Eigenvalues Extraction sums of squared loadings
Total % of variance Cumulative % Total % of Variance Cumulative %
1 3.717 74.348 74.348 3.717 74.348 74.348

2 0.648 12.970 87.317
3 0.493 9.859 97.177
4 0.124 2.475 99.652
5 0.017 0.348 100.000
influencing geotechnical properties. The PCA results for geotechnical parameters can
be found in Tables 5 and 6 (A: friction angle, B: shear strength). The variance values,
correlation matrix, and covariance for both φ and τ are robust, indicating promis-
ing results for the development of new models using statistical and computational
methods.
Table 6 Results of principal components analysis (PCA) for shear strength

KMO and Bartlett’s Test
Kaiser–Meyer–Olkin Measure of Sampling Adequacy 0.622

Bartlett’s Test of Sphericity Approx. Chi-Square 81.01
df 10
Sig 0.000
Anti-image Matrices
Torque Thrust RPM τ Ch. P
Anti-image Covariance Torque 0.674 0.001 0.043 0.079 0.000

Thrust 0.001 0.024 − 0.034 0.058 − 0.031
RPM 0.043 − 0.034 0.081 − 0.091 0.030
τ 0.079 0.058 − 0.091 0.798 − 0.074
Ch. P 0.000 − 0.031 0.030 − 0.074 0.052
Anti-image Correlation Torque 0.950 0.006 0.184 0.108 − 0.003
Thrust 0.006 0.568 − 0.791 0.422 − 0.885
RPM 0.184 − 0.791 0.664 − 0.358 0.464
τ 0.108 0.422 − 0.358 0.101 − 0.362
Ch. P − 0.003 − 0.885 0.464 − 0.362 0.632
Total variance explained
Component Initial Eigenvalues Extraction Sums of Squared Loadings
Total % of Variance Cumulative % Total % of Variance Cumulative %
1 3.241 64.823 64.823 3.241 64.823 64.823

2 1.014 20.276 85.099 1.014 20.276 85.099
3 0.606 12.121 97.220
4 0.124 2.476 99.696
5 0.015 0.304 100.000
Fig. 6 Normal P–P plot of regression standardized residual, a shear strength, b friction angle
The P-P plot compares data distribution with various theoretical models, utilizing
the empirical cumulative distribution functions of normal, Laplace, and uniform dis-
tributions. A model that fits the data well should roughly align with the y = x line [12,
21]. The P–P plot results for shear strength and friction angle are depicted in Fig. 6.
Fig. 7 3D diagrams based on two main EPB operating parameters and geotechnical features, a friction angle,
b shear strength
Table 7 The developed models to estimate geotechnical parameters using actual EPB-TBM data
Inputs Out Relationship R Reg. type
Th, Tq, Ch.P, RPM φ 41.56 + 0.416*Tq +c0.534*Th– 0.67 Forward stepwise
13.68*RPM-6.63*Ch.P
Th, Tq, Ch.P, RPM τ 325.117 + 9.065*Tq + 7.684*Th– 0.68 Forward stepwise
195.372* RPM–49.964*Ch.P
It is evident that the outcomes of the P–P diagram, derived from determined factors,
closely adhere to the x = y line, indicating reliable results. The 3D representations of
the three primary parameters for each output geomechanical parameters (φ and τ) are
illustrated in Fig. 7. These visualizations demonstrate the correlation between inputs
and target values in a 3D perspective.
In this research, multivariable regression was utilized to establish an empirical equa-
tion linking geotechnical parameters with operational parameters of EPB-TBM. The
empirical formula is crucial for tunneling, derived from real-time tunnel data. Machine
specifications (thrust force, torque, RPM) and chamber pressure were considered as
independent variables, while geotechnical parameters were the dependent variables.
Linear regression analysis assessed the impact of each input variable on geological
strength index parameters.
Several analyses were conducted using input parameters from the database to deter-
mine the key factors influencing geotechnical characteristics. New equations for friction
angle and shear strength of soil materials were proposed, as detailed in Table 7. Table 8
displays the ANOVA analysis and regression coefficients for these equations, which were
found to be valid and reliable. The t-test and F-test results confirmed the accuracy of
the correlations and coefficients. The outcomes of the MVR analysis, utilizing empirical
equations for geological strength index parameters, are illustrated in Figs. 8 and 9.
Sensitivity analysis is a method to determine the impact of input uncertainty on tar-
get variables related to independent input variables. This method focuses on quantifying
uncertainty and evaluating power. The results of sensitivity analysis are often displayed
in a tornado diagram, showing the range of output parameter values based on changes in
input parameters.
Table 8 (a) Variables of the generated models for stepwise regression analysis; (b) Significance of
r-value; (c) Analysis of variance for the significance of regression based on the law of total variance
Internal friction angle
(a) Model summary

R R2 Adj. R2 Std. error
0.67 0.45 0.42 4.67

(b) Coefficients
Unstandardized coefficients Standardized coefficients t Sig.
B Std. error Beta
(Constant) − 0.03 7.63 − 0.005 0.9

Formula 1.00 0.27 0.67 3.65 0.00
(c) Anova model
Sum of squ. df Mean square F Sig.
Regression 291.44 1 291.441 13.327 0.00

Residual 349.90 16 21.869
Total 641.34 17
Shear strength
(a) Model summary R R2 Adj. R2 Std. error
0.68 0.47 0.44 27.6

(b) Coefficients Unstandardized coefficients Standardized coefficients t Sig.
B Std. error Beta
(Constant) 98.6 22.8 0.001 0.9

formula 0.45 0.12 0.68 3.58 0.00
(c) Anova Sum of squ. df Mean square F Sig.
model
Regression 10976.0 1 10976.0 14.3 0.00

Residual 12210.0 16 763.1
Total 23186.1 17
In this study, recalculating outputs under alternative assumptions to assess the

impact of EPB operating parameters in sensitivity analysis can help identify the most
influential variables on geological strength index parameters of soil materials.
Changes in friction angle due to variations in input parameters are shown in
Fig. 10a. The tornado diagram indicates that thrust force and cutterhead torque have
a minor impact on the friction angle of soil materials, while RPM and chamber pres-
sure are the primary parameters affecting friction angle.
A similar analysis was conducted to determine the most influential parameters
on shear strength, as shown in Fig. 10b. The tornado diagram reveals that cutter-
head torque has a minimal effect on the geotechnical parameters of soil materials,
whereas RPM and chamber pressure are the critical parameters.
Fig. 8 Comparison of the measured and calculated geotechnical parameters using MVR, a shear strength, b
friction angle
Fig. 9 Correlation between measured and calculated geotechnical parameters using MVR, a shear strength,
b friction angle
Fig. 10 Tornado graph resulted of sensitivity analysis of MVR model, a friction angle, b shear strength
Supervised learning algorithms

Support vector regression (SVR) is a widely used supervised learning network for
estimating continuous variables. SVR follows the same principle as the Support
Vector Machine (SVM) algorithm. In this computational model, the hyperplane that
maximizes the number of marks serves as the best fit line [22].
A feed-forward neural network (FFNN) is a type of neural network that operates in
a single forward direction without loops in its structure. Starting from input nodes,
the network progresses through hidden layers to predict output factors [1].
A comprehensive dataset of borehole logging results along the tunnel path was
gathered, with 85% of the data randomly selected for training and the rest reserved
for testing the model. Figures 11 and 12 display the measured data alongside the cor-
responding predicted values. The results from SVR and FFNN demonstrate satisfac-
tory performance in achieving higher accuracy in regression, affirming the efficiency
and reliability of the models. The proposed soft computing technique proves feasible
and reliable in predicting geological strength index parameters using operational data
from tunnel boring machines.
Fig. 11 Comparison of the measured and predicted geotechnical parameters using FNN and SVR networks,
a friction angle, b shear strength
Fig. 12 Correlation between measured and predicted geotechnical parameters using FNN and SVR
networks, a friction angle, b shear strength
Results obtained from various loss functions during training and validation stages for
the two applied smart networks, focusing on shear strength and internal friction angle,
are presented in Tables 9 and 10, respectively. These results highlight the strong cor-
relation between measured geological strength index parameters and the accurate pre-
diction of these factors using machine learning algorithms. Additionally, the function
results related to the five sets in the validation stage are depicted in Fig. 13.
A comparison between machine learning and MVR methods with measured data from
the test dataset is illustrated in Fig. 14. The small deviation intervals of the predicted
values indicate that the predictions from FFNN, SVR, and MVR align well with the
measured values of geotechnical parameters. Essentially, these models, particularly neu-
ral networks, can effectively predict geotechnical parameters at the tunnel face, making
them suitable for accurate predictions based on EPB operating data.
The internal friction angle (φ) values (Min: 18, Max: 36) are categorized into four
classes for the confusion matrix representation (Class I: 18–22, Class II: 22–26, Class
III: 26–30, Class IV: 30–36). The confusion matrix for each dataset scale is depicted
in Fig. 15. Each row in the matrix represents an actual value, while each column
corresponds to the predicted φ using the three applied methods (MVR, SVR, FNN;
respectively). Figure 15 illustrates that a significant portion of class I (φ: 18–22) was
incorrectly predicted as other classes, attributed to the limited representation of class
I in the dataset. Additionally, the confusion matrix for shear strength (τ) (Min: 100
Table 9 The results of loss functions in the development of ML-based algorithms for shear strength
Training subset
Loss functions FFNN SVR
Root mean square error (RMSE) 2.272 0.330

Mean square error (MSE) 0.050 5.9675
Mean absolute percentage error (MAPE) 0.047 0.068
Mean absolute deviation (MAD) 3.918 3.030
Relative root mean square error (RRMSE) 0.644 1.332
Normalized root mean square error (NRMSE) 0.008 0.001
Relative squared error (RSE) 0.361 0.776
Root relative squared error (RRSE) 0.601 0.881
Testing subset

Table 10 The results of loss functions in the development of ML-based algorithms for friction angle
Training subset

Testing subset

Fig. 13 Loss function results in reference to five sets in the validation stage, a friction angle, b shear strength
Fig. 14 Comparison of computing machinery and MVR methods with actual data based on the test dataset,
a friction angle, b shear strength
Fig. 15 Confusion matrices of φ, four classes, three methods: a MVR, b SVR, c FFNN
kN/m2, Max: 300 kN/m2) based on the outcomes of the three methods is shown in
Fig. 16, with four defined classes (Class I: 100–150, Class II: 150–200, Class III: 200–
250, Class IV: 250–300). It is evident that a considerable number of class I (τ: 100–
150) instances were misclassified as other classes due to their scarcity in the dataset.
Fig. 16 Confusion matrices of τ, four classes, three methods: a MVR, b SVR, c FFNN
Fig. 17 The assessment of changes in the values of geotechnical parameters (four classes, τ, φ) on EPB-TBM
performance
Figure 17 presents the variations in main EPB-TBM parameters (thrust force, cham-
ber pressure) across the four specified shear strength and internal friction angle
classes. The results reveal that increasing the shear strength range (200–300 kN/m2)
leads to higher thrust force and chamber pressure values. Similarly, expanding the
friction angle range (30–36) results in decreased values of EPB operating parameters,
which are deemed reliable for tunneling projects.
Key limitation
This study has limitations related to the database used, focusing on determining
geotechnical properties like friction angle (φ) and shear strength (τ) for engineer-
ing geology units ET1 to ET5 (fine-grained and coarse-grained soil) using EPB-TBM
operational and driving parameters. To enhance the comprehensiveness and applica-
bility of the machine learning (ML) models in this study, expanding the database is
crucial. The authors are actively working on improving the database by incorporating
data from additional laboratory tests, in-situ analysis, EPB data, and including a wider
range of conditions and tunneling projects. It is expected that developing ML and
deep learning (DL) models based on this expanded dataset will produce more effec-
tive results. Additionally, it is worth noting that the proposed models could be appli-
cable for similar ground and TBM tunneling conditions.
Conclusions
In this research, various methods have been introduced to forecast geotechnical proper-
ties of soft soil in urban areas based on EPB operational data. To create a new predic-
tive model, two geotechnical factors, friction angle, and soil shear strength, are analyzed
using EPB operational data such as RPM, thrust force, cutterhead torque, and chamber
pressure. Computational tools and statistical techniques were employed to construct
these prediction models.
Data from EPB operations in Tehran metro line 6 southern extension sector (TML-
SE6) and the east–west section of line 7 (TML-EW7) were gathered to form a dataset
encompassing diverse engineering geological units. 85% of the samples were randomly
chosen for training, with the remainder used for final testing through support vector
regression (SVR) and feed-forward neural network (FFNN) algorithms. The models
demonstrated effective prediction of abrupt geological unit changes during tunneling in
areas with varying geological characteristics. Validation was performed using statistical
metrics and loss functions, confirming the reliability and efficiency of the models. Addi-
tionally, empirical formulas were devised based on data from two tunneling projects in
different geological units to estimate soil geotechnical properties in urban settings.
The outcomes reveal a strong correlation between soil mechanics parameters and EPB
machine performance and operational factors. The R2 values in the SVR neural network
model are 0.96 for shear strength and 0.70 for friction angle, while in the FFNN model,
they are 0.91 for shear strength and 0.66 for friction angle. The MAD loss function val-
ues for the parameters φ and τ in the SVR network are 0.42 and 0.06, respectively. Hence,
utilizing EPB operational variables is advised for estimating soil mechanics parameters φ
and τ. The study suggests that computational techniques, particularly the SVR model,
offer more accurate estimations of target parameters compared to statistical methods.
Author contribution
HS: conceptualization, methodology, software, investigation, visualization, writing-original draft. JH: conceptualization,
supervision, validation, formal analysis, resources, writing-review and editing, project administration. JR: supervision,
validation, writing-review and editing, project administration.
Data availability
The data, models, and code from the study are available upon request from the corresponding author.
Declarations
Competing interests
The authors proclaim that they have no known contending financial interests or personal relationships that could have
appeared to influence the work reported in this paper.
Received: 13 April 2023 Accepted: 25 August 2024
References
1. Andreas, Z. 1994. Simulation neuronaler netze-simulation of neural networks. Addison-Wesley. p. 73, ISBN
3-89319-554-8.
2. Chen Z, Bezuijen A, Fang Y, Wang K, Deng R (2022) Experimental study and field validation on soil clogging of EPB
shields in completely decomposed granite. Tunn Undergr Space Technol 120:104300. https://doi.org/10.1016/j.tust.
2021.104300
3. Elhaik E (2022) Principal Component Analyses (PCA)-based findings in population genetic studies are highly biased
and must be reevaluated. Sci Rep 12:1. https://doi.org/10.1038/s41598-022-14395-4
4. Gao X, Shi M, Song X, Zhang C, Zhang H (2020) Recurrent neural networks for real-time prediction of TBM operating
parameters. Autom Constr 98:225–235. https://doi.org/10.1016/j.autcon.2018.11.013
5. Hassanpour J, Rostami J, Khamehchiyan M, Bruland A, Tavakoli HR (2010) TBM performance analysis in pyroclastic
rocks, a case history of Karaj Water Conveyance Tunnel (KWCT). Rock Mech Rock Eng 43:427–445. https://doi.org/10.
1007/s00603-009-0060-2
6. Hassanpour J, Rostami J, Zhao J (2011) A new hard rock TBM performance prediction model for project planning.
Tunnel Underground Space Technol 29:595–603. https://doi.org/10.1016/j.tust.2011.04.004
7. Hassanpour J, Firouzei Y, Hajipour G (2021) Actual performance analysis of a double shield TBM through sedimen-
tary and low to medium grade metamorphic rocks of Ghomrood water conveyance tunnel project (lots 3 and 4).
Bull Eng Geol Env 80:1419–1432. https://doi.org/10.1007/s10064-020-01947-z
8. Jolliffe LT, Cadima J (2016) Principal component analysis: a review and recent developments. Philos Trans Royal Soc
A Math Phys Eng Sci. https://doi.org/10.1098/rsta.2015.0202
9. Liu X, Shao C, Ma H, Liu R (2011) Optimal earth pressure balance control for shield tunneling based on LS-SVM and
PSO. Autom Constr 20:321–327. https://doi.org/10.1016/j.autcon.2010.11.002
10. Liu B, Wang R, Guan Z, Li J, Xu Z, Guo X, Wang Y (2019) Improved support vector regression models for predicting
rock mass parameters using tunnel boring machine operational data. Autom Constr. https://doi.org/10.1016/j.tust.
2019.04.014
11. Mahmoodzadeh A, Taghizadeh M, Mohammed AH, Ibrahim HH, Samadi H, Mohammadi M, Rashidi S (2022) Tunnel
wall convergence prediction using optimized LSTM deep neural network. Geomech Eng. 316:545–556. https://doi.
org/10.12989/gae.2022.31.6.545
12. Michael JR (1983) The stabilized probability plot. Biometrika 70(1):11–17
13. Oliveira DGG, Thewes M, Diederichs MS, Langmaack L (2018) EPB tunnelling through clay-sand mixed soils: pro-
posed methodology for clogging evaluation. Geomechanik und Tunnelbau 11(4):375–387. https://doi.org/10.1002/
geot.201800009
14. Oliveira DGG, Thewes M, Diederichs MS (2019) Clogging and flow assessment of cohesive soils for EPB tunnelling:
proposed laboratory tests for soil characterization. Tunn Undergr Space Technol 94:103–110. https://doi.org/10.
1016/j.tust.2019.103110
15. Peila D, Picchio A, Chieregato A (2013) Earth pressure balance tunnelling in rock masses: laboratory feasibility study
of the conditioning process. Tunn Undergr Space Technol 35:55–66. https://doi.org/10.1016/j.tust.2012.11.006
16. Samadi H, Hassanpour J, Farrokh E (2021) Maximum surface settlement prediction in EPB TBM tunneling using soft
computing techniques. J Phys Conf Ser 1973(1):012195. https://doi.org/10.1088/1742-6596/1973/1/012195
17. Samadi H, Hassanpour J, Tarigh Azali S. Developing GEP technique for prediction of EPB-TBM performance in lime-
stone strata. TBMDigs 2022. Austria, 2022:158–164.
18. Samadi H, Mahmoodzadeh A, Hussein Mohammed A, Alenizi FA, Hashim Ibrahim H, Nematollahi M, Babeker Elhag
A (2023) Application of several fuzzy-based techniques for estimating tunnel boring machine performance in meta-
morphic rocks. Rock Mech Rock Eng. https://doi.org/10.1007/s00603-023-03602-x
19. Samadi H, Hassanpour J, Rostami J (2023) Prediction of earth pressure balance for EPB-TBM using machine learning
algorithms. Int J Geo-Eng 14:21. https://doi.org/10.1186/s40703-023-00198-7
20. Sapigni M, Berti M, Behtaz E, Busillo A, Cardone G (2002) TBM performance estimation using rock mass classification.
Int J Rock Mech Min Sci 39:771–788. https://doi.org/10.1016/S1365-1609(02)00069-
21. Shorack GR, Wellner JA. Empirical processes with applications to statistics. Wiley. 1986. ISBN 0-471-86725-X, 248–250.
22. Smola AJ, Scholkopf B (2004) A tutorial on support vector regression. Stat Comput 14:199–222
23. Vinai R, Oggeri C, Peila D (2008) Soil conditioning of sand for EPB applications: a laboratory research. Tunn Undergr
Space Technol 23(3):308–317. https://doi.org/10.1016/j.tust.2007.04.010
24. Yagiz S (2008) Utilizing rock mass properties for predicting TBM performance in hard rock condition. Tunn Undergr
Space Technol 23(3):326–339. https://doi.org/10.1016/j.tust.2007.04.011
25. Yeh I (1997) Application of neural networks to automatic soil pressure balance control for shield tunneling. Autom
Constr 5(5):421–426. https://doi.org/10.1016/S0926-5805(96)00165-3
26. Zhou C, Ding L, Zhou Y, Zhang H, Skibniewski MJ (2019) Hybrid support vector machine optimization model for
prediction of energy consumption of cutter head drives in shield tunneling. J Comput Civil Eng. https://doi.org/10.
1061/(ASCE)CP.1943-5487.0000833
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Assessment of Shear Strength of Fine Grained and Coarse Grained Soil Using Actual EPB TBM Operating Data

Uploaded by

Copyright:

Available Formats

Assessment of Shear Strength of Fine Grained and Coarse Grained Soil Using Actual EPB TBM Operating Data

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Assessment of Shear Strength of Fine Grained and Coarse Grained Soil Using Actual EPB TBM Operating Data

Uploaded by

Copyright:

Available Formats

Samadi et al.

ORIGINAL RESEARCH Open Access

Assessment of shear strength of fine‑grained

Keywords: Soil strength parameters, EPB-TBM, Computing machinery, Sensitivity

Table 1 General characteristics of tunneling projects

1 Line 6 metro 6.8 5 EPB (Herrenknecht) 9.19 0.35

Results and discussion

Developing new empirical equations

497 BH-1 CH 21.2 110.4 2100 8700 1.1 0.8 15 TML-SE6

Table 4 Summary of basic information of the model parameters

Kaiser–Meyer–Olkin Measure of Sampling Adequacy 0.727

Anti-image Covariance Torque 0.648 0.000 0.043 − 0.005 − 0.132

1 3.717 74.348 74.348 3.717 74.348 74.348

Table 6 Results of principal components analysis (PCA) for shear strength

Kaiser–Meyer–Olkin Measure of Sampling Adequacy 0.622

Anti-image Covariance Torque 0.674 0.001 0.043 0.079 0.000

1 3.241 64.823 64.823 3.241 64.823 64.823

(a) Model summary

0.67 0.45 0.42 4.67

Unstandardized coefficients Standardized coefficients t Sig.

B Std. error Beta

(Constant) − 0.03 7.63 − 0.005 0.9

Regression 291.44 1 291.441 13.327 0.00

0.68 0.47 0.44 27.6

(Constant) 98.6 22.8 0.001 0.9

Regression 10976.0 1 10976.0 14.3 0.00

In this study, recalculating outputs under alternative assumptions to assess the

Supervised learning algorithms

Root mean square error (RMSE) 2.272 0.330

Root mean square error (RMSE) 11.080 23.327

Root mean square error (RMSE) 2.402 2.125

Root mean square error (RMSE) 0.083 0.491

Received: 13 April 2023 Accepted: 25 August 2024

You might also like