Professional Documents
Culture Documents
Abstract: A safety monitoring system is usually applied in deep excavations in order to control the construction risk and to ensure the
Downloaded from ascelibrary.org by George Washington University on 04/04/16. Copyright ASCE. For personal use only; all rights reserved.
serviceability of adjacent facilities. Considering the mass data collected by different sensors, a reasonable assessment method on the mon-
itoring results is necessary to evaluate the safety state of both the deep excavation itself and the surrounding environment. By introducing the
conception of data fusion, a comprehensive assessment method is presented to find the anomaly in the safety monitoring results in this paper.
Data fusion analyses on both a single monitoring item and the correlation of multiple monitoring items are proposed and studied. The one-
class support vector machines (SVMs) are used to improve the data fusion analysis between a single monitoring item and different excavation
parameters, and then developed to three-dimensional (3D) fusion analysis on a single item and multiple parameters of an excavation. The
mechanical and geometric patterns between different monitoring items are studied to propose a data fusion analysis on multiple monitoring
items and then to build the assessment criteria. Based on these two kinds of data fusion analysis, the mass monitoring data can be analyzed
completely to assess the safety state of deep excavations. An application in two cases of deep excavation in Shanghai, China, shows that
the proposed method is effective in data anomaly assessment. DOI: 10.1061/(ASCE)AS.1943-5525.0000593. © 2015 American Society of
Civil Engineers.
Introduction adjacent facilities (Mana and Clough 1981; Clough and O’Rourke
1990; Woo and Moh 1990; Ou et al. 1993; Wong et al. 1997). How-
Field monitoring is an important means to ensure the safety state of ever, these criteria are mostly represented through linear relation-
deep excavations and their surrounding structures (Bhalla et al. ship and are single-parameter related. Such traditional monitoring
2005; Liu et al. 2005, 2006; Su et al. 2006; Tan and Wei 2011; criteria just describe local features of the deep excavation safety
Zhang et al. 2013; Chen et al. 2015). During deep excavation, re- state and are vulnerable to the external environment. Additionally,
taining structures of deep excavations and the surrounding environ- they do not consider deep excavation as a whole system and do not
ment is a whole system. Therefore, in the design and construction make full use of the potential value of mass monitoring data. These
of a braced excavation, it is essential to consider not only the sta- have led to false positive and false negative in safety monitoring
bility and safety issue but also the potential serviceability problem and cannot play an effective role in risk control to achieve the
of adjacent buildings and pipes caused by excessive wall deflec- purpose of monitoring (Liu and Wang 2009). For such reason, a
tions and ground movements (Xu 2007; Wang et al. 2010; Zhang 33-m-deep excavation adjacent to Nicoll Highway collapsed in
et al. 2013, 2015; Li et al. 2014; Wang et al. 2015). The typical Singapore with continuous alarms, which masked real potential
response of the whole system during deep excavation is shown dangers (Magnus et al. 2005b).
as Fig. 1 (Wang et al. 2010; Zhang et al. 2015). All of these differ- There are two aspects to solving false positive and false negative
ent responses have to be monitored as different items, including
problems according to present researches. One is to introduce new
displacement of wall, ground settlement, axial force of strut, and
monitoring techniques (Bhalla et al. 2005; Liu et al. 2006; Su et al.
vertical displacement of column and wall (Liu et al. 2005; Tan
2006). Liu et al. (2006) studied the feasibility of applying Brillouin
and Wei 2011).
optical time domain reflectometry (BOTDR) to deep excavation
After the field monitoring data are obtained, the safety state of
monitoring. Su et al. (2006) explored the use of three-dimensional
the deep excavation and surrounding environment, both soil and
(3D) laser scanning technology to accurately capture construction
adjacent underground structure, should be assessed to control
activities during excavation. The other aspect is to develop new
the construction risk. Many monitoring criteria for risk warning
assessment methods based on analysis of monitoring data such
have been proposed based on performances of excavations and
as statistics, nonlinear regression, and data fusion. There is a need
1
Associate Professor, Dept. of Civil Engineering, Shanghai Jiao
to integrate information from the various instruments and to relate
Tong Univ., 800 Dongchuan Rd., Shanghai 200240, China. E-mail: the crucial information to what is happening on the worksite as well
chenjj29@sjtu.edu.cn as the quality of each of the elements in the construction (Magnus
2 et al. 2005a). Similarly, the conception of data fusion is the process
Graduate Student, Dept. of Civil Engineering, Shanghai Jiao Tong Univ.,
800 Dongchuan Rd., Shanghai 200240, China. E-mail: zhangwei2609@ of integration of multiple data points and knowledge representing
gmail.com the same real-world object based on the optimization criteria or
3
Professor, Dept. of Civil Engineering, Shanghai Jiao Tong Univ., 800 algorithms into a consistent, accurate, and useful representation
Dongchuan Rd., Shanghai 200240, China (corresponding author). E-mail: (Hall and Llinas 1997), and it has been applied in different fields.
wjh417@sjtu.edu.cn
Based on the conception of data fusion, a comprehensive
Note. This manuscript was submitted on May 10, 2015; approved on
October 9, 2015; published online on December 31, 2015. Discussion anomaly assessment method is presented in this paper through data
period open until May 31, 2016; separate discussions must be submitted fusion analysis of a single monitoring item and multiple monitoring
for individual papers. This paper is part of the Journal of Aerospace En- items separately. Application in two case studies indicates the
gineering, © ASCE, ISSN 0893-1321. effectiveness of this method.
Fig. 5. Boundary of δ hm and ρs calculated by one-class SVM Fig. 7. Boundary of δ hm , H, and ρs calculated by one-class SVM
and most of the data are less than 0.2%H. Xu (2007) used FEM to fell outside region of existing similar engineering, it is worth ana-
obtain the upper limit of the relationship between δ vm and H. lyzing the reason why anomaly data appeared. The most likely rea-
Training data are from the top-down method of deep excavation son is the lack of similar engineering data, but that may not meet
used in concrete diaphragm walls in Shanghai, China (Xu 2007). some implied mechanical laws in region. Region representation
Fig. 6 shows the boundary of relationship between δ vm and H among multiple parameters calculated by the one-class SVM algo-
calculated by one-class SVM. rithm describes the region more accurately so it can solve the false
positive problem in deep excavation monitoring. The region can be
updated to obtain better performance by presenting new training
3D Fusion Analysis by One-Class SVM examples as new data become available.
Monitoring items are not only affected by a single excavation
parameter, but also related by other parameters. Generalization
to region representation of the relationship between a single Data Fusion Analysis of Multiple Monitoring Items
monitoring item and multiple parameters makes the region more
precise. And another advantage of using the one-class SVM algo- During excavation, there are mechanical and geometric patterns be-
rithm is that it is easy to implement generalization to multiple tween various monitoring items because the retaining structure and
dimensions. Combined with geometric parameter and mechanical surrounding environment work as a whole system. Soil unloading
parameter, the proposed region relationship among δ hm , H, and ρs causes bottom heave and displacement of retaining walls, thereby
is shown in Fig. 7. ground movements behind walls occur, which cause settlements of
pipelines and buildings. Among the monitoring items, the relation-
ship between lateral displacement of retaining wall and settlement
Discussions ground surface describes the soil flow inside and outside of the pit,
Data fusion analysis of monitoring item and excavation parameters indirectly reflecting the settlement and bottom heave of pit and the
is an experience analogy method. If data of a new deep excavation settlement pipelines and buildings; the relationship between verti-
cal displacements of supporting columns and retaining walls con-
tributes an additional factor on the stability of struts; the amount of
compression of struts and lateral displacement of the retaining wall
must be coordinated geometrically, otherwise there is a risk of
buckling failure of struts or punching failure of walls.
Similarly, there is a relationship between maximum rate of the traditional method, there is hidden risk because limit-value dis-
change of settlement Δδ vm and maximum rate of change of lateral placement still gives rise to additional displacement for struts.
displacement Δδ hm . According to relevant research (Xu 2007), if
Δδ vm is bigger than Δδ hm or Δδ hm is small and Δδ vm is large, a
mismatch is considered to have happened. Axial Force of Strut and Lateral Displacement of
Without outside soil flowing and groundwater seepage, the vol- Retaining Wall
ume of soil loss behind retaining walls caused by displacement of Generally, risk control for struts uses the value of axial force as the
walls is equal to the sum of the volume of ground surface settlement only assessment criteria, such as 80% (Liu and Wang 2009) of de-
trough and volume of bottom heave caused by unloading. Statis- sign value of axial force. Statistical results show that the force of
tical analysis in Shanghai, China, shows that the relationship be- struts is rarely beyond the design value, while local buckling of
tween volume of soil loss trough behind retaining walls and struts or punching failure to retaining wall brings great hidden risk
volume of ground surface settlement trough is V s ¼ 0.76V w (Xu to deep excavation engineering. From the design of the supporting
2007). This coefficient is affected by factors including dewatering system, struts deformation and lateral displacement of retaining
outside the pit, nonignorable lateral deformation at the toe of walls should be coordinated geometrically. If that is a mismatch,
retaining walls, overload on ground surface, and abnormal bottom the force of struts will be small and lateral displacement of the re-
heave. taining wall will be too large.
According to material mechanics, Eq. (9) can be derived 4. Collect references for determining coefficient between multiple
monitoring items by calculation, literature, or statistics;
FN L
¼ Kδ ð9Þ 5. Obtain assessment criteria on multiple monitoring items,
Es A e.g., δ vm ¼ 0.5 − 1.5δ hm and K ¼ 0.5 − 1.5 in “Data Fusion
Analysis of Multiple Monitoring Items” can be used as assess-
where FN = axial force of strut; L = length of strut; Es = elastic
ment criteria;
modulus of strut material; A = cross-sectional area of strut; δ =
6. Assess safety monitoring results through these two kinds of
compression deformation calculated by lateral displacement of
assessment criteria; and
retaining wall; and K = coefficient between strut axial force and
7. Obtain anomaly data after assessment and possible reasons for
lateral displacement calculation value.
these data.
In the assessment relationship, it is not necessary to draw the
Data Fusion Analysis on Results of Inclinometer region in practice. Once w and b in Eq. (9) are discovered, a model
There is a mechanical relation for the same kind of monitoring is obtained based on one-class SVM. Then, w and b are substituted
item, so it can also apply data fusion analysis. An inclinometer in into Eq. (1) to obtain the value of yðiÞ. Concretely, in deep exca-
a deep excavation can obtain lateral displacement of the wall at a vation the features of training data include δ hm , H, and ρs or others.
different depth. Then the internal force of the retaining wall is indi- Then calculate the monitoring data with the model, which is learnt
rectly felt through these lateral displacements such as using curva- from training data. The value of y determines this engineering is in
ture of displacement curve to calculate bending moment of retaining a safety state or anomaly state.
wall. Consequently, the assessment criterion can be obtained by
selecting the internal force control parameters. There are some Application in Case 1 of Deep Excavation
methods to calculate internal force through lateral displacement.
Project Description
Discussions The depth of excavation of the Hanzhong Road Station in Shanghai
Metro Line No. 13 is 31.1 to 33.1 m. The thickness of the dia-
Essentially, relationships between monitoring items mean geomet- phragm walls is 1,200 mm, and the depth is 58 m in the standard
ric or physical patterns in deep excavation. If new monitoring data segment of station and 62 m in the work shafts of the shield tunnel.
violate these geometric and physical patterns, they are considered There are eight levels of struts along depth of pit. The fourth and
as anomaly data. This is a situation in which the traditional mon-
itoring method is not considered because it uses a single monitoring
item. What is more, it is more obvious in physical meaning than the
traditional method, so it can solve false positive and false negative
problems in deep excavation monitoring.
Fig. 13. Ratio of δ hm and δ vm in excavation Fig. 15. Coefficient K at point ZC18-2
Conclusions
In order to realize more effective use of the mass monitoring data
and control the construction risks, a data fusion analysis method is
proposed to assess the safety monitoring results of deep excava-
tions. Data fusion analysis on both a single monitoring item and
correlation of multiple monitoring items are studied. The following
conclusions and suggestions can be obtained:
• In data fusion analysis on a single monitoring item, assessment
by region is an experience analogy method. If new deep exca-
vation data fall outside the region of existing similar engineer-
ing, the reason for the anomaly data should be analyzed.
• In data fusion analysis on multiple monitoring items, relation-
ships between monitoring items essentially mean geometric or
physical patterns in deep excavation. If new monitoring data
violate these patterns, they should be considered as anom-
aly data.
Fig. 14. Axial force changes at point ZC18-2 • An application of deep excavation shows the performance of the
anomaly assessment method, which solves the false positive and
false negative problem in deep excavation monitoring.
The anomaly assessment method based on data fusion analysis
Application in Case 2 of Deep Excavation can be updated to obtain better results by presenting new train-
ing data.
Project Description
Case 2 is an excavation of Kaixuan Road Station in Shanghai Metro
Line No. 10. This deep excavation is 150.25 m long, 24 m wide, Acknowledgments
and 15.44 m deep. The depth of diaphragm walls is 33 m in the The financial support from the National Natural Science Founda-
standard segment of station and 29 m in the work shafts of shield tion of China (NSFC) (Grant Nos. 41172251 and 41330633) and
tunnel. There are five levels of struts along depth of the pit. The first the Science and Technology Commission of Shanghai Municipality
levels of struts are concrete struts and the others are steel struts. (Funding No. 14231200702) are gratefully acknowledged.
Data Fusion Analysis of Multiple Monitoring Items
The steel strut monitored by the ZC18-2 sensor is 18.6 m long and Notation
located at −4.5 m. The cross-section shape of this strut is a pipe
whose external diameter is 609 mm and thickness is 16 mm. Ac- The following symbols are used in this paper:
cording to the traditional method, the limit value of axial force is A = cross-sectional area of strut;
1,600 kN. The axial force of this strut is always under the limit b = parameter of hyperplane;
value shown in Fig. 14. Calculated by Eq. (9), the coefficient be- Es = elastic modulus of strut material;
tween strut axial force and lateral displacement calculation value Ew = elastic modulus of retaining wall;
can be obtained. Fig. 15 shows the coefficient during excavating. FN = axial force of strut;
However, it can been seen that on December 16, K equals 0.41, f = hyperplane;
which less than the criteria K ¼ 0.5 − 2, while nothing happened H = excavation depth;
in the traditional method, so it belongs to false negative data. h = average spacing of struts in vertical direction;
Ou, C., Hsieh, P., and Chiou, D. (1993). “Characteristics of ground surface
γ = minimum of γ ðiÞ ; settlement during excavation.” Can. Geotech. J., 30(5), 758–767.
γ = distance from xðiÞ to the hyperplane;
ðiÞ Schölkopf, B., Platt, J. C., Shawe-Taylor, J., Smola, A. J., and Williamson,
γ w = unit weight of water; R. C. (2001). “Estimating the support of a high-dimensional distribu-
Δδ hm = maximum rate of change of lateral displacement; tion.” Neural Comput., 13(7), 1443–1471.
Δδ vm = maximum rate of change of settlement; Schölkopf, B., Smola, A. J., Williamson, R. C., and Bartlett, P. L. (2000).
δ = compression deformation calculated by lateral “New support vector algorithms.” Neural Comput., 12(5), 1207–1245.
displacement of retaining wall; Su, Y. Y., Hashash, Y., and Liu, L. Y. (2006). “Integration of construction
δ hm = maximum lateral displacement of wall; as-built data via laser scanning with geotechnical monitoring of urban
excavation.” J. Constr. Eng. Manage., 10.1061/(ASCE)0733-9364
δ vm = maximum lateral displacement of wall; and
(2006)132:12(1234), 1234–1241.
ρs = system stiffness of deep excavation.
Tan, Y., and Wei, B. (2011). “Observed behaviors of a long and deep
excavation constructed by cut-and-cover technique in Shanghai soft
clay.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)GT.1943-5606
References .0000553, 69–88.
Vapnik, V. (2000). The nature of statistical learning theory, Springer,
Bhalla, S., Yang, Y. W., Zhao, J., and Soh, C. K. (2005). “Structural health New York.
monitoring of underground facilities—Technological issues and chal- Wang, J. H., Xu, Z. H., and Wang, W. D. (2010). “Wall and ground
lenges.” Tunnell. Underground Space Technol., 20(5), 487–500. movements due to deep excavations in Shanghai soft soils.” J. Geotech.
Bishop, C. M. (2006). Pattern recognition and machine learning, Springer, Geoenviron. Eng., 10.1061/(ASCE)GT.1943-5606.0000299, 985–994.
New York. Wang, W. D., Wang, J. H., Li, Q., and Xu, Z. H. (2015). “Design and
Chang, C., and Lin, C. (2011). “{LIBSVM}: A library for support vector
performance of large excavations for Shanghai Hongqiao International
machines.” ACM Trans. Intell. Syst. Technol., 2(3), 1–27.
Airport Transport Hub using combined retaining structures.” J. Aerosp.
Chen, J. J., Zhu, Y. F., Li, M. G., and Wen, S. L. (2015). “Novel excavation
Eng., 10.1061/(ASCE)AS.1943-5525.0000435, A4014002.
and construction method of an underground highway tunnel above
Wong, I., Poh, T., and Chuah, H. (1997). “Performance of excavations
operating metro tunnels.” J. Aerosp. Eng., 28(6), A4014003.
for depressed expressway in Singapore.” J. Geotech. Geoenviron.
Clough, G. W., and O’Rourke, T. D. (1990). “Construction induced move-
ments of in situ walls.” Proc., ASCE Conf. on Design and Performance Eng., 10.1061/(ASCE)1090-0241(1997)123:7(617), 617–625.
of Earth Retaining Structures, ASCE, Reston, VA, 439–470. Woo, S. M., and Moh, Z. C. (1990). “Geotechnical characteristics of soils
Hall, D. L., and Llinas, J. (1997). “An introduction to multisensor data in the Taipei basin.” Proc., 10th Southeast Asian Geotechnical Conf.,
fusion.” Proc. IEEE, 85(1), 6–23. 51–65.
Li, M. G., Chen, J. J., Xu, A. J., Xia, X. H., and Wang, J. H. (2014). “Case Xu, Z. H. (2007). “Deformation behavior of deep excavations supported by
study of an innovative top-down construction method with channel-type permanent structures in Shanghai soft deposit.” Ph.D. thesis, Shanghai
excavation.” J. Constr. Eng. Manage., 10.1061/(ASCE)CO.1943-7862 Jiao Tong Univ., Shanghai, China (in Chinese).
.0000828, 05014003. Zhang, H. B., Chen, J. J., Zhao, X. S., Wang, J. H., and Hu, H. (2015).
Liu, G. B., Ng, C. W., and Wang, Z. W. (2005). “Observed performance of a “Displacement performance and simple prediction for deep excavations
deep multistrutted excavation in Shanghai soft clays.” J. Geotech. supported by contiguous bored pile walls in soft clay.” J. Aerosp. Eng.,
Geoenviron. Eng., 10.1061/(ASCE)1090-0241(2005)131:8(1004), 10.1061/(ASCE)AS.1943-5525.0000481, A4014008.
1004–1013. Zhang, J. F., Chen, J. J., Wang, J. H., and Zhu, Y. F. (2013). “Prediction
Liu, G. B., and Wang, W. D. (2009). Handbook for excavation engineering, of tunnel displacement induced by adjacent excavation in soft soil.”
China Architecture and Building Press, Beijing (in Chinese). Tunnell. Underground Space Technol., 36, 24–33.