Identifying characteristics of pipejacking parameters to assess geological conditions using optimisation algorithm-based support vector machines

Highlights

• Potential for AI techniques to identify changes in geology is explored.

• Data decomposition into feature-based sub-series accentuates their features.

• The PSO-SVM model can be useful in providing accurate predictions.

Abstract

Detecting sudden changes in geological conditions (e.g., karst cavern and fault zone) during tunnelling is a complex task. These changes can cause shield machines to jam or even induce geo-hazards such as water ingress and surface subsidence. Tunnelling parameters that relate closely to the surrounding geology have proliferated in recent years and present a substantial opportunity for the application of data-driven artificial intelligent (AI) techniques that can infer patterns from data without reference to known, or labelled, outcomes. This study explores the potential for support vector machines (SVM) to identify changes in soil type during tunnelling towards reducing the possibility of jamming and geo-hazard development. All tunnelling data were pre-processed to convert time series data into feature-based sub-series. A selection of the most popular parameter optimisation algorithms was explored to improve the accuracy of the AI predictions. Their relative merits were evaluated through comparisons with a recent pipejacking case history undertaken in gravel and clayey gravel soils. The results highlight an exciting potential for the use of optimisation algorithm-based SVMs to identify changes in soil conditions during pipejacking.

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 F_T required to advance a shield machine consists of the face resistance F₀ and the soil-structure frictional resistance F_s. A significant body of research has indicated that effective lubrication can significantly reduce F_s, whereas work stoppages and pipe misalignment can lead to significant and transient increases in F_s. 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 F_T. Although previous studies performed over the past three decades have greatly enhanced our understanding of the influencing factors and their influence on F_T, 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.