Identifying characteristics of pipejacking parameters to assess geological conditions using optimisation algorithm-based support vector machines
Introduction
Pipejacking is a common means of installing utility (e.g., wastewater, natural gas, electricity) pipelines and/or obtaining soil samples for the purpose of geological evaluation (Cui et al., 2015, Ong and Choo, 2016, Wang et al., 2019a). It is an increasingly popular alternative to open-cut construction, especially in densely populated urban areas, due to the more efficient construction processes and reduced environmental impact (Chen et al., 2015, Tan and Lu, 2017, Tan and Wei, 2012, Tan et al., 2015, Wang et al., 2018a, Cheng et al., 2020a). Identifying the type of soils encountered during tunnel construction not only reduces the potential of geo-hazard (e.g. water ingress and surface subsidence; Fu et al., 2019, Cheng et al., 2019a) but also prevents unplanned downtimes and operation costs (e.g. jamming of shield; Barla et al. 2006). While geo-hazard prevention techniques exist, there remains significant motivation in the industry to sense a variety of complex geological conditions that is likely to confront during tunnel excavation (Shen et al., 2017, Qiu et al., 2018, Wang et al., 2018b, Zhang et al., 2018, Cheng et al., 2019b, Modoni et al., 2019, Wang et al., 2019b; Cheng et al., 2020b).
In pipejacking, tunnelling parameters such as jacking force, cutter wheel torque, flow rate of feedline, pressure in slurry circulating system, and slurry density are highly variable due to their dependence on a number of influencing factors including surrounding geology, lubrication performance, work stoppage and pipe misalignment (Norris and Milligan, 1992, Milligan and Marshall, 1998, Milligan and Norris, 1999, Chapman and Ichioka, 1999, Pellet-Beaucour and Kastner, 2002, Rahjoo et al., 2012, Reilly and Orr, 2012, Choo and Ong, 2015, Sheil et al., 2016, Cheng et al., 2017, Cheng et al., 2018, Cheng et al., 2019c, Ochmański et al., 2018, O’Dwyer et al., 2018, Ren et al., 2018, Phillips et al., 2019, O’Dwyer et al., 2019, Zhang et al., 2019a; Wei et al., 2019). For instance, the total jacking loads FT required to advance a shield machine consists of the face resistance F0 and the soil-structure frictional resistance Fs. A significant body of research has indicated that effective lubrication can significantly reduce Fs, whereas work stoppages and pipe misalignment can lead to significant and transient increases in Fs. Conversely, jacking into clayey gravel from gravel can cause an increase in F0 due to increased contact of the shield face and therefore an increase in FT. Although previous studies performed over the past three decades have greatly enhanced our understanding of the influencing factors and their influence on FT, the relationship between surrounding geology and tunnelling parameters remains unclear. Systematic research to reveal the surrounding geology-tunnelling parameters relationship is therefore essential (Lai et al., 2018, Qiu et al., 2019; Wu et al., 2019, Wu and Shao, 2019, Song et al., 2020a, Song et al., 2020b). In particular, identifying the response of tunnelling parameters to sudden change of geological condition (e.g., karst cavern and fault zone) is crucial for shield operators to adopt appropriate and timely countermeasures towards preventing geo-hazard from happening.
Traditionally, the manipulation of tunnelling parameters during pipejacking is highly dependent on the shield operator’s accumulated site experience, yet the effectiveness of this process determines the safety of tunnel construction and adjacent properties. This may explain the reoccurrence of geo-hazards that is often seen in daily news and social media worldwide and suggest that people may not be adequately aware of the importance of geo-hazard prevention despite the complexity of real-life geological lithologies (Ong and Choo, 2011, Mehdizadeh et al., 2017) and soil-structure interaction (Ong et al., 2003, Ong and Choo, 2018, Choo and Ong, 2020). The proliferation of tunnelling parameters retrieved from modern tunnel shield machines presents substantial opportunity for the application of data-driven artificial intelligent (AI) techniques that can identify patterns in data without reference to known labels (Sheil et al. 2020). This study examines the potential for data-driven AI techniques to identify the type of encountered soils, reducing the possibility of jamming and potential of geo-hazard. A selection of the most popular AI techniques proposed in the literature were considered for this purpose (Zhang et al., 2017, Zhang et al., 2019b, Zhang et al., 2020). Their relative merits were assessed by comparing AI predictions to monitored data from a recent case history of pipejacking in gravel and clayey gravel soils.
Section snippets
Overview
Data-driven approaches identify characteristics of the measured system by utilising information retrieved from the measured data, rather than by modelling the system response. However, in most practical problems the measured data (i.e. the outputs) are not labelled. For this reason, ‘unsupervised’ machine learning algorithms, used to infer patterns in data without reference to known outcomes, is popular. The aim of this study is to develop an improved understanding of existing pipejacking
Implementation
The Support Vector Machine (SVM) algorithm discussed above was implemented using the Python module Scikit-learn (Pedregosa et al. 2011). All data were pre-processed to maximise the efficiency and performance of the learning process and to ensure that the importance of the input dataset is equalised. A ‘min-max scaler’ was introduced to scale the dataset (Masters 1993), so that all data were laid between our specified range of minimum and maximum. The min-max scaler transforms the features of
Project overview
Cheng et al., 2017, Cheng et al., 2018, Cheng et al., 2019a, Cheng et al., 2019c) describe a total of four drives in the soft alluvial deposits of the Shulin district in Taipei County, Taiwan. Two out of the four drives were considered here to assess the selected AI techniques, namely drives C and D. Fig. 6a shows the location of Drives C and D and associated geological boreholes. The length for Drives C and D was 75 m and 102 m respectively. Overburden depth relative to the tunnel crown for
Classification results: drives C and D
The three hyperparameter optimisation algorithms were first considered to evaluate their performance for this problem. To assess the feasibility of the optimisation algorithms, the fitness at each generation was traced, as shown in Fig. 9. It can be observed that for gravel identification at drive C (Fig. 9a), the GA achieved an optimal solution immediately whereas for PSO it was achieved within four generations. In contrast, for drive D the GA technique required 19 generations to achieve an
Conclusions
This paper has examined the potential for the use of AI techniques to identify geological conditions encountered during pipejacking. A selection of the most popular parameter optimisation algorithms was considered to improve the accuracy and efficiency of the AI predictions, namely genetic algorithms, particle swarm optimisation and grid search. Based on the results and discussion, some main conclusions can be drawn as follows:
- (a)
Decomposition of the data was implemented to transform the
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This study would not have been possible without financial supports from the Special fund for Basic Scientific Research of Central Colleges, Chang’an University, under Grant No. 300102269502. The third author is funded by the Royal Academy of Engineering under the Research Fellowship Scheme.
Author contribution
This paper represents a result of collaborative teamwork. Dr. Wen-Chieh Cheng and Dr. Brian B. Sheil prepared the original manuscript and provided edits to the revised manuscript. Mr. Xue-Dong Bai performed a series of analyses and prepared the revised manuscript under the first author’s supervision. Mr. Fei Wang, and Miss Ge Li revised the figures and tables and gave edits to the revised manuscript.
References (61)
- et al.
Analysis of jacking forces during microtunnelling in limestone
Tunn. Undergr. Space Technol.
(2006) - et al.
Prediction of jacking forces for microtunnelling operations
Tunn. Undergr. Space Technol.
(1999) - et al.
A simple approach for characterising tunnel bore conditions based upon pipe-jacking data
Tunn. Undergr. Space Technol.
(2018) - et al.
Lubrication performance of pipejacking in soft alluvial deposits
Tunn. Undergr. Space Technol.
(2019) - et al.
Recent massive incidents for subway construction in soft alluvial deposits of Taiwan: a review
Tunn. Undergr. Space Technol.
(2020) - et al.
Modelling liner forces response to very close-proximity tunnelling in soft alluvial deposits
Tunnelling Underground Space Technol.
(2020) - et al.
Assessment of non-linear rock strength parameters for the estimation of pipe-jacking forces. Part 2. Numerical modelling
Eng. Geol.
(2020) - et al.
Field performance of concrete pipes during jacking in cemented sandy silt
Tunn. Undergr. Space Technol.
(2015) - et al.
Strength of sandy and clayey soils cemented with single and double fluid jet grouting
Soils Found.
(2019) - et al.
Automated numerical modelling for the control of EPB technology
Tunn. Undergr. Space Technol.
(2018)
Back-analysis and finite element modeling of jacking forces in weathered rocks
Tunn. Undergr. Space Technol.
Assessment of non-linear rock strength parameters for the estimation of pipe-jacking forces. Part 1. Direct shear testing and backanalysis
Eng. Geol.
Experimental and analytical study of friction forces during microtunneling operations
Tunn. Undergr. Space Technol.
Experiences of utility microtunnelling in Irish limestone, mudstone and sandstone rock
Tunn. Undergr. Space Technol.
Enhancing discharge of spoil to mitigate disturbance induced by horizontal jet grouting in clayey soil: theoretical model and application
Comput. Geotech.
Fundamentals and applications of microwave energy in rock and concrete processing - a review
Appl. Therm. Eng.
Multivariate adaptive regression splines for inverse analysis of soil and wall properties in braced excavation
Tunn. Undergr. Space Technol.
Clay dosage and water/cement ratio of clay-cement grout for optimal engineering performance
Appl. Clay Sci.
Real-time analysis and regulation of EPB shield steering using random forest
Autom. Constr.
Random forests
Machine Learning
STL: a seasonal-trend decomposition procedure based on loess
J. Off. Statist.
Failure investigation at a collapsed deep excavation in very sensitive organic soft clay
J. Perform. Constr. Facil
The use of tunnelling parameters and spoil characteristics to assess soil types: a case study from alluvial deposits at a pipejacking project site
Bull. Eng. Geol. Environ.
Using post-harvest waste to improve shearing behaviour of loess and its validation by multiscale direct shear tests
Appl. Sci.
Evaluation of pipe-jacking forces based on direct shear testing of reconstituted tunneling rock spoils
J. Geotech. Geoenviron. Eng.
Multivariate adaptive regression splines
Ann. Statist.
Cracking performance of an operational tunnel lining due to local construction defects
Int. J. Geomech.
Extreme deformation characteristics and countermeasures for a tunnel in difficult grounds in southern Shaanxi, China
Environ. Earth Sci.
Cited by (43)
Evaluation and machine learning prediction on thermal performance of energy walls in underground spaces as part of ground source heat pump systems
2024, Tunnelling and Underground Space TechnologySafety evaluation of buildings adjacent to shield construction in karst areas: An improved extension cloud approach
2023, Engineering Applications of Artificial IntelligenceInvestigation into the flow characteristics of slurry shield pipeline system under sandy pebble stratum: Model test and CFD-DEM simulation
2023, Powder TechnologyCitation Excerpt :With the massive development of infrastructure construction, a large number of underground projects are in full swing around the world [1–4].
Prediction of geology condition for slurry pressure balanced shield tunnel with super-large diameter by machine learning algorithms
2023, Tunnelling and Underground Space TechnologyCitation Excerpt :Previous studies have confirmed the correlation between TBM operational parameters and geological conditions, and the application of machine learning shows enormous potential to address such complex nonlinear problems (Liu et al., 2020b; Yan et al. 2021). Many researchers have introduced machine learning algorithms to predict geological conditions with operational parameters (Cheng et al., 2020; Erharter et al., 2019, 2020; Erharter and Marcher, 2020; Hou et al., 2022; Liu et al., 2019, 2020a; Liu et al., 2021). For example, Liu et al. (2020b) proposed a rock classification model based on the AdaBoost algorithm by adopting statistical indicators of hard-rock TBM operational data as input variables.
A novel Bayesian inference method for predicting optimum strength gain in sustainable geomaterials for greener construction
2022, Construction and Building Materials