1 Digital twins for urban underground space 1 2 Nandeesh Babanagar1, Brian Sheil1, Jelena Ninic2, Qianbing Zhang3, Stuart Hardy4 3 4 1 Department of Engineering, University of Cambridge, United Kingdom 5 2 School of Engineering, University of Birmingham, United Kingdom 6 3 Department of Civil Engineering, Monash University, Australia 7 4 Arup, London, United Kingdom 8 9 10 11 12 13 14 Initial submission: June 2024 15 Revised submission based on reviewers’ comments: September 2024 16 Figures: 14 17 Tables: 5 18 2 ABSTRACT 19 Digital twins (DTs) offer promising benefits to address several inherent problems in 20 underground construction, yet confusion surrounds the concept due to its context-specific 21 nature which hinders more widespread adoption. This paper seeks to clarify DT-related 22 terminologies from a built environment perspective and define the features and maturity levels 23 of DTs for underground spaces, considering their unique challenges. A layered architecture 24 for constructing DTs is proposed, offering various options based on functionality, 25 technological advancement, and expected value. Additionally, a comprehensive literature 26 review of technologies enabling the development of digital twins for underground spaces is 27 presented, including data-driven ground modelling, site investigation and design process 28 integration with BIM, computational BIM, and advanced sensing and instrumentation. The 29 paper identifies synergies between DTs and the observational method in geotechnical 30 engineering, highlights research gaps, and proposes a transition to a prescriptive, knowledge-31 based DT. Furthermore, exemplar use cases of underground DTs throughout their lifecycle are 32 explored, demonstrating their potential value. 33 Keywords: BIM, digital twins, observational method, underground construction. 34 1 INTRODUCTION 35 Escalating demand for infrastructure, including housing, energy, water, sanitation, and 36 transport, combined with above-ground congestion has propelled recent significant increases 37 in underground construction. For instance, 10% of our rail network in Europe is located 38 underground, and in London, this figure rises to 45% (Pritchard and Preston 2018). However, 39 traditional underground construction methods struggle to meet the rising demand due to issues 40 such as low productivity, prolonged construction timelines, high costs, and safety concerns 41 (Quigley et al. 2016; Sun et al. 2023). Uncertainties in ground conditions and various design 42 and construction stages exacerbate these challenges. 43 Digital technologies emerge as one of the promising enablers to overcome these limitations 44 and revolutionise underground construction by offering unprecedented capabilities in planning, 45 monitoring, and execution. The adoption of building information modelling (BIM) has 46 improved delivery and performance, fostered collaboration and innovation, and introduced 47 opportunities for a new level of automation. Although widely used in the construction industry, 48 the current adoption level and capabilities of BIM does not fully exploit the full range of 49 benefits of the digital revolution. Digital twin (DT) is an emerging solution within the industry 50 3 4.0 framework, which was first proposed as a conceptual model for product lifecycle 51 management by Grieves (2002). It refers to a set of virtual information representations that 52 mimics the structure, context, and behaviour of a natural, engineered, or social system which 53 is dynamically updated with data from its physical counterpart and has a predictive capability 54 to inform decisions for realising value (AIAA 2020). It offers real-time monitoring, 55 performance analysis, and predictive maintenance, contributing to enhanced productivity, 56 safety, and operational efficiency (Amthiou et al. 2023). 57 The term DT has become commonplace jargon due to its expansive applicability across various 58 industries and contexts particularly in the fields of manufacturing, production, and operations 59 (Camposano et al. 2021). There can be different scales of DTs ranging from physical products 60 (objects/ components), system or processes. Accordingly the contemporary digital 61 representation may capture physical entities such as aircraft engines, social constructs and 62 processes such as stock market operation, and composite systems which encompass both of 63 these, such as construction projects (Kritzinger et al. 2018). The basic premise of DTs is to 64 solve information silo problems by combining and linking isolated data or models in different 65 field spaces, time domains and cognitive contexts of complex systems (Hu et al. 2023). 66 Due to the broad applicability of DTs, different organisations define and interpret DT concepts 67 based on domain-specific functions and characteristics (Kritzinger et al. 2018). These different 68 interpretations have led to significant confusion and misunderstandings associated with DTs. 69 Tao et al. (2019a) state that a DT has three main elements: a physical artefact, a digital 70 counterpart that mirrors the physical artefact, and the connection that binds the two. It has also 71 been proposed that a true DT should provide a service, such as simulation, monitoring and 72 decision-making, and/or control of the physical object (Tao et al. 2019b). In construction, DTs 73 are commonly misinterpreted as digital models of assets (e.g., only 3D BIM models) or as 74 simulations for prediction of behaviour. However, DTs go beyond these as they are virtual 75 replicas of real-world physical products or systems (Naderi and Shojaei 2023). 76 In the realm of digital representations, the connection between elements differentiates a DT 77 from related concepts such as a digital model and a digital shadow, each serving distinct 78 purposes (see Figure 1). A DT incorporates a bidirectional and near real-time exchange of data, 79 information, and knowledge between the physical and virtual counterparts, enabled by data 80 acquisition technologies (DAX) like advanced sensing, internet of things (IoT), high-speed 81 networking, and machine learning (ML) and data science technologies (data processing 82 4 technologies) (Sacks et al. 2020a), which may even have a human in-the-loop to make 83 decisions (Agrawal et al. 2023; Zhang et al. 2022). DTs can also involve multiple interoperable 84 information models such as geometric model, analytical model, behavioural models to create 85 living transdisciplinary simulations that update and change in real-time in response to changes 86 of the physical counterpart over the whole lifecycle. Such information models can then be 87 employed for optimising processes, supporting decision making, virtual control, and analysis. 88 A digital model, on the other hand, is a static, often only a geometric representation of an object 89 or system. It supplies a simplified, non-real-time view for design, visualisation, and analysis. 90 Lastly, a digital shadow refers to a digital representation encompassing historical data and 91 behaviours, offering insights into past performance and aiding in predictive analytics but 92 missing near-real time data processing, interoperability between different information models 93 and feedback to the physical twin (Fuller et al. 2020). 94 95 Figure 1: Conceptual representation of (a) digital model, (b) digital shadow, and (c) digital twin 96 This paper identifies the challenges in developing DT for underground space and presents a 97 conceptual definition and maturity levels of underground DTs. Further, research in 98 technologies which will be future enablers for underground DT has also been recognised. A 99 key academic contribution of this paper is a vision for transitioning from a well-developed and 100 codified observational method in geotechnical engineering to knowledge-based prescriptive 101 DT. Further, this study demonstrates value delivered by underground DT throughout the 102 lifecycle by providing exemplar use cases. 103 5 2 DIGITAL TWINS IN CONSTRUCTION INDUSTRY 104 While DT technology has gained attention in various sectors, its application in construction is 105 still in its infancy (Wu et al. 2022a). The Centre for Digital Built Britain (CDBB) defines DTs 106 as “a realistic digital representation of assets, processes or systems in the built or natural 107 environment” (Bolton et al. 2018). Boje et al. (2020) notes that a comprehensive definition of 108 a DT should delineate both its ontological dimensions detailing its inherent properties, and its 109 epistemological aspects, explaining the methods and processes through which knowledge 110 about the DT is acquired and validated. It should be purpose driven and should identify the 111 fundamental information and technology elements, clarify the interconnections among these 112 elements, and discern their individual and collective functionalities. 113 Jiang et al. (2021) interpreted a construction DT to comprise five parts: (i) the physical 114 components/ systems, (ii) the digital representation, (iii) physical-digital connections, (iv) data 115 and (v) service. Figure 2 describes the various mandatory and optional components required 116 for a construction DT according to Jiang et al. (2021). More recent frameworks like Torzoni 117 et al. (2024) define predictive capabilities as being a core requirement for civil engineering 118 applications; these are considered as offerings of the service component of DT framework as 119 shown in the Figure 2. 120 121 122 Figure 2: Essential and optional connections of DT according to Jiang et al. (2021) 123 There has been a marked increase in DT research applied to construction since 2018, with 124 scholars exploring various phases of the construction lifecycle. In the engineering and design 125 6 phase, the focus has been on structural and layout optimisation, quality assessment and 126 improvement in design management by optimising information flows to enhance productivity 127 by reducing design time and rework costs (Moshood et al. 2024; Opoku et al. 2021). BIM is 128 considered a cornerstone for creating a DT for collaborative information-sharing and decision-129 making regarding material section, energy management (Kaewunruen and Lian 2019) and 130 procurement. However, some researchers consider these as secondary applications of DT, as 131 they are fundamentally aligned with the virtual design and construction (VDC) concepts which 132 was first introduced 2001 as part of the by the Centre for Integrated Facility Engineering at 133 Stanford University (Rafsanjani and Nabizadeh 2023). 134 For the construction phase, previous research has primarily centred on assessing the integrity 135 of structural systems e.g. Angjeliu et al. (2020), Gerhard et al. (2020). Additional research has 136 considered construction progress, budgeting, value, quality, sustainability and carbon footprint 137 assessment (Huang et al. 2023a), and safety monitoring and management (Jiang et al. 2021; 138 Teizer et al. 2022). DTs have been most widely applied to the operation and maintenance phase, 139 particularly for facilities management, involving advanced building management systems for 140 optimising energy use and structural health monitoring of buildings (Kaewunruen and Lian 141 2019). However, there is a notable lack of literature related to the application of DT to the end-142 of-life/demolition phase (Opoku et al. 2021). 143 The digital twin consortium, which is a global organisation that works to develop digital twin 144 technology, highlights the required capabilities of a DT in six key categories - data services, 145 integration, intelligence, user experience, management, and trustworthiness (McKee 2023). 146 The Gemini principles provided by CDBB advocates the requirement of similar principles. 147 These principles, range from ensuring a clear purpose and public good to prioritise data security 148 and openness to emphasising the importance of establishing reliable and functional DTs. 149 Federation is identified as a key element i.e., the linking of multiple DTs from different sources 150 or organisations to create a more comprehensive and interconnected network. 151 The construction of a DT requires the integration of various enabling technologies. Khajavi et 152 al. (2019) proposed the “House of DT” framework (Figure 3) to identify such technologies. It 153 is crucial for the model to hold both graphical and semantic information, which is possible with 154 BIM. IoT and sensor technologies capture real-time data from the object or structure of interest. 155 These data streams are then analysed using data analytics and ML to derive meaningful 156 insights. Visual programming extends BIM models, allowing computations and connections 157 7 with the sensor network. Cloud-based servers often serve as databases for storing extensive 158 data and virtual models. The advancements in BIM, including increased interoperability, data 159 accuracy, collaboration abilities, and enhanced visualisation and simulation capabilities, 160 contribute to the development of DT from BIM (Nguyen and Adhikari 2023). 161 162 Figure 3: House of digital twins defined by Khajavi et al. (2019) 163 3 UNDERGROUND DIGITAL TWINS 164 3.1 Challenges in underground space 165 Within the specialised domain of underground space, DT research is still regarded as “new 166 territory” (Wu et al. 2022a). Given the domain specificity of DT definition and maturity levels, 167 underground construction requires a specialised approach to DT concepts due to the unique 168 challenges associated with subterranean environments. For example, subsurface conditions are 169 inherently uncertain, non-uniform and anisotropic. The intricate relationship between the 170 ground and structure introduces complexity through non-linear and temporally varying 171 parameters, distinguishing it from conventional structural analysis for system identification. 172 This necessitates a denser and more continuous network of sensors to collect data. The acquired 173 data are also often multivariate, uncertain, sparse, incomplete, and potentially corrupted with 174 spatial/temporal dimensions, which adds complexity to follow-on analysis (Phoon et al. 175 2022a). Geological data introduces a three-dimensional (3D) variation that poses measurement 176 challenges not encountered in above-surface infrastructure. Moreover, the sensors required for 177 underground construction are usually buried, or embedded within structures which are difficult 178 8 to access, necessitating the use of sensors with high durability and low signal interference from 179 various dynamic factors such as settlements, water seepages and chemical exposure. 180 Data management also faces unique issues of format standardisation - for instance industry 181 foundation classes (IFC) and CityGML are not finalised due to the complexity of geological 182 data and structure interaction, differences of excavation methods and supporting strategies 183 (Huang et al. 2021). The distinctive characteristics of geological objects require the 184 simultaneous maintenance of multiple solid representation schemes. These issues have also led 185 to a lagged implementation of BIM in underground space compared to above-ground 186 structures. However, the upcoming IFC version aims to address some of these challenges (Rives 187 et al. 2020). 188 Unlike DTs for surface structures, the numerical modelling, 3D geological visualisation, 189 mapping of surrounding utilities in the subsurface itself involves the application of 190 sophisticated sensing/ exploratory methods, complex simulations, data, and analytics. These 191 complexities, including a multi-disciplinary nature, a high degree of customisation, and long-192 term cycles of underground infrastructure, call for a unique paradigm of DTs for underground 193 space (Li et al. 2024b), requiring separate maturity levels, and further research and exploration. 194 3.2 Maturity dimensions of underground DT 195 The maturity levels of DTs indicate their development stages within a sector, facilitating the 196 definition, management, and integration of their progress. Researchers have assessed maturity 197 based on factors like information flow (Fuller et al. 2020), technological advancement, and 198 functionality by include aspects like model, data, content, level of control, human-machine 199 interaction, and computing capability (Uhlenkamp et al. 2022). DTs cover a broad spectrum of 200 dimensions, including functional completeness, federation, and value, rather than merely being 201 a technological overlay or following a linear progression. Based on these factors the maturity 202 model for the construction industry can consists of four dimensions: descriptive twin, reflective 203 twin, predictive twin, and prescriptive twin as defined in the Table 1 along with the description 204 of each dimension. Further Li et al. (2024) proposed several rubrics for evaluating these 205 dimensions in the context of underground space, and the degree of development is graded as 206 different levels of maturity, as summarised in Table 2. 207 208 209 9 210 Table 1: Dimension of DT maturity pertaining to construction 211 Maturity dimensions Description Key Questions Addressed Descriptive Twin Collects geometric, material, performance, and historical data to recreate entities digitally. Lacks real-time updates and analysis (Gürdür Broo et al. 2022; Sacks et al. 2020a) “What is it?” Reflective Twin Uses sensors and instant communication for real- time status updates, visualising construction progress, operations, and structural responses. Promotes efficient decision-making with sensing and control (Agrawal et al. 2023) “What is happening?” Predictive Twin Predicts future states like mechanical responses and risks using statistical, mechanistic, and knowledge models. Continuously optimises with field feedback (Torzoni et al. 2024) “Why is it happening?” and “What will happen?” Prescriptive Twin Analyses and optimises construction and operations, providing real-time decisions. Supports remote and automated control of intelligent devices (Wang et al. 2021). “What should be done?” 10 Table 2: Summary of rubric grading for DT maturity evaluation in underground infrastructure (Li et 212 al. 2024b) 213 214 Note: If the rubric does not meet the requirements of Level 1, the level should be 0 and not 215 qualified as DTs. 216 Accordingly, the concept of underground DT is not confined to a rigid definition; it 217 encompasses a spectrum of possibilities and variations. Each dimension of a DT can have 218 different levels of advancement based on the context of application. The degree of 219 advancement depends on the specific use case and the expected value from adopting a DT 220 approach, as outlined in the rubrics in the table. 221 Dimension Maturity Rubric Level-1 Level-5 D es cr ip ti v e tw in Data & semantic richness Single-scale, multiple components, basic data → Multi-scale, multiple components, linked data structures Model quality Actual data, low (level of detail) LOD, low fidelity model with acceptable error → Actual data, real-time, multi-LOD models Model completeness Geometric (OR) process → Geometric (AND) behavioural (AND) process R ef le ct iv e tw in Sensing capabilities Manual, partial sensing → Autonomous sensing Interoperability Data & model partially standardised and federated → Fully standardised, highly federated, and high- security sharing Connection capabilities High manual intervention, low fidelity → Integrated and automated connectivity and high fidelity Model update capabilities Partially updatable attributes, delayed update → Fully updatable attributes, real-time updates P re d ic ti o n tw in Analysis capabilities Actual data supported, manually constructed analysis. → Multi-physics and automatic analysis model enhanced by feedback Prediction capabilities Data-driven fitting prediction → Scenario deduction for complex systems P re sc ri p ti v e tw in Decision capabilities Artificial control, parametric design enabled → Feedback-optimised autonomous control, generative design based 11 3.3 Construction of an underground DT 222 The underground DT should capture both the physical entities and the processes involved for 223 design, construction, operation, and end-of life phases of underground structures. Figure 4 224 conceptualises a layered architecture for an underground DT. This architecture begins with a 225 physical layer, that encapsulates entities and processes integral to underground construction 226 such as: (i) site environment capturing the geophysics, hydrology, geology, and topography of 227 the site, (ii) foundation structures - including various components such as piles, rafts and 228 retaining wall, (iii) temporary support structures- such as struts and walers, anchorages, tie rods 229 and shotcrete, (iv) dewatering arrangements and, (v) utilities and services. The data richness, 230 model quality, and completeness of these entities can vary depending on the context and use-231 case. 232 Data from the physical layer is captured by the sensing and data acquisition layer, which utilises 233 various sensing equipment. This equipment ranges from survey data and laboratory test for 234 various geological and engineering parameters to satellite imagery, strain sensors, 235 inclinometers, piezometer, geophysical equipment (e.g., ground penetration radars (GPR)), and 236 vision data (computer and/or human). The maturity of the underground DT depends on the 237 degree of automation in sensing and the reliability and accuracy of the data collected. These 238 data are then transmitted to the data management layer by the transmission layer. Depending 239 on the data type, range of data transfer, and connection requirements, the transmission layer 240 can employ a broad spectrum of methods, from wired connections to high-fidelity wireless 241 connections. IoT-enabled sensing and advancements in edge computing can also integrate the 242 sensing, transmission, and data management layers to a high degree. 243 In the data management layer, collected data are stored and processed. This layer is capable of 244 fusing and processing data from multiple sources, performing preliminary analytics, and 245 storing data in databases. It should seamlessly interface with the visualisation and modelling 246 layers, where BIM plays a pivotal role. Additionally, the data management layer enables data 247 exchange with various software packages or plugins used in the application layer. Leveraging 248 these inputs, a virtual model is created using a multi-physics approach to depict various details 249 of the subsurface environments. This model encompasses diverse semantic information and 250 reflects engineering properties, serving as the foundation throughout the structure's lifecycle. 251 Dynamic display features of the DT simulate real-time state changes, enabling the prediction 252 of potential states in actual underground construction (Wu et al. 2022a). 253 12 The application layer of the underground DT provides a range of solutions and simulations, 254 from descriptive twin applications to more advanced prescriptive phases. Example solutions 255 include data-driven ground modelling, real-time model updating, structural health monitoring 256 and warning systems, and automated analyses such as settlement, stability, and constructability 257 assessments. It also includes risk prognosis, design, cost, and carbon optimisation simulations. 258 Stakeholders use this layer to make real-time decisions based on multi-scale federated views 259 of the entities and processes in the physical layer. The decision-making and decision 260 implementation may also be automated in some applications, such as in the concept of 261 hypertunnel, where swarm robotics is used for tunnelling works autonomously informed by the 262 prescriptive digital twin (King 2022). 263 The data management layer plays a central role in the proposed architecture, where all the data 264 are stored. Figure 5 which is a high-level abstraction of figure 4, illustrates the interaction 265 between various layers within an underground DT and other DTs. The data storage, in a 266 common data environment, should be seamlessly accessed and modified by the data 267 visualisation and modelling layer, where BIM is pivotal. The data management layer must also 268 support data exchange with various software packages or plugins in the application layer. 269 Stakeholders can always access the data management layer to view raw or semi-processed data, 270 the visualisation layer to view and edit BIM models, and the application layer to run 271 simulations or analyses on the BIM models based on data from the data management layer. 272 The architecture also includes a controlled federation with secured two-way data exchange 273 capabilities with other DTs, such as the city level DT (an ecosystem of DTs connected via 274 securely shared data; Bolton et al. 2018), superstructure DT, equipment DTs etc., aligned with 275 the Gemini principles for DT development. The “National Underground Asset Register 276 (NUAR)” (2024) is a project which is developing a basic DT which integrates data of various 277 underground assets in the UK. Federation of the underground DT with these DTs will be 278 crucial. Similar project has also been undertaken in Singapore (“Singapore-ETH Centre” 279 2023). Federation is not limited to these external DTs, but each component in the physical layer 280 can have its own DT (e.g., equipment DTs), contributing to the formation of the underground 281 DT (see Figure 4). While federation is desirable for DT development (Huang et al. 2023c), the 282 computational complexity and cost associated with exchanging data across all federated DTs 283 may be challenging for a universally applicable DT (Li et al., 2024). 284 13 285 Figure 4: A conceptualisation of an underground digital twin architecture 286 14 287 Figure 5: Interaction between different layers within underground DT and federation across DTs 288 To address this challenge, a trade-off between fidelity and computational efficiency must be 289 considered. A recommended approach is to adopt a multi- LOD model of appropriate 290 abstraction level of and technology, aligned with specific scenarios and purpose for data 291 federation. The resulting multi-scale modelling and multi-level LOD representation approach 292 will provide model precision according to the need of a particular application whilst 293 simplifying representations for other applications. Figure 6 illustrates an example, where Ninić 294 et al. (2020) has adopted this multi-LOD approach to establish efficient data exchange between 295 various models to suit analysis requirements of different stakeholders. 296 One of the primary functions of the data management layers is data fusion and processing. The 297 key challenge lies in handling heterogeneous data from different sensors, along with temporal 298 and spatial misalignments, and complex models due to the multi-physics nature of underground 299 spaces. Probabilistic fusion models, such as Bayesian networks, are effective in environments 300 with inherent uncertainty, combining data with varying confidence levels to predict outcomes 301 or assess risks (Macías et al. 2024). Semantic data fusion becomes essential when the meaning 302 and context behind data are critical (Li et al. 2024c). Ontologies and knowledge graphs enable 303 the integration of heterogeneous data by providing a formal structure for defining concepts, 304 relationships, and rules within a domain. This ensures consistent interpretation and allows for 305 15 advanced querying and inference, leading to more accurate and context-aware decisions, which 306 are vital for developing digital twins in complex infrastructures like underground spaces (Yu et 307 al. 2021). 308 309 Figure 6:Example of a multi-LOD approach for federation of DTs (reproduced from (Ninić et al. 310 2020b)) 311 4 IDENTIFYING KEY RESEARCH AREAS INFLUENCING THE 312 DEVELOPMENT OF UNDERGROUND DTS 313 Underground DT literature is highly fragmented, with applications ranging from tunnelling 314 operations to risk prognosis during deep excavations. Figure 7 presents a keyword co-315 occurrence map based on a comprehensive concatenated Scopus search incorporating DT and 316 underground construction which provides insights into the prevailing research landscape. This 317 visual representation serves as a valuable guide in recognising the focal points of underground 318 DT research which feature BIM, the finite element method (FEM), ML, sensing, and 319 monitoring systems as prominent areas. 320 Figure 8 presents a Sankey diagram constructed using Latent Dirichlet Allocation-based topic 321 modelling of the titles and abstracts from a Scopus database search with the keywords ‘BIM’, 322 ‘digital twin’, ‘underground’, ‘tunnels’, and ‘geotechnics’, to understand the evolution of 323 research topics in this area. Design evaluation, risk assessment, and the application of computer 324 vision for sensing feature as emergent areas. 325 16 326 Figure 7: Mapping of co-occurrence of keywords (based on 275 select papers; minimum number of 327 occurrence was set to 4; search criteria ( TITLE-ABS-KEY ( underground OR excavation OR geotech* ) AND 328 TITLE-ABS-KEY ( digital AND twin OR bim ) ) AND ( LIMIT-TO ( SUBJAREA , "ENGI" ) OR LIMIT-TO ( 329 SUBJAREA , "COMP" )) 330 331 332 Figure 8: Evolution of research topics in BIM and DT pertaining to underground, tunnels and 333 geotechnics- adopted from Ninić (2024) 334 17 A comprehensive literature review is described below to identify priority technical challenges 335 and ongoing research areas for underground DTs, informed by Figure 7 & 8. These challenges 336 include BIM integration with site investigation data, BIM-based ground modelling methods, 337 BIM-FEM interoperability, and advanced sensing, instrumentation, and monitoring. Another 338 area of active research in construction is computational BIM which can significantly influence 339 the development of underground DTs. A proactive understanding of these advancements is 340 essential for the development of a well-integrated underground DT. 341 4.1 Data-driven ground modelling 342 Ground models are a digitised model characterising each sub-surface element with parameters 343 that define the ground behaviour, such as layer thickness, geo-structural features, and 344 mechanical/hydraulic properties. Obtaining a precise site representation from limited disparate 345 data is challenging. Interpolation methods are broadly categorised as deterministic (e.g., 346 splines, inverse distance weighting, polynomial interpolation) and probabilistic (e.g., kriging, 347 sequential simulation). Deterministic methods are convenient but overlook natural ground 348 variability and uncertainty in field and lab procedures (Phoon 2018). Probabilistic geostatistical 349 methods, especially kriging, address these limitations by offering linear, unbiased, and 350 minimum estimation variance, making them widely adopted (Che and Jia 2019). The 351 proliferation of ML has given rise to numerous data-driven methods which effectively integrate 352 sparse observation data and prior knowledge in geotechnical site characterisation, an example 353 of which is shown in Figure 9. Approaches such as iterative convolution XGBoost (Shi and 354 Wang 2021a, 2023), multiple point statistics (Wang et al. 2022b; Zhou et al. 2024a; b), sparse 355 Bayesian learning, hierarchical Bayesian modelling, Gaussian process regression (Yoshida et 356 al. 2021), and geotechnical Lasso (Phoon et al. 2022b) have been shown to be capable of 357 automatically improving geotechnical site characterisation results. Object oriented parametric 358 geological modelling using conditional random fields approach has also been explored (Xie et 359 al. 2023). These data-driven models automatically improve as the measurement dataset grows, 360 reducing the influence of prior knowledge on geotechnical site characterisation over time. 361 However, there approaches are complex for practical application and require further research 362 and development. Access to real-world data to develop and test reliable models, data 363 standardisation and protection remain crucial to address these challenges (Phoon et al. 2023). 364 18 365 Figure 9: Ground model developed from sparse data using CNN based iterative interpolation; 366 reproduced from Shi and Wang (2021b) 367 4.2 BIM integration of site investigation and geospatial data 368 Ground models, representing the subsurface in a BIM environment, provide a better quality of 369 geological prognosis and also enable visualisation and coordination among all stakeholders 370 (Berdigylyjov and Popa 2019). The first step in building the ground model is the integration of 371 geotechnical and geological site investigation data into BIM for visualisation and stratigraphy 372 analysis. These models requires constant updates using new geological, hydrogeological, and 373 geotechnical information (Klinc et al. 2021). Acknowledging the dynamic nature of ground 374 conditions throughout site investigation, design, and construction phases, automation with 375 seamless interoperability of data across various formats becomes imperative for developing a 376 high-fidelity DT of the underground space. 377 Currently, manual entry of paper-based borehole log reports into 2D/3D drafting tools is 378 common; subsurface geological cross-sections are subsequently developed by drawing lines or 379 splines to connect adjacent boreholes (Shi and Wang 2021a). Inherent uncertainties coupled 380 with heterogeneity of the ground strata make the modelling process highly challenging. 381 Erharter et al. (2023) proposed a framework for splitting the model into factual model, 382 geotechnical model, and geotechnical synthetic modelling. AGS (Association of Geotechnical 383 and Environmental Specialists) is a standard data format designed to enable information 384 exchange across different geotechnical software systems with data structures to store 385 19 information such as borehole data, laboratory, and field tests (AGS 2023). Geospatial data are 386 available in digital standards like Open Geospatial Consortium (OGC) or CityGML, facilitating 387 the exchange of geological map data, time scales, boreholes, and laboratory metadata (“OGC 388 Geoscience Markup Language (GeoSciML)” n.d.). OGC is extending schemas to integrate 389 geotechnical models to maintain a common concept for the geology/geotechnics domain that 390 will be implemented by both IFC and OGC standards in the future (Fonsati et al. 2023). Despite 391 various open formats, there is a lack of integration of the geotechnical/geological data with the 392 BIM system. Converting data between formats poses interoperability and data preservation 393 challenges (El Sibaii et al. 2022; Fonsati et al. 2023) causing reduced effectiveness of the 394 ground modelling process. While custom workflows (e.g. El Sibaii et al. 2022) have 395 endeavoured to address these issues by proposing standardisation through product data 396 templates, the need for data format standardisation across the industry still remains as a 397 challenge. 398 For seamless integration of the ground models with BIM, accurate 3D visualisation of the 399 resulting ground conditions is essential. Representation methods for 3D geological models fall 400 into two primary categories: bin-based representation (triangulated irregular network and 401 boundary representation) and voxel-based representation (Borrmann et al. 2022). A 402 comparative analysis of these representation types is outlined in Table 3, which aids in 403 understanding the strengths and limitations of each type based on various criteria. From the 404 literature, recent research attempting BIM-based ground modelling are summarised in Table 2 405 along with their main features. Various tasks necessitate distinct forms of model 406 representations. For example, surface-based models are best for visualisation and planning, 407 while volumetric models are needed to capture the detailed spatial variability and material 408 properties essential for accurate simulations and risk assessments. A comprehensive ground 409 data management system that integrates and updates multiple representations in real time is 410 crucial to address this issue, providing accurate and accessible information for everyone 411 involved (Hegemann 2015). While Table 3 highlights the benefits of adopting a voxel-based 412 representation, Table 4, which ranks ground modelling representations from the literature based 413 on visualisation and analytical methods, reveals gaps in research. Specifically, there is a need 414 for data-driven methods combined with voxel representation in BIM and for hybrid models 415 capable of merging and converting different methods. 416 417 20 Table 3: Comparison of various 3D representation methods 418 Feature / Model Representation TIN (Triangulated Irregular Network) BRep (Boundary Representation) Voxel (Volumetric Pixel) Type of model Surface-based Surface-based Volume-based Geometry Polygonal facets (triangles) Polygonal facets (various shapes) Cubes or rectangular blocks Complexity handling capacity High High Moderate (Resolution dependent) Ability to hold semantic data Data related to surface Volumetric data Volumetric data Computation Time Moderate High High (due to large data) Detail Level Surface detail only Surface detail only Internal detail (volumetric) Spatial Variability Capture Low Low High Interoperability IFC scheme available IFC scheme available Expected in IFC 4.0 Resolution Management Fixed by triangle size Fixed by facet size Adjustable (Through voxel size and hierarchical structure) Advantages Flexible, efficient for complex surfaces Accurate for boundary details Detailed internal properties, adaptable resolution Disadvantages Non-smooth approximation High complexity, high computation High data storage, complex merging, and processing 419 Table 4: Compilation of ground modelling representations from the literature, ranked based on 420 visualisation method adopted with corresponding analytical methods for modelling 421 Reference Visualisation method Analytical method Project Software (Che and Jia 2019) TIN Weighted kriging Qianjiaying Coal Mine, Tangshan, China Not disclosed in the article (Fabozzi et al. 2021) TIN Spatial interpolation Naples underground Line 1 and 6 Bentley OpenRail designer (Huang et al. 2022) TIN Spatial interpolation Not specified – Example demonstration Revit, Dynamo (Wang et al. 2022a) TIN Kriging Interpolation Liangshuijing Tunnel Civil 3D, Revi, Dynamo (Xie et al. 2023) TIN Ordinary Kriging, uncertainty quantification using Conditional Random Fields Not specified – Example demonstration Revit, Dynamo (Li et al. 2022) BRep Spatial interpolation Suki Kinari underground powerhouse caverns Catia (Haryono et al. 2022) BRep Propriety tool used for ground Not specified Leapfrog, Revit 21 modelling requiring manual input to BIM (Hung et al. 2022) Propriety tool is used for ground modelling which must manually be input to BIM models Taipei mass rapid transit project GMS (Groundwater Modelling System): (Hegemann et al. 2013; Koch et al. 2017) Hybrid between BRep and voxel Simple Kriging Wehrhahn-Linie subway tunnelling project, Germany X3dom, 3D CAD (Mahmoudi et al. 2021) Octree Voxel Ordinary Kriging with uncertainty quantification using Optimal experimental design Not specified – Example demonstration Revit, Dynamo (Khan et al. 2023) BRep model, converted to voxel Kriging Interpolation Demonstrated with Peshawar city data Revi, Dynamo, ArcGIS 422 4.3 Computational BIM 423 Computational BIM is a tool that integrates advanced computer processing with BIM, utilising 424 algorithms and parameters to automate design solutions through a user-friendly visual scripting 425 approach within the BIM environment (Wei et al. 2020). Parametric modelling is a process of 426 creating models whose geometry and characteristics can be altered or manipulated through the 427 adjustment of parameters (Edmonds et al. 2022). It serves as the fundamental building block 428 for computational design, forming the basis for subsets such as parametric design, generative 429 design, and algorithmic design, each offering different levels of user interaction (see Figure 430 10Error! Reference source not found.). 431 432 22 Figure 10: Subsets of computational design (incorporated from (Caetano et al. 2020)) 433 Generative design uses algorithms to automatically generate diverse design alternatives 434 iteratively based on predefined user input criteria with specific goals and penalty functions (Ma 435 et al. 2021). Algorithmic design is a subset of generative design which demonstrates 436 traceability of options, indicating an evident correlation between the algorithm and the resulting 437 design. Parametric design, on the other hand, involves defining geometry through parameters 438 and rules, enabling efficient modifications (Caetano et al. 2020). 439 Parametric design is increasingly used in the construction industry for design support, 440 automation, topology optimisation, design review and checking compliance to standards and 441 codes (Sacks et al. 2020b). Even in the field of geotechnical engineering, the application of 442 parametric modelling, is actively explored, particularly in research. For instance, Koch et al. 443 (2017) and Ninić et al. (2020) introduced a tunnel information modelling framework that 444 creates and connects the ground model, tunnel lining model, tunnel excavation model, and the 445 built environment. Díaz et al. (2021) demonstrated the use of generative design for automating 446 code-compliant design of a retaining wall. These approaches support efficient design and 447 process optimisation in conventional design, allowing for the exchange of information between 448 parametric analysis and numerical simulations with reduced computational effort 449 (Hedayatzadeh et al. 2024). Ninić et al. (2024) demonstrates a parametric modelling approach 450 for visualisation of ground settlements and building damage risk, enabled by integrating 451 empirical and analytical models using surrogate models. Computational BIM workflows which 452 (i) are integrated with geometric models, ground models and analytical models and, (ii) at the 453 same time, are able to modify those models based on real-time sensing data will be crucial for 454 future underground DT development. A lower level of user interaction, enabled by generative 455 design through BIM and multi-modal data, leads to the advancement of more mature predictive 456 and prescriptive DTs. 457 The use of Computational BIM in underground construction for the construction and 458 operational phases has been relatively underexplored in existing literature. Nevertheless, some 459 researchers have attempted to leverage computational BIM to integrate real-time sensor data 460 with BIM models, aiming to facilitate BIM-based structural health monitoring and 461 visualisation. For example, Chang et al. (2018) developed a platform that translates sensor data 462 into visual outputs using color-coded representations within BIM models. Similarly, 463 Valinejadshoubi et al. (2019) proposed a framework for transferring real-world sensor data into 464 BIM, allowing for the detection and visualisation of structural damage through colour coding, 465 23 utilising computational BIM tools. Davila Delgado et al. (2018) combined fibre optic sensor 466 data with BIM, employing color-coded visualisations and dynamic charts to depict structural 467 deformations. This integration was achieved using an external gaming engine software, which 468 communicated with the BIM model via computational BIM tools. Additionally, Boddupalli et 469 al. (2019) introduced a framework for handling large volumes of vibration sensor data from 470 bridges, connecting the processed data with sensor parameters within the BIM environment. 471 4.4 BIM-numerical modelling software interoperability 472 Geotechnical analysis is a crucial step in the design of underground structures, which involves 473 the use of ground models and ground constitutive models for realistic simulations using various 474 computational methods e.g. FEM (Wang and Tian 2023). While FEM pre-processing steps are 475 mostly streamlined and automated, challenges arise in integrating geometry into numerical 476 models, where interoperability issues persist between BIM and numerically modelling software 477 (Alsahly et al. 2020; Fabozzi et al. 2021; Giangiulio et al. 2023; Klinc et al. 2021; Lou et al. 478 2021). 479 The process of creating 3D geometry for numerical models involves importing information 480 through various file formats such as .dwg, .ifc, and cloud points. Manual operations are often 481 necessary for model definition, posing challenges for efficiency and accuracy (Wu et al. 482 2022a). Some researchers (Fabozzi et al. 2021; Huang et al. 2022; Ninić et al. 2019a; 483 Tschuchnigg and Lederhilger 2020) have attempted to automate these steps by adopting 484 advanced Python scripting with structured data inputs but these steps can be challenging to 485 execute. For tunnelling applications, Ninić et al. (2021) introduced an integrated framework, 486 employing a “BIM-to-FEM” approach through a custom platform (“SATBIM”), where 487 parametric modelling and numerical analysis were adopted to minimise user interaction and 488 real-time analysis capabilities, albeit requiring substantial programming effort. Huang et al. 489 (2022) proposed a BIM-to-finite differences method workflow for multi-LOD underground 490 metro stations, which enhances interoperability, automation, and error-free design-to-design 491 process to effectively explore design solutions and construction optimisat65ion. 492 The challenges of mesh dependency and need for remeshing in FEM while importing from 493 BIM can be addressed by adopting iso-geometric analysis framework proposed by Ninić et al. 494 (2020). With this approach higher order geometry in BIM can be directly utilised to create high 495 order computational models. Recent studies have demonstrated the reduction of computational 496 time by up to ten times (whilst maintaining accuracy) by adopting this approach (Bui et al. 497 24 2024). The choice of geometry representation significantly affects FEM mesh quality and 498 flexibility during creation. Transitioning from analytical to discretised geometry in FEM mesh 499 preparation must align with model requirements for specific analysis. However, pre-discretised 500 geometry, like the triangulated model in IFC, limits flexibility, potentially making the FEM 501 mesh unsuitable. Insufficient or excessive density in discretised geometry points can lead to 502 distortion or numerical challenges. Unsuitable geometry may also introduce issues during the 503 transfer of 3D elements to 2D or 1D. Various representations, such as explicit geometry with 504 faceted B- Rep, triangulated face sets, procedural geometry with a swept profile, and octree or 505 voxel- based volume geometry, pose challenges for discretisation (Eastman 2011; Huang et 506 al. 2022). 507 The disconnect between BIM and numerical analysis programs inevitably leads to repeated 508 manual adjustments, highlighting the need for interoperability to enhance cross-disciplinary 509 coordination and workflow efficiency. (Klinc et al. 2021). Several case studies have 510 highlighted the data integration problems resulting from inconsistent information types and 511 data formats (Fabozzi et al. 2019; Klinc et al. 2021). Whenever modifications occur, a new 512 model is created using disparate geometrical representation and parameterisation native to the 513 numerical modelling program instead of directly referring to the as-designed or as-built BIM 514 model. The temporal changes in the ground behaviour due to the stages of construction and the 515 impact of long-term effects like consolidation, creep, etc., must be analysed by simulating the 516 distinct phases of the project. These processes become cumbersome with manual iterations 517 (Giangiulio et al. 2023; Huang et al. 2022). Leveraging the IFC scheme to enable seamless 518 FEM-BIM integration, even in the case of 4D simulation, is a key requirement for underground 519 DTs which should be explored further (Li et al. 2020; Satyanaga et al. 2023). 520 4.5 Sensing and monitoring 521 Traditional drawbacks of sensing and monitoring in underground construction include a lack 522 of sensor robustness, time-consuming installation and post-processing, and discrete and noisy 523 measurements (Hong et al. 2022). In particular, fibre optic technology has enabled 524 unprecedented real-time distributed strain and temperature monitoring of large infrastructure 525 (e.g. Soga and Luo 2018; Suhail 2017) as well as novel approaches for the ground-structure 526 contact stress sensing (Templeman and Sheil 2024). Micro electromechanical system (MEMS) 527 is another promising high-precision, low-power wireless technology (e.g., for measuring 528 inclinations), and IoT integration enables smart sensing networks (Royston et al. 2022). 529 Scalability of measurement can also now be achieved using computer vision techniques when 530 25 coupled with laser scanning (e.g. LiDAR) and imaging (e.g. using drones) which can help 531 identify defects and deviations in structures (Huang et al. 2021; Lin et al. 2024; Romanovich 532 et al. 2021). In situations where expansive areas need monitoring, Interferometric Synthetic 533 Aperture Radar technology becomes particularly advantageous providing a holistic view of 534 ground deformation and subsidence with minimal on-site instrumentation (e.g., Bayaraa et al. 535 2023). Geophysical techniques like GPR are used to detect geological discontinuities and 536 assess defects such as voids and delamination in underground structures through dielectric 537 properties. Figure 11 describes a framework where the GPR scan data is integrated with BIM 538 models (Zhu et al. 2024). 539 540 Figure 11: BIM centred modelling by laser scanning and GPR scans reproduced by (Zhu et al. 2024) 541 Owing to the inherent non-linearity in ground-structure interaction, the fields of system 542 identification and inverse analysis are leveraging artificial neural networks and physics-543 informed machine learning (Jafari 2020; Ouyang et al. 2024). These approaches aim to 544 establish a digital nervous system for structures, facilitating accurate modelling of their 545 behaviour. These advancements pave the way for the development of underground DT with 546 optimised sensor utilisation. 547 5 LINKS BETWEEN OBSERVATIONAL METHOD AND DT CONCEPTS 548 The Observational Method (OM) in geotechnical engineering, introduced by Peck in 1969, is 549 a dynamic 'learn-as-you-go' process that spans design, construction control, monitoring, and 550 Geometry Scanning Geo-integrated BIM Model Geology Integration 0 .7 2 m 0 Interior Scanning Retrieve Internal geometry SegmentThickness t Project-scale tunnel model 1 2 3 • Internal diameter D • Ring configuration + = Tunnel fitout S egm ental lining S tation Tunnel S tation S ystem level: m etro netw ork + = A sset level: tunnel Tunnel M aterial level: lining C oncrete R einforcem ent E xcavation m achinery A 1 -A 3 P ro d u c tio n T ra n s p o rt A 4 A 5 C o n s tru c tio n A B C E n d U s e L ife c y c le m o d u le s E N 1 5 9 7 8 C ra d le -to -C o m p le tio n D B e y o n d + = Tunnel fitout Segmental lining Station Tunnel Station System level: metro network + = Asset level: tunnel Tunnel Material level: lining Concrete Reinforcement Excavation machinery A1-A3 Production Transport A4 A5 Construction A B C End Use Life cycle m odules EN 15978 Cradle-to-Completion D Beyond Step-frequency GPR recording 0 .3 8 m t Rebar 3D geological model W E FR SW HW Soil Borehole Quantity take-off Adaptive alignment and instantiation Geo M4 BIM 26 review, aiming to reduce uncertainties and ensure efficient, yet safe, designs (Spross 2014). 551 Application of OM can take two forms: 552 (1) “ab initio” or from the start: involves initiating OM from project inception with the 553 most probable design values and contingency measures for deviations. 554 (2) “Ipso tempore” or the best way out: involves adaptability to unforeseen challenges 555 during construction to prevent catastrophe (Spross and Johansson 2017). 556 Several case studies highlight the benefits of the OM in geotechnical engineering, including 557 improved construction control, enhanced safety, and collaboration between designers and 558 constructors (Nicholson 1999). OM application has demonstrated substantial savings, 559 especially in temporary works and construction method optimisation, impacting both 560 temporary and permanent structures (Powderham 2002). Notable projects like the Heathrow 561 airport terminal building and several stations of the Crossrail projects have exemplified the 562 advantages of OM (Chen et al. 2015; Gaba et al. 2017; Hardy et al. 2021; Powderham and 563 O’Brien 2021; Yeow et al. 2014). 564 In the context of design approaches, CIRIA C185 (Nicholson 1999) elucidates the principles 565 and applications of OM, thereby presenting an opportunity for design optimisation (Hardy et 566 al. 2018). Eurocode 7 (CEN 2004) includes OM as an accepted alternative verification method 567 for conventional design of geotechnical structures. The code emphasises an adaptive 568 geotechnical approach, with principles focusing on continuous monitoring, pre-construction 569 planning, and prompt contingency actions as necessary, ensuring compliance in OM 570 application (Spross 2014). 571 5.1 Recent advancements in OM 572 Advancements in sensing and computational technologies have propelled OM from basic 573 onsite observations to sophisticated instrumentation and computer-based back analysis 574 techniques. Variables of interest may be measured directly (e.g., displacements) or indirectly 575 (e.g., stresses) through back-analysis using mathematical models such as FEM. The evolving 576 state of the system as construction progresses adds complexity to the back-analysis. In this 577 context, a practical approach involves probabilistic design with a Bayesian perspective on 578 statistics (Huber 2016; Mohammadi and Parsapour 2024; Spross 2014). For example, recent 579 work by Jin et al. (2021) employed the Markov chain Monte Carlo technique to estimate an 580 improved set of (posterior) ground parameters given a prior assumption (original design model) 581 and new evidence (monitored data). Such techniques have also helped to overcome over-582 27 reliance on the monitored data by using prior distributions over model parameters. Sensitivity 583 analysis of monitored variables provides a rational approach to account for geotechnical 584 uncertainties. However, this process can be computationally expensive, especially when 585 sophisticated numerical models are involved with repeated simulations (e.g. Li et al. 2018). 586 To guide the decision on utility and safety of applying OM, Roper et al. (2024) introduced a 587 risk-based decision framework that uses expected utility theory, integrating risk, cost, 588 construction timelines, and engineering judgment within an economic decision model through 589 a probabilistic approach. Further advancements include the integration of ML with numerical 590 analysis (Mitelman et al. 2023). Considering the time and stage dependant behaviour of 591 underground structures Bismut et al. (2023) formalised a geotechnical problem as a sequential 592 decision problem and applied a heuristics-based method to finding optimal strategies. These 593 analytical techniques can be viewed as a simplified form of a preconstruction DT, establishing 594 a connection between the physical object and digital models. To become a true DT, models 595 must continuously and automatically update based on on-the-fly site data from the physical 596 object flowing into the virtual prototypes of the structure. 597 5.2 Challenges in the application of OM 598 While the OM presents several advantages for the construction industry, its broader adoption 599 faces many challenges; primary factors leading to stakeholder hesitancy include (a) the absence 600 of a standardised implementation code, (b) effort required for implementation, and (c) 601 misconceptions about heightened risk (Hardy et al. 2018), (d) lack of economic motivation due 602 to insufficient data/ quality of information in the early design stage. Traditional contractual 603 conditions, which often separate design and construction, impede a collaborative approach to 604 risk management (Powderham and O’Brien 2021). The iterative nature of the OM, especially 605 when addressing non-linear ground-structure interaction behaviours demands substantial time 606 and resources, and can be an obstacle to strict project timelines (Hardy et al. 2018; Powderham 607 and O’Brien 2021). Historical concerns regarding the accuracy and dependability of sensors 608 also fosters a perception of increased risk (Spross 2014). 609 5.3 Complementarities between OM and DT 610 Recent advancements in OM highlighted in section 5.1 have explored the integration of 611 advanced sensing technologies and ML. These technologies enhance data processing and 612 improve understanding of uncertainty in collected data through statistical analysis, which is 613 crucial for accurate multidisciplinary engineering evaluations. However, despite these 614 28 advancements, current practice still requires significant manual intervention for monitoring, 615 interpreting, and analysing site conditions, limiting the optimisation of design and construction. 616 Underground DTs offer a solution to these challenges by incorporating features that streamline 617 and automate many aspects of OM. Underground DTs provide capabilities such as real-time 618 data acquisition data integration, and advanced analytics and visualisation tools that enable 619 dynamic and continuous analysis of site conditions. These features reduce the need for manual 620 computations and facilitate the testing of various design scenarios, thereby optimising project 621 costs and mitigating risks. Furthermore, underground DTs address concerns regarding 622 measurement precision and instrumentation reliability through comprehensive sensitivity and 623 reliability analyses, which are essential for reassuring stakeholders and overcoming barriers to 624 the adoption of OM. 625 Table 5 outlines the synergies between the requirements of OM, as specified in Eurocode 7, 626 and the features of underground DTs. The table demonstrates how underground DTs can 627 enhance OM by supporting key activities such as establishing acceptable behaviour limits, 628 enabling continuous monitoring, and facilitating rapid contingency planning. Additionally, 629 underground DTs improve risk management and cost optimisation by addressing challenges 630 related to time constraints and the iterative nature of analysis. The integration of OM with 631 underground DTs can vary in levels of automation, ranging from descriptive twins that 632 primarily provide visualisation, to reflective twins that enable real-time model updates, and 633 further to advanced predictive and prescriptive twins. The latter more sophisticated 634 underground DTs can perform various types of analyses and offer recommendations, such as 635 design revisions or enhancements to monitoring protocols, thereby providing a more proactive 636 approach to managing construction projects. The full potential of advancements in OM 637 research can be realised by transitioning from conventional OM applications to integrated 638 predictive or prescriptive underground DTs. 639 640 641 642 643 644 645 29 Table 5: Synergy between requirements of OM and features of underground DT 646 MATURITY DIMENSIONS Prescriptive Predictive Reflective Descriptive FEATURES OF DT REQUIREMENTS AND CHALLENGES IN OM D at a in te g ra ti o n a n d m o d el li n g M u lt i- sc al e v is u al is at io n R ea l ti m e m o n it o ri n g d ig it al m o d el u p d at in g M u lt i- p h y si cs an al y si s S af et y A ss es sm en t R el ia b il it y & s en il it y an al y si s C o st o p ti m is at io n si m u la ti o n s R is k p ro g n o si s V ir tu al c o n tr o l an d g en er at iv e d es ig n R E Q U IR E M E N T S O F O M A S P E R E U R O C O D E 7 Establish acceptable limits of behaviour as triggers ✔ ✔ Assess range of possible behaviours ✔ ✔ Establish monitoring for frequent update to enable contingency plan ✔ ✔ ✔ ✔ Rapid monitoring to capture real time change ✔ ✔ ✔ ✔ Define at contingency actions when trigger values exceed ✔ ✔ ✔ ✔ ✔ Continuous monitoring during construction ✔ ✔ ✔ ✔ ✔ ✔ Assessment of monitoring to enable timely contingency action ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ Maintain reliability and quality of data ✔ ✔ ✔ ✔ ✔ ✔ ✔ C H A L L E N G E S I N O M A P P L IC A T IO N Insufficient time for design program ✔ ✔ ✔ ✔ ✔ ✔ ✔ Iterative recalibrated analysis process ✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔ Contractual issue- sharing of risks and opportunities ✔ ✔ ✔ ✔ Apprehensions of reliability and precision of instrumentation ✔ ✔ ✔ Apprehensions about safety of structure ✔ ✔ ✔ ✔ Despite the clear synergy, current research on OM has not considered application of DTs. A 647 co-occurrence map of keywords generated through a Scopus search of “observational method 648 in geotechnical engineering” lacks the keywords “digital twin” entirely, even though closely 649 related topics like computer simulations are evident (see Figure 12Error! Reference source 650 not found.). These results highlight the significant gap in the existing literature exploring DT 651 application in the context of OM. To foster widespread growth of DT in geotechnical 652 engineering, there is a need for increased collaboration, particularly those focused on machine 653 learning, field monitoring, and OM. 654 30 655 Figure 12: Co-occurrence map of OM literature demonstrating gaps in DT application (based on 219 656 selected articles from Scopus database with search criteria ( TITLE-ABS-KEY ( observational AND method 657 ) AND TITLE-ABS-KEY ( geotech* OR underground ) 658 6 VALUE OF UNDERGROUND DTS 659 The Gemini principles emphasise the importance of having a clear purpose and delivering value 660 to users when designing and developing DTs. Here, we identify some advantages of employing 661 underground DTs and present exemplar use-cases for underground DT services at various 662 lifecycle stages to demonstrate its value. 663 6.1 Reduction in uncertainty costs by breaking information silos 664 Underground construction projects typically involve several stakeholders throughout the 665 design and construction phases. The exchange of information between these stakeholders 666 typically undergoes numerous iterations and transformations, influenced by factors such as 667 design changes and conflicting requirements. These iterative processes often suffer from low-668 fidelity information flow and a delayed or reactive sharing of data among stakeholders, leading 669 to a significant loss of information across the value chain. These challenges are exacerbated by 670 the existence of information silos inherent in conventional methods, as illustrated in Figure 671 13a. For instance, during the planning stage, data are traditionally stored in tables, 2D drawings, 672 31 and physical paperwork. This is particularly the case for buried utilities, where their 673 identification and diversion are a key challenge in underground construction. During 674 construction, data are also generated by various machinery (e.g., piling rigs) which is 675 undervalued in current projects, despite its potential to optimise both design and construction 676 processes. The presence of silos in the system hinders the full utilisation of automated data 677 collection (Hoyme and Maurer 2021). These uncertainties inherent in the conventional 678 approach contributes to the conservatism embedded in underground construction. Whereas the 679 underground DT approach breaks the information silos enabling real-time exchange of data 680 and on-the-fly analysis/ simulations capabilities as shown in the Figure 13b. 681 Figure 14 highlights the differences in uncertainty costs between conventional, BIM-based, 682 and DT approaches, due to information flow across project phases. Conventional methods, with 683 incremental data collection, face delays in real-time sharing and substantial information loss at 684 the end of each phase, resulting in higher uncertainty costs, as shown at point ‘B’. On the 685 contrary, BIM-based methods improve information flow as described by Borrmann et al. 686 (2018) which involve sharing data among stakeholders, but real-time decisions are limited by 687 delayed data transfers. The dependence on subjective engineering judgment, coupled with 688 delayed and fragmented information, hinders timely evidence-based decisions. At each stage 689 there is a delay between discovery of findings and their communication to other stakeholders. 690 Most of the available information is also withheld until the end of each phase after the 691 information is transferred to the subsequent stakeholders in bulk. This bulk transfer of 692 information leads to an instantaneous drop in uncertainty cost at the end of each phase as 693 indicated as ‘C’ in the Figure. However, during the end-of-operations phase, the assessment of 694 reusability will involve various destructive and non-destructive tests which bring additional 695 uncertainty about the ground properties and structural condition, resulting in increased 696 uncertainty costs as shown in ‘D’. Another advantage of BIM process, marked by point ‘A’, is 697 that it retains information from previous projects, reducing uncertainty at the start, unlike 698 conventional methods that begin with limited prior knowledge. 699 The DT approach, unlike conventional and BIM methods, in addition to eliminating the 700 information silos through real-time data exchange, the analysis capabilities of the DT with real-701 time visualisation, machine learning, and analysis to reveal patterns and insights, enhancing 702 decision-making for downstream processes. This helps in significantly lowering uncertainty 703 costs throughout the project lifecycle (Li et al. 2024a). Continuous data flow supports timely 704 32 data collection and collaboration, enabling faster reductions in uncertainty compared to BIM. 705 For example, the data collected in the ground investigation or pile load test data dynamically 706 updates the ground models which helps to reduce the uncertainty in designs. Additionally, the 707 application of observational method for automated analysis of data in real-time is enabled by 708 the DT approach which is missing in the BIM approach. Due to which the uncertainty cost 709 reduces at a higher rate. Despite this, residual risks remain due to project complexity and 710 multiple stakeholder involvement. 711 (a) (b) Figure 13: Information flow between stakeholders in (a) conventional process; (b) DT process 712 33 713 Figure 14: Conceptual illustration of the uncertainty costs due to information silos in a conventional 714 and BIM approach compared to that enabled by underground DTs 715 6.2 Risk prognosis and safety assessment during construction 716 Underground construction faces several inherent risks which are exacerbated by uncertain 717 weather, geological factors, and the hydrogeological environment. Human factors, like 718 collisions with utilities during excavation, pose additional dangers which have led to several 719 historical accidents (Liu et al. 2023). Environmental factors, such as rainfall and temperature, 720 and factors such as uncontrolled dewatering (e.g. Zeng et al. 2023) can impact the safety of 721 underground construction significantly. 722 Given changing ground conditions during construction, risk analysis and mitigation requires 723 timely acquisition of spatiotemporal risk information for effective risk prediction. Current field 724 practices rely on visual measurements and inspections, which are time-consuming and demand 725 extensive field experience. FEM models based only on initial design documents often fall short 726 in accurately representing as-built conditions during construction if not dynamically updated. 727 Therefore, updating models using diverse data is crucial to establish a high-fidelity virtual 728 model, enabling the examination of various unsafe scenarios through simulation approaches. 729 A virtual model with geometric, physical, and behavioural components, integrating 730 multidisciplinary knowledge, aids risk management. 731 34 Several case studies illustrate practical applications of intelligent foundation engineering and 732 risk prognosis of deep excavations using a DT approach. Sun et al. (2023) demonstrated the 733 application of a DT in a case study of Wuhan Metro Line 7, by capturing and processing diverse 734 information in the physical space through advanced sensing techniques and data processing 735 algorithms. Data analysis tools were developed to inform prognosis and control of unsafe 736 events during foundation pit excavation, allowing for risk assessment and introducing control 737 decisions for enhanced safety in the physical environment. Cao et al. (2022) assessed the risk 738 of tunnelling-induced building damage by combining FEM of tunnelling with building models, 739 accounting for ground-structure interaction, and using artificial neural networks for real-time 740 damage prediction. Liu et al. (2023) developed a computer vision-based approach for avoiding 741 accidents in excavation by analysing the position of excavators. By integrating DT, deep 742 learning and mixed reality technologies, Wu et al. (2022b) developed a real-time visual alarm 743 system that enables construction workers to proactively judge their safety status and avoid 744 accidents. 745 6.3 Time, cost and carbon optimisation of underground construction 746 Underground DTs have the potential to transform underground construction by optimising both 747 design and construction processes. Their key advantage lies in bringing diverse sources of data 748 from different domains, such as the environment, man-made structures, and archaeological 749 artefacts, into a unified framework using advanced analytics, numerical modelling, and ML to 750 perform real-time simulations and analyses. This enables continuous model refinement and 751 real-time design optimisation, providing a more holistic understanding of complex interactions 752 and facilitating multi-disciplinary optimisation of design and maintenance measures through 753 the simultaneous evaluation of different criteria and physics. For instance, Xie et al. (2024) 754 introduced a BIM-based multi-model framework designed to incorporate multiple LOD for 755 TBM machinery, 3D geological models, and numerical analysis models, enabling real-time 756 thrust calculations and predictive analysis of shield tunnelling. In terms of ground-structure 757 interaction (GSI), such as the pile installation, ground resistance and applied torque can be 758 transmitted to the underground DT, allowing automatic updates to the ground model. This 759 dynamic modelling enhances understanding of ground conditions, leading to more accurate 760 predictions of ground behaviour, and reducing the inherent uncertainty in geotechnical 761 engineering. As Randolph (2003) observed, even with advanced models, estimating axial pile 762 capacity can be challenging, often within ±30%. But with underground DTs this uncertainty 763 can be reduced significantly, leading to safer and more cost-effective designs. Databases like 764 35 the DINGO project (Voyagaki et al. 2022) further improve statistical assessments of pile 765 performance. By simulating design scenarios using real-time data, underground DTs create 766 site-specific designs. Continuous response from the structures under varying loads informs 767 real-time adjustments, reducing uncertainty and aligning designs with actual conditions, 768 enhancing safety, and cutting costs. Moreover, underground DTs use historical and real-time 769 data to predict potential issues, such as faults in underground excavation and structures 770 installation before they escalate. ML models refine GSI predictions, minimising risks and 771 preventing delays. Future underground DT developments could integrate generative design, 772 allowing engineers to explore various scenarios and optimise foundations to site-specific 773 conditions, reducing on-site adjustments and costs. Additionally, underground DTs could 774 incorporate modular and prefabricated design elements, streamlining the construction process 775 and further cutting time and costs. Huang et al. (2023) integrated BIM to evaluate carbon 776 emissions and construction feasibility, including GSI assessment and the assembly process, in 777 prefabricated stations. Chen et al. (2024) expanded this digitalisation framework by 778 incorporating carbon assessment standards, numerical modelling, and optimisation techniques 779 to assess carbon emissions and establish benchmarks for TBM tunnel construction products 780 and processes. In tunnelling operations, underground space DTs can help in automated steering 781 of TBM operational parameters based on the data and models to minimise the environmental 782 impact. 783 6.4 Lifetime monitoring of underground structures and assets 784 Current maintenance practices for underground spaces and assets are predominantly reactive, 785 with excavation in urban areas often facing risks such as pipeline bursts, explosions, and safety 786 hazards. In the UK alone, accidental strikes on underground pipes and cables are estimated to 787 cost approximately £1.2 billion annually (Wang and Yin 2022). The adoption of DTs for 788 underground spaces can greatly enhance the value of these assets by delivering essential 789 information and services. In addition to adding value, continuous monitoring of foundation 790 structures is crucial but presents significant challenges, as these structures are rarely 791 replaceable and key areas are often inaccessible for visual inspection (Bhalla et al. 2005). 792 Underground structures pose unique challenges, including changing earth pressures, 793 construction-induced ground movements, underground formations, and fluctuations in 794 groundwater. Predicting the long-term behaviours of these structures during the design stages 795 is challenging due to the wide variety of surrounding geologic conditions and non-linear ground 796 properties (Dutta and Kurup 2018). As the ground around the structure undergoes temporal 797 36 changes, such as consolidation, the corresponding loads applied to the underground 798 engineering structure also change significantly. Moreover, the service life and safety of 799 underground structures are impacted by changes in geometric and material properties due to 800 operational incidents and the complexity of underground environmental conditions (Abbas et 801 al. 2023). Additionally, increased congestion in underground spaces means that the 802 construction of new underground structures can distort or damage existing underground 803 structures due to changing stresses in the ground (Yu and Geng 2019) and complex two-way 804 relationships (Wan et al. 2023). 805 Recent efforts to address these issues include sophisticated numerical models for mechanised 806 tunnelling and ground-structure interaction (Boldini et al. 2018; Ninić et al. 2014; Yiu et al. 807 2017). In addition, geophysics techniques like GPR are adopted to understand anamolies such 808 as voids amd water leakage from burried utilities (Zhu et al. 2024). However, advanced 3D 809 computational models often involve extensive detail and prolonged computation times which 810 prompts parallelisation strategies for high performance computing (Ninić et al. 2019b). 811 Furthermore, the BIM framework has gained prominence in large infrastructural projects, 812 serving as a robust tool for information management, processing, visualisation, and analysis 813 across the project lifecycle. This is especially beneficial during the early design phases of 814 intricate multidisciplinary systems. 815 In existing monitoring systems, sparsely distributed point measurements are gathered around 816 structures to assess whether the recorded displacements surpass predetermined thresholds 817 established during design. Despite the widespread use of monitoring, concerns persist about 818 the effectiveness of conventional techniques due to the limited number of monitoring points 819 for each structure hindering accurate evaluation of actual building deformations. Moreover, the 820 reliability of damage inference using monitored displacement data and equivalent-beam 821 models is constrained due to data sparsity. Additionally, GSI response is influenced by the 822 extent of pre-existing cracking and distortions in the supported superstructure (Acikgoz et al. 823 2022). Ninić et al. (2024) proposed a BIM-based real-time prediction of non-linear structural 824 response using meta models for prediction of complex phenomena which includes assessment 825 of structural response in BIM. These recent developments coupled with lifetime monitored data 826 of the underground structure, will enable life-time monitoring using underground DT. 827 37 6.5 Reuse of underground structures 828 The concept of reusing old foundation structure is growing in popularity for its programme, 829 material, carbon, and construction cost savings. However, a major obstacle for reuse is the 830 uncertainty and risk associated with the performance of the old foundations (Chapman et al. 831 2008) even though research has shown that ageing can improve both foundation stiffness and 832 capacity (Sheil 2017). Non- availability of records with high reliability, indicating various key 833 details such as the foundation location, sizes, capacities, integrity, and structural details is 834 another challenge for evaluating the feasibility of reuse (Tayler 2020). Unknowns, such as 835 unidentified construction defects and any deterioration that may have occurred after 836 construction, also increases risk (Chapman et al. 2006). Assessing foundation reuse feasibility 837 involves collaborative decision-making between geotechnical engineers, structural engineers, 838 architects, construction engineers, and other stakeholders based on the analysis of the previous 839 records, estimates of the remaining service life of the foundations, prediction of settlements 840 and various design options proposed for the new superstructure. 841 The current method of evaluating the feasibility of reuse involves unreliable estimates of the 842 various parameters above, which leads to overly conservative designs to account for residual 843 uncertainties and risks. A DT of the basement structure will capture the system's data through 844 its entire lifecycle and provide real-time updates about the ground-structure interaction and 845 structural health of the structure, which are critical for the decision-making. This will be a novel 846 solution to address the risks associated with basement/ underground construction reuse. The 847 rich data will also help in predicting the behaviours of the structure and surrounding ground 848 for various current and future design scenarios. 849 7 LIMITATIONS OF THE STUDY 850 The components of the underground DT architecture presented in Figure 4 are not exhaustive. 851 For example, the physical entities and processes layer could encompass additional elements, 852 such as structural components from neighbouring constructions. Similarly, the application 853 layer could offer a wider range of features and analytical capabilities. The conceptualisation of 854 the digital twin architecture in this study focuses on features and applications that are 855 commonly adopted in current industry practices. However, DTs can reach higher levels of 856 maturity, such as autonomous twins (beyond prescriptive twins), in industries like 857 manufacturing and aerospace. These advanced twins are enabled by automated decision-858 making and actuation. Since DTs for underground space are still in their infancy, this study 859 does not cover specific use cases of these advanced features. For instance, an example of 860 38 autonomous digital twins in underground construction could be autonomous excavators that 861 operate based on design models and real-time feedback from computer vision data. While such 862 developments are progressing in the mining industry, they are not explored here in the context 863 of urban underground construction. Additionally, this study focuses only on a few exemplary 864 high-level use cases of digital twins to demonstrate their immediate value. Although there are 865 numerous emerging areas influencing the development of DTs—such as research on data 866 fusion and the application of blockchain technology for secure information management ‒ this 867 study prioritises the most pertinent areas identified in the literature from underground space 868 and construction. 869 8 CONCLUSIONS 870 The expansion of underground construction presents a myriad of challenges, involving low 871 productivity, prolonged construction timelines, heightened costs, safety concerns, and 872 uncertainties in ground conditions. These obstacles collectively impede the overall efficiency 873 of underground construction processes. However, DTs emerge as a promising solution, 874 offering potential benefits such as real-time monitoring and visualisation, improved 875 collaboration, intelligent operation throughout the lifecycle of the structure. 876 In the current landscape, there exists a considerable degree of ambiguity surrounding the 877 concept of DTs, which vary depending on the context and the sector. Given the intricate 878 challenges inherent to underground spaces, a distinct definition and maturity framework for 879 underground DT are imperative. This definition should also identify the specific use cases of 880 underground DTs. Such clarity facilitates the definition, management, integration, and 881 optimisation of underground DTs, enabling both academic and practical benchmarking of 882 projects and technology systems. Establishing clear definitions and identifying features and 883 derived value, enables benchmarking with other sectors allowing the geotechnical community 884 to draw inspiration from diverse developments. 885 With the context of underground DTs, a spectrum of possibilities and variations are possible 886 with varying dimension of maturity ranging from basic descriptive twins to sophisticated 887 prescriptive twins. Each dimension is further capable of differing levels of advancement based 888 on specific applications and expected value. A layered architecture serves as a generic 889 framework for building an underground DT. The maturity of a particular underground DT 890 hinges on factors like model quality and completeness, stakeholder interaction, fidelity and 891 39 automation levels, and federation degree, ranging from basic prescriptive twins to advanced 892 iterations. 893 Research in underground space has explored areas such as machine learning, BIM-FEM 894 interoperability, and GIS integration, revealing applications such as data-driven ground 895 modelling, predictive ground behaviour modelling, and real-time monitoring of geotechnical 896 structures. Despite their potential to enhance accuracy, efficiency, and safety, practical 897 implementation of these advancement in real projects faces challenges due to technical 898 complexities and scepticism with black-box application of technologies like ML. Addressing 899 these issues entails providing researchers access to real-world data to refine models and 900 integrating disparate technology developments through a DT approach, which mitigates 901 stakeholder interaction and process silos. 902 Although research and technology development are ongoing, the adoption of digital twins for 903 underground spaces faces significant management and regulatory barriers due to low 904 technology readiness levels. Key challenges include complex stakeholder coordination, high 905 upfront and maintenance costs, data integration issues, and a lack of specialised skills. 906 Additionally, barriers such as data standardisation, privacy, and security concerns must be 907 systematically addressed. 908 By integrating the underground space DTs in readily adaptable solutions like OM, the benefit 909 of this technology is realised by the broader geotechnical community while the above obstacles 910 will be gradually overcome. Currently, OM involves frequent monitoring and model updates 911 based on trigger values, functioning as a basic descriptive twin. By focusing ongoing OM 912 research on advanced instrumentation and computer-based back analysis within an integrated 913 DT approach, the practice can evolve into a more mature, prescriptive, knowledge-based DT 914 system. To fully leverage UGDTs' potential across all lifecycle stages of underground 915 structures, professional bodies and the geotechnical community should adopt a holistic 916 approach to integrating other technological advancements with DTs. 917 9 ACKNOWLEDGEMENT 918 The first author gratefully acknowledges the UK Engineering and Physical Sciences Research 919 Council (EPSRC) for funding this research through the EPSRC Centre for Doctoral Training 920 in Future Infrastructure and Built Environment: Resilience in a Changing World (EPSRC grant 921 reference number EP/S02302X/1). 922 40 10 DECLARATION OF GENERATIVE AI AND AI-ASSISTED TECHNOLOGIES IN 923 THE WRITING PROCESS 924 During the preparation of this work, the first author used Grammarly and ChatGPT to review 925 grammar and proofread certain sections of the article. After utilising these tools, the author 926 reviewed and edited the content as needed, and takes full responsibility for the final content of 927 the publication. 928 11 REFERENCES 929 Abbas, N., T. Umar, R. Salih, M. Akbar, Z. Hussain, and X. Haibei. 2023. “Structural Health 930 Monitoring of Underground Metro Tunnel by Identifying Damage Using ANN Deep 931 Learning Auto-Encoder.” Appl. Sci., 13 (3): 1332. 932 https://doi.org/10.3390/app13031332. 933 Acikgoz, S., A. Luciano, M. Dewhirst, M. J. Dejong, and R. Mair. 2022. “Innovative 934 monitoring of the response of a heritage masonry building to nearby tunnelling in 935 London Clay.” Géotechnique, 72 (3): 200–215. https://doi.org/10.1680/jgeot.19.P.243. 936 Agrawal, A., R. Thiel, P. Jain, V. Singh, and M. Fischer. 2023. “Digital Twin: Where do 937 humans fit in?” Autom. Constr., 148: 104749. 938 https://doi.org/10.1016/j.autcon.2023.104749. 939 AGS. 2023. “AGS – Association of Geotechnical and Geoenvironmental Specialists – Working 940 Groups – Data Management.” Accessed December 10, 2023. 941 https://www.ags.org.uk/person/phil-wade/. 942 AIAA. 2020. Digital twin: Definition & Value. Aerospace Industries Association. 943 Alsahly, A., F. Hegemann, M. König, and G. Meschke. 2020. “Integrated BIM‐to‐FEM 944 approach in mechanised tunnelling.” Geomech. Tunn., 13 (2): 212–220. 945 https://doi.org/10.1002/geot.202000002. 946 Amthiou, H., M. Arioua, and T. Benbarrad. 2023. “Digital Twins in Industry 4.0: A Literature 947 Review.” ITM Web Conf., (Y. Mejdoub and A. Elamri, eds.), 52: 01002. 948 https://doi.org/10.1051/itmconf/20235201002. 949 Angjeliu, G., D. Coronelli, and G. Cardani. 2020. “Development of the simulation model for 950 Digital Twin applications in historical masonry buildings: The integration between 951 numerical and experimental reality.” Comput. Struct., 238: 106282. 952 https://doi.org/10.1016/j.compstruc.2020.106282. 953 Bayaraa, M., C. Rossi, F. Kalaitzis, and B. Sheil. 2023. “Entity Embeddings in Remote 954 Sensing: Application to Deformation Monitoring for Infrastructure.” Remote Sens., 15 955 (20): 4910. https://doi.org/10.3390/rs15204910. 956 Berdigylyjov, M., and H. Popa. 2019. “The implementation and role of geotechnical data in 957 BIM process.” E3S Web Conf., (M. C. Balan, F. Bode, C. Croitoru, A. Dogeanu, A. 958 Georgescu, C. Georgescu, I. Nastase, and M. Sandu, eds.), 85: 08009. 959 https://doi.org/10.1051/e3sconf/20198508009. 960 Bhalla, S., Y. W. Yang, J. Zhao, and C. K. Soh. 2005. “Structural health monitoring of 961 underground facilities – Technological issues and challenges.” Tunn. Undergr. Space 962 Technol., 20 (5): 487–500. https://doi.org/10.1016/j.tust.2005.03.003. 963 Bismut, E., D. Cotoarbă, J. Spross, and D. Straub. 2023. “Optimal adaptive decision rules in 964 geotechnical construction considering uncertainty.” Géotechnique, 1–12. 965 https://doi.org/10.1680/jgeot.22.00408. 966 41 Boddupalli, C., A. Sadhu, E. Rezazadeh Azar, and S. Pattyson. 2019. “Improved visualization 967 of infrastructure monitoring data using building information modeling.” Struct. 968 Infrastruct. Eng., 15 (9): 1247–1263. https://doi.org/10.1080/15732479.2019.1602150. 969 Boje, C., A. Guerriero, S. Kubicki, and Y. Rezgui. 2020. “Towards a semantic Construction 970 Digital Twin: Directions for future research.” Autom. Constr., 114: 103179. 971 https://doi.org/10.1016/j.autcon.2020.103179. 972 Boldini, D., N. Losacco, S. Bertolin, and A. Amorosi. 2018. “Finite Element modelling of 973 tunnelling-induced displacements on framed structures.” Tunn. Undergr. Space 974 Technol., 80: 222–231. https://doi.org/10.1016/j.tust.2018.06.019. 975 Bolton, A., L. Butler, I. Dabson, M. Enzer, M. Evans, T. Fenemore, F. Harradence, E. Keaney, 976 A. Kemp, A. Luck, N. Pawsey, S. Saville, J. Schooling, M. Sharp, T. Smith, J. 977 Tennison, J. Whyte, A. Wilson, and C. Makri. 2018. Gemini Principles. CDBB. 978 Borrmann, A., M. König, C. Koch, and J. Beetz. 2018. “Building Information Modeling: Why? 979 What? How?” Build. Inf. Model., A. Borrmann, M. König, C. Koch, and J. Beetz, eds., 980 1–24. Cham: Springer International Publishing. 981 Borrmann, A., M. Rives, S. Muhic, L. Wikström, and J. Weil. 2022. “The IFC-Tunnel Project 982 – Extending the IFC Standard to Enable High-Quality Exchange of Tunnel Information 983 Models.” Adv. Inf. Technol. Civ. Build. Eng., Lecture Notes in Civil Engineering, S. 984 Skatulla and H. Beushausen, eds., 487–516. Cham: Springer International Publishing. 985 Bui, H.-G., J. Ninić, C. Koch, K. Hackl, and G. Meschke. 2024. “Integrated BIM-based 986 modeling and simulation of segmental tunnel lining by means of isogeometric 987 analysis.” Finite Elem. Anal. Des., 229: 104070. 988 https://doi.org/10.1016/j.finel.2023.104070. 989 Caetano, I., L. Santos, and A. Leitão. 2020. “Computational design in architecture: Defining 990 parametric, generative, and algorithmic design.” Front. Archit. Res., 9 (2): 287–300. 991 https://doi.org/10.1016/j.foar.2019.12.008. 992 Camposano, J. C., K. Smolander, and T. Ruippo. 2021. “Seven Metaphors to Understand 993 Digital Twins of Built Assets.” IEEE Access, 9: 27167–27181. 994 https://doi.org/10.1109/ACCESS.2021.3058009. 995 Cao, B. T., M. Obel, S. Freitag, L. Heußner, G. Meschke, and P. Mark. 2022. “Real-Time Risk 996 Assessment of Tunneling-Induced Building Damage Considering Polymorphic 997 Uncertainty.” ASCE-ASME J. Risk Uncertain. Eng. Syst. Part Civ. Eng., 8 (1): 998 04021069. https://doi.org/10.1061/AJRUA6.0001192. 999 Chang, K.-M., R.-J. Dzeng, and Y.-J. Wu. 2018. “An Automated IoT Visualization BIM 1000 Platform for Decision Support in Facilities Management.” Appl. Sci., 8 (7): 1086. 1001 https://doi.org/10.3390/app8071086. 1002 Chapman, T., M. Bryan, and F. Andy. 2006. “Foundations for the future.” Proc. ICE, 107–109. 1003 Chapman, T., A. S, and W. J. 2008. Reuse of foundations (C653). CIRIA. 1004 Che, D., and Q. Jia. 2019. “Three-Dimensional Geological Modeling of Coal Seams Using 1005 Weighted Kriging Method and Multi-Source Data.” IEEE Access, 7: 118037–118045. 1006 https://doi.org/10.1109/ACCESS.2019.2936811. 1007 Chen, X., M. Huang, Y. Bai, and Q.-B. Zhang. 2024. “Sustainability of underground 1008 infrastructure – Part 1: Digitalisation-based carbon assessment and baseline for TBM 1009 tunnelling.” Tunn. Undergr. Space Technol., 148: 105776. 1010 https://doi.org/10.1016/j.tust.2024.105776. 1011 Chen, Y., D. Nicholson, P. Ingram, S. Hardy, H. L. Liew, I. Farooq, and G. Biscontin. 2015. 1012 “Application of the Observational Method on Crossrail projects.” Crossrail Spec. 1013 Found. Constr. Lessons Learnt Conf., 63–72. 1014 Davila Delgado, J. M., L. J. Butler, I. Brilakis, M. Z. E. B. Elshafie, and C. R. Middleton. 2018. 1015 “Structural Performance Monitoring Using a Dynamic Data-Driven BIM 1016 42 Environment.” J. Comput. Civ. Eng., 32 (3): 04018009. 1017 https://doi.org/10.1061/(ASCE)CP.1943-5487.0000749. 1018 Díaz, G., R. F. Herrera, F. Muñoz-La Rivera, and E. Atencio. 2021. “Generative Design for 1019 Dimensioning of Retaining Walls.” Mathematics, 9 (16): 1918. 1020 https://doi.org/10.3390/math9161918. 1021 Dutta, S., and P. Kurup. 2018. “Fiber Optic Sensing Technologies for Structural Health 1022 Monitoring of Underground Infrastructure.” North Am. Soc. Trenchless Technol. 1023 NASTT. 1024 Eastman, C. M. (Ed.). 2011. BIM handbook: a guide to building information modeling for 1025 owners, managers, designers, engineers and contractors. Hoboken, NJ: Wiley. 1026 Edmonds, A., T. Mourtis, and M. Boyle. 2022. “Parametric Design—A Drive Towards a 1027 Sustainable Future.” Innov. Constr. Springer. 1028 El Sibaii, M., J. Granja, L. Bidarra, and M. Azenha. 2022. “Towards efficient BIM use of 1029 geotechnical data from geotechnical investigations.” J. Inf. Technol. Constr., 27: 393–1030 415. https://doi.org/10.36680/j.itcon.2022.019. 1031 Erharter, G. H., J. Weil, L. Bacher, F. Heil, and P. Kompolschek. 2023. “Building information 1032 modelling based ground modelling for tunnel projects – Tunnel Angath/Austria.” Tunn. 1033 Undergr. Space Technol., 135: 105039. https://doi.org/10.1016/j.tust.2023.105039. 1034 Fabozzi, S., S. A. Biancardo, R. Veropalumbo, and E. Bilotta. 2021. “I-BIM based approach 1035 for geotechnical and numerical modelling of a conventional tunnel excavation.” Tunn. 1036 Undergr. Space Technol., 108: 103723. https://doi.org/10.1016/j.tust.2020.103723. 1037 Fabozzi, S., G. Cipolletta, and E. Capano. 2019. “BIM-FEM interoperability for the modelling 1038 of a traditional excavated tunnel.” Tunn. Undergr. Cities Eng. Innov. Meet Archaeol. 1039 Archit. Art. Taylor & Francis Group. 1040 Fonsati, A., R. M. Cosentini, C. Tundo, and A. Osello. 2023. “From Geotechnical Data to 1041 GeoBIM Models: Testing Strategies for an Ex-Industrial Site in Turin.” Buildings, 13 1042 (9): 2343. https://doi.org/10.3390/buildings13092343. 1043 Fuller, A., Z. Fan, C. Day, and C. Barlow. 2020. “Digital Twin: Enabling Technologies, 1044 Challenges and Open Research.” IEEE Access, 8: 108952–108971. 1045 https://doi.org/10.1109/ACCESS.2020.2998358. 1046 Gaba, A., S. Hardy, L. Doughty, W. Powrie, and D. Selemetas. 2017. Guidance on embedded 1047 retaining wall design (C760). CIRIA C 760. London: CIRIA. 1048 Gerhard, D., M. Neges, and M. Wolf. 2020. “Transfer of digital twin concepts to the production 1049 of precast concrete parts in the construction industry.” Z. Für Wirtsch. Fabr., 115 (s1): 1050 58–61. https://doi.org/10.3139/104.112313. 1051 Giangiulio, M., A.-N. Granitzer, F. Tschuchnigg, and J. Hoffmann. 2023. “BIM-to-FEM: 1052 Development of a Software Tool to Increase the Operational Efficiency of Dam 1053 Construction Projects.” Trends Constr. Digit. Era, Lecture Notes in Civil Engineering, 1054 A. Gomes Correia, M. Azenha, P. J. S. Cruz, P. Novais, and P. Pereira, eds., 182–195. 1055 Cham: Springer International Publishing. 1056 Grieves, M. 2002. “Completing the cycle: Using plm information in the sales and service 1057 functions [slides].” SME Manag. Forum. 1058 Griffiths, J. S. 2016. “Incorporating geomorphology in engineering geological ground models.” 1059 Geol. Soc. Lond. Eng. Geol. Spec. Publ., 27 (1): 159–168. 1060 https://doi.org/10.1144/EGSP27.14. 1061 Gürdür Broo, D., M. Bravo-Haro, and J. Schooling. 2022. “Design and implementation of a 1062 smart infrastructure digital twin.” Autom. Constr., 136: 104171. 1063 https://doi.org/10.1016/j.autcon.2022.104171. 1064 43 Hardy, S., K. Carter, and D. C. Patel. 2021. “Victoria & Albert Museum: Ground Movements 1065 Due to Pile Installation and Basement Excavation for the Exhibition Road Quarter.” 1066 Piling 2020, 351–356. ICE Publishing. 1067 Hardy, S., D. Nicholson, P. Ingram, A. Gaba, Y. Chen, and G. Biscontin. 2018. “New 1068 observational method framework for embedded walls.” Geotech. Res., 5 (3): 122–129. 1069 https://doi.org/10.1680/jgere.18.00013. 1070 Haryono, .S., A. Saw, S. Lee, and L. Wong. 2022. “Digital Twin for Geotechnical Engineering 1071 Applications.” HKIE Geotech. Div. 41st Annu. Semin. 1072 Hedayatzadeh, M., J. Rostami, V. Sarhosis, M. Nematollahi, N. Hajiantilaki, and A. 1073 Shafiepour. 2024. “Use of GIS and BIM for the integration of tunnel design and 1074 construction process in conventional tunneling.” Undergr. Space, 16: 261–278. 1075 https://doi.org/10.1016/j.undsp.2023.10.009. 1076 Hegemann, F. 2015. “A hybrid ground data management concept for tunneling projects.” 1077 Doctor of Engineering. Ruhr-University at Bochum. 1078 Hegemann, F., P. Manickam, K. Lehner, C. Koch, and M. König. 2013. “Hybrid Ground Data 1079 Model for Interacting Simulations in Mechanized Tunneling.” J. Comput. Civ. Eng., 27 1080 (6): 708–718. https://doi.org/10.1061/(ASCE)CP.1943-5487.0000291. 1081 Hong, C., J. Zhang, and W. Chen. 2022. “An Integrated Intelligent Approach for Monitoring 1082 and Management of a Deep Foundation Pit in a Subway Station.” Sensors, 22 (22): 1083 8737. https://doi.org/10.3390/s22228737. 1084 Hoyme, C., and J. Maurer. 2021. “The Need of Connecting Information on Special Foundation 1085 Projects.” Piling 2020, 549–553. ICE Publishing. 1086 Hu, F., X. Qiu, G. Jing, J. Tang, and Y. Zhu. 2023. “Digital twin-based decision making 1087 paradigm of raise boring method.” J. Intell. Manuf., 34 (5): 2387–2405. 1088 https://doi.org/10.1007/s10845-022-01941-0. 1089 Huang, M. Q., X. L. Chen, J. Ninić, Y. Bai, and Q. B. Zhang. 2023a. “A framework for 1090 integrating embodied carbon assessment and construction feasibility in prefabricated 1091 stations.” Tunn. Undergr. Space Technol., 132: 104920. 1092 https://doi.org/10.1016/j.tust.2022.104920. 1093 Huang, M. Q., X. L. Chen, J. Ninić, Y. Bai, and Q. B. Zhang. 2023b. “A framework for 1094 integrating embodied carbon assessment and construction feasibility in prefabricated 1095 stations.” Tunn. Undergr. Space Technol., 132: 104920. 1096 https://doi.org/10.1016/j.tust.2022.104920. 1097 Huang, M. Q., J. Ninić, and Q. B. Zhang. 2021. “BIM, machine learning and computer vision 1098 techniques in underground construction: Current status and future perspectives.” Tunn. 1099 Undergr. Space Technol., 108: 103677. https://doi.org/10.1016/j.tust.2020.103677. 1100 Huang, M. Q., H. M. Zhu, J. Ninić, and Q. B. Zhang. 2022. “Multi-LOD BIM for underground 1101 metro station: Interoperability and design-to-design enhancement.” Tunn. Undergr. 1102 Space Technol., 119: 104232. https://doi.org/10.1016/j.tust.2021.104232. 1103 Huang, Y., A. Ghelmani, and A. Hammad. 2023c. “Future research directions of construction 1104 digital twins.” Proc. 2023 Eur. Conf. Comput. Constr. 40th Int. CIB W78 Conf. 1105 https://doi.org/10.35490/EC3.2023.320. 1106 Huber, M. 2016. “Reducing forecast uncertainty by using observations in geotechnical 1107 engineering.” Probabilistic Eng. Mech., 45: 212–219. 1108 https://doi.org/10.1016/j.probengmech.2016.02.002. 1109 Hung, H. C., Y. Y. Liu, J. F. Chang, and C. R. Chou. 2022. “Application of BIM in Deep 1110 Excavation Projects.” HKIE Geotech. Div. 41st Annu. Semin., 100–112. 1111 Jafari, M. 2020. “System Identification of a Soil Tunnel Based on a Hybrid Artificial Neural 1112 Network–Numerical Model Approach.” Iran. J. Sci. Technol. Trans. Civ. Eng., 44 (3): 1113 889–899. https://doi.org/10.1007/s40996-020-00405-w. 1114 44 Jiang, F., L. Ma, T. Broyd, and K. Chen. 2021. “Digital twin and its implementations in the 1115 civil engineering sector.” Autom. Constr., 130: 103838. 1116 https://doi.org/10.1016/j.autcon.2021.103838. 1117 Jin, Y., G. Biscontin, and P. Gardoni. 2021. “Adaptive prediction of wall movement during 1118 excavation using Bayesian inference.” Comput. Geotech., 137: 104249. 1119 https://doi.org/10.1016/j.compgeo.2021.104249. 1120 Kaewunruen, S., and Q. Lian. 2019. “Digital twin aided sustainability-based lifecycle 1121 management for railway turnout systems.” J. Clean. Prod., 228: 1537–1551. 1122 https://doi.org/10.1016/j.jclepro.2019.04.156. 1123 Khajavi, S. H., N. H. Motlagh, A. Jaribion, L. C. Werner, and J. Holmstrom. 2019. “Digital 1124 Twin: Vision, Benefits, Boundaries, and Creation for Buildings.” IEEE Access, 7: 1125 147406–147419. https://doi.org/10.1109/ACCESS.2019.2946515. 1126 Khan, M. S., I. S. Kim, and J. Seo. 2023. “A boundary and voxel-based 3D geological data 1127 management system leveraging BIM and GIS.” Int. J. Appl. Earth Obs. 1128 Geoinformation, 118: 103277. https://doi.org/10.1016/j.jag.2023.103277. 1129 King, M. 2022. “Innovative British tunnelling system uses swarms of lining robots to cut cost 1130 and risk.” Proc. Inst. Civ. Eng. - Civ. Eng., 175 (1): 12–12. 1131 https://doi.org/10.1680/jcien.2022.175.1.12. 1132 Klinc, R., D. Gabršček, J. Česnik, M. Žibert, M. Hostnik, and J. Logar. 2021. “Development 1133 of a Semiautomatic Parametric Method for Creation of an I-BIM Model of a Tunnel for 1134 Use in FEM Software.” J. Adv. Transp., (S. A. Biancardo, ed.), 2021: 1–18. 1135 https://doi.org/10.1155/2021/8843277. 1136 Koch, C., A. Vonthron, and M. König. 2017. “A tunnel information modelling framework to 1137 support management, simulations and visualisations in mechanised tunnelling 1138 projects.” Autom. Constr., 83: 78–90. https://doi.org/10.1016/j.autcon.2017.07.006. 1139 Kritzinger, W., M. Karner, G. Traar, J. Henjes, and W. Sihn. 2018. “Digital Twin in 1140 manufacturing: A categorical literature review and classification.” IFAC-Pap., 51 (11): 1141 1016–1022. https://doi.org/10.1016/j.ifacol.2018.08.474. 1142 Li, D.-Q., F.-P. Zhang, Z.-J. Cao, X.-S. Tang, and S.-K. Au. 2018. “Reliability sensitivity 1143 analysis of geotechnical monitoring variables using Bayesian updating.” Eng. Geol., 1144 245: 130–140. https://doi.org/10.1016/j.enggeo.2018.07.026. 1145 Li, H., W. Chen, X. Tan, and E. Chen. 2022. “Digital design and stability simulation for large 1146 underground powerhouse caverns with parametric model based on BIM-based 1147 framework.” Tunn. Undergr. Space Technol., 123: 104375. 1148 https://doi.org/10.1016/j.tust.2022.104375. 1149 Li, T., X. Li, Y. Rui, J. Ling, S. Zhao, and H. Zhu. 2024a. “Digital twin for intelligent tunnel 1150 construction.” Autom. Constr., 158: 105210. 1151 https://doi.org/10.1016/j.autcon.2023.105210. 1152 Li, T., Y. Rui, S. Zhao, Y. Zhang, H. Zhu, and X. Li. 2024b. “A quantitative digital twin 1153 maturity model for underground infrastructure based on D-ANP.” Tunn. Undergr. 1154 Space Technol., 146: 105612. https://doi.org/10.1016/j.tust.2024.105612. 1155 Li, T., Y. Rui, H. Zhu, L. Lu, and X. Li. 2024c. “Comprehensive digital twin for infrastructure: 1156 A novel ontology and graph-based modelling paradigm.” Adv. Eng. Inform., 62: 1157 102747. https://doi.org/10.1016/j.aei.2024.102747. 1158 Li, X., X. Zhang, Y. Shen, Y. Bai, and G. Yang. 2020. “Application Status and Analysis of 1159 BIM in Pit Engineering.” J. Phys. Conf. Ser., 1710 (1): 012006. 1160 https://doi.org/10.1088/1742-6596/1710/1/012006. 1161 Lin, W., B. Sheil, X. Xie, Y. Zhang, and Y. Cao. 2024. “Semantic segmentation of large-scale 1162 segmental lining point clouds using 3D deep learning.” IOP Conf. Ser. Earth Environ. 1163 Sci., 1337 (1): 012026. https://doi.org/10.1088/1755-1315/1337/1/012026. 1164 45 Liu, D., C. Sun, J. Chen, and L. Liu. 2023. “Multisensory and BIM-Integrated Digital Twin to 1165 Improve Urban Excavation Safety.” J. Comput. Civ. Eng., 37 (5): 04023025. 1166 https://doi.org/10.1061/JCCEE5.CPENG-5354. 1167 Lou, J., W. Lu, and F. Xue. 2021. “A Review of BIM Data Exchange Method in BIM 1168 Collaboration.” Proc. 25th Int. Symp. Adv. Constr. Manag. Real Estate, X. Lu, Z. 1169 Zhang, W. Lu, and Y. Peng, eds., 1329–1338. Singapore: Springer Singapore. 1170 Ma, W., X. Wang, J. Wang, X. Xiang, and J. Sun. 2021. “Generative Design in Building 1171 Information Modelling (BIM): Approaches and Requirements.” Sensors, 21 (16): 5439. 1172 https://doi.org/10.3390/s21165439. 1173 Macías, A., D. Muñoz, E. Navarro, and P. González. 2024. “Data fabric and digital twins: An 1174 integrated approach for data fusion design and evaluation of pervasive systems.” Inf. 1175 Fusion, 103: 102139. https://doi.org/10.1016/j.inffus.2023.102139. 1176 Mahmoudi, E., M. Stepien, and M. König. 2021. “Optimisation of geotechnical surveys using 1177 a BIM-based geostatistical analysis.” Smart Sustain. Built Environ., 10 (3): 420–437. 1178 https://doi.org/10.1108/SASBE-03-2021-0045. 1179 McKee, D. 2023. Platform Stack Architectural Framework: An Introductory Guide. White 1180 Paper. Digital Twin Consortium. 1181 Mitelman, A., B. Yang, A. Urlainis, and D. Elmo. 2023. “Coupling Geotechnical Numerical 1182 Analysis with Machine Learning for Observational Method Projects.” Geosciences, 13 1183 (7): 196. https://doi.org/10.3390/geosciences13070196. 1184 Mohammadi, D., and D. Parsapour. 2024. “Structural Response of Large Span Underground 1185 Spaces Due to Adjacent Excavation.” Geotech. Geol. Eng., 42 (2): 1269–1295. 1186 https://doi.org/10.1007/s10706-023-02618-y. 1187 Moshood, T. D., J. Ob. Rotimi, W. Shahzad, and J. A. Bamgbade. 2024. “Infrastructure digital 1188 twin technology: A new paradigm for future construction industry.” Technol. Soc., 77: 1189 102519. https://doi.org/10.1016/j.techsoc.2024.102519. 1190 Naderi, H., and A. Shojaei. 2023. “Digital twinning of civil infrastructures: Current state of 1191 model architectures, interoperability solutions, and future prospects.” Autom. Constr., 1192 149: 104785. https://doi.org/10.1016/j.autcon.2023.104785. 1193 “National Underground Asset Register (NUAR).” 2024. GOV.UK. Accessed May 30, 2024. 1194 https://www.gov.uk/guidance/national-underground-asset-register-nuar. 1195 Nguyen, T. D., and S. Adhikari. 2023. “The Role of BIM in Integrating Digital Twin in 1196 Building Construction: A Literature Review.” Sustainability, 15 (13): 10462. 1197 https://doi.org/10.3390/su151310462. 1198 Nicholson, D. (Ed.). 1999. The observational method in ground engineering: principles and 1199 applications. CIRIA Report. London: Construction Industry Research and Information 1200 Association. 1201 Ninić, J. 2024. “Automated reconstruction of digital twins for underground infrastructure.” 1202 Rankine day seminar at Imperial College London. 1203 Ninić, J., A. Alsahly, A. Vonthron, H.-G. Bui, C. Koch, M. König, and G. Meschke. 2021. 1204 “From digital models to numerical analysis for mechanised tunnelling: A fully 1205 automated design-through-analysis workflow.” Tunn. Undergr. Space Technol., 107: 1206 103622. https://doi.org/10.1016/j.tust.2020.103622. 1207 Ninić, J., H. G. Bui, and G. Meschke. 2020a. “BIM-to-IGA: A fully automatic design-through-1208 analysis workflow for segmented tunnel linings.” Adv. Eng. Inform., 46: 101137. 1209 https://doi.org/10.1016/j.aei.2020.101137. 1210 Ninić, J., H.-G. Bui, C. Koch, and G. Meschke. 2019a. “Computationally Efficient Simulation 1211 in Urban Mechanized Tunneling Based on Multilevel BIM Models.” J. Comput. Civ. 1212 Eng., 33 (3): 04019007. https://doi.org/10.1061/(ASCE)CP.1943-5487.0000822. 1213 46 Ninić, J., H.-G. Bui, C. Koch, and G. Meschke. 2019b. “Computationally Efficient Simulation 1214 in Urban Mechanized Tunneling Based on Multilevel BIM Models.” J. Comput. Civ. 1215 Eng., 33 (3): 04019007. https://doi.org/10.1061/(ASCE)CP.1943-5487.0000822. 1216 Ninić, J., A. Gamra, and B. Ghiassi. 2024. “Real-time assessment of tunnelling-induced 1217 damage to structures within the building information modelling framework.” Undergr. 1218 Space, 14: 99–117. https://doi.org/10.1016/j.undsp.2023.05.010. 1219 Ninić, J., C. Koch, A. Vonthron, W. Tizani, and M. König. 2020b. “Integrated parametric 1220 multi-level information and numerical modelling of mechanised tunnelling projects.” 1221 Adv. Eng. Inform., 43: 101011. https://doi.org/10.1016/j.aei.2019.101011. 1222 Ninić, J., J. Stascheit, and G. Meschke. 2014. “Beam–solid contact formulation for finite 1223 element analysis of pile–soil interaction with arbitrary discretization.” Int. J. Numer. 1224 Anal. Methods Geomech., 38 (14): 1453–1476. https://doi.org/10.1002/nag.2262. 1225 Norbury, D. 2021. “Ground models; a brief overview.” Q. J. Eng. Geol. Hydrogeol., 54 (2): 1226 018. https://doi.org/10.1144/qjegh2020-018. 1227 “OGC Geoscience Markup Language (GeoSciML).” n.d. Open Geospatial Consort. Accessed 1228 December 10, 2023. https://www.ogc.org/standard/geosciml/. 1229 Opoku, D.-G. J., S. Perera, R. Osei-Kyei, and M. Rashidi. 2021. “Digital twin application in 1230 the construction industry: A literature review.” J. Build. Eng., 40: 102726. 1231 https://doi.org/10.1016/j.jobe.2021.102726. 1232 Ouyang, W., G. Li, L. Chen, and S.-W. Liu. 2024. “Machine learning-based soil–structure 1233 interaction analysis of laterally loaded piles through physics-informed neural 1234 networks.” Acta Geotech. https://doi.org/10.1007/s11440-023-02179-7. 1235 Phoon, K.-K. 2018. “Probabilistic Site Characterization.” ASCE-ASME J. Risk Uncertain. Eng. 1236 Syst. Part Civ. Eng., 4 (4): 02018002. https://doi.org/10.1061/AJRUA6.0000992. 1237 Phoon, K.-K., Z.-J. Cao, Z. Liu, and J. Ching. 2023. “Report for ISSMGE 1238 TC309/TC304/TC222 Third ML dialogue on ‘Data-Driven Site Characterization 1239 (DDSC)’: 3 December 2021, Norwegian Geotechnical Institute, Oslo, Norway 1240 (Online).” Georisk Assess. Manag. Risk Eng. Syst. Geohazards, 17 (1): 227–238. 1241 https://doi.org/10.1080/17499518.2022.2105366. 1242 Phoon, K.-K., J. Ching, and Z. Cao. 2022a. “Unpacking data-centric geotechnics.” Undergr. 1243 Space, 7 (6): 967–989. https://doi.org/10.1016/j.undsp.2022.04.001. 1244 Phoon, K.-K., J. Ching, and T. Shuku. 2022b. “Challenges in data-driven site characterization.” 1245 Georisk Assess. Manag. Risk Eng. Syst. Geohazards, 16 (1): 114–126. 1246 https://doi.org/10.1080/17499518.2021.1896005. 1247 Powderham, A. J. 2002. “The observational method—learning from projects.” Proc. Inst. Civ. 1248 Eng. - Geotech. Eng., 155 (1): 59–69. https://doi.org/10.1680/geng.2002.155.1.59. 1249 Powderham, A. J., and A. O’Brien. 2021. The observational method in civil engineering: 1250 maximising value, minimising risk. Boca Raton: CRC Press. 1251 Pritchard, J. A., and J. Preston. 2018. “Understanding the contribution of tunnels to the overall 1252 energy consumption of and carbon emissions from a railway.” Transp. Res. Part 1253 Transp. Environ., 65: 551–563. https://doi.org/10.1016/j.trd.2018.09.010. 1254 Quigley, P., F. Buggy, J. Clancy, M. Condron, C. Finnerty, D. Gill, G. Hayes, D. Luby, and M. 1255 Ryan. 2016. “Some trends from recent geotechnical projects in Ireland.” Proc. Inst. Civ. 1256 Eng. - Forensic Eng., 169 (3): 83–93. https://doi.org/10.1680/jfoen.16.00015. 1257 Rafsanjani, H. N., and A. H. Nabizadeh. 2023. “Towards digital architecture, engineering, and 1258 construction (AEC) industry through virtual design and construction (VDC) and digital 1259 twin.” Energy Built Environ., 4 (2): 169–178. 1260 https://doi.org/10.1016/j.enbenv.2021.10.004. 1261 Randolph, M. F. 2003. “Science and empiricism in pile foundation design.” Géotechnique, 53 1262 (10): 847–875. https://doi.org/10.1680/geot.2003.53.10.847. 1263 47 Rives, M., A. Borrmann, A. Alsahly, and H. Atzbache. 2020. IFC-Tunnel Project- 1264 Requirements analysis report v1. Draft. buildingSMART International. 1265 Romanovich, M., M. Kuzmenkova, V. Breskich, and K. Kulakov. 2021. “Using the laser 1266 scanning method in the reconstruction of metro stations.” Transp. Res. Procedia, 54: 1267 819–826. https://doi.org/10.1016/j.trpro.2021.02.135. 1268 Roper, T., R. Bertuzzi, D. Oliveira, and J. Karlovšek. 2024. “A framework to demonstrate the 1269 utility and safety of the observational method.” Geomech. Tunn., 17 (2): 91–103. 1270 https://doi.org/10.1002/geot.202300041. 1271 Royston, R., B. B. Sheil, and B. W. Byrne. 2022. “Monitoring the construction of a large-1272 diameter caisson in sand.” Proc. Inst. Civ. Eng. - Geotech. Eng., 175 (3): 323–339. 1273 https://doi.org/10.1680/jgeen.19.00266. 1274 Sacks, R., I. Brilakis, E. Pikas, H. S. Xie, and M. Girolami. 2020a. “Construction with digital 1275 twin information systems.” Data-Centric Eng., 1: e14. 1276 https://doi.org/10.1017/dce.2020.16. 1277 Sacks, R., M. Girolami, and I. Brilakis. 2020b. “Building Information Modelling, Artificial 1278 Intelligence and Construction Tech.” Dev. Built Environ., 4: 100011. 1279 https://doi.org/10.1016/j.dibe.2020.100011. 1280 Satyanaga, A., G. D. Aventian, Y. Makenova, A. Zhakiyeva, Z. Kamaliyeva, S.-W. Moon, and 1281 J. Kim. 2023. “Building Information Modelling for Application in Geotechnical 1282 Engineering.” Infrastructures, 8 (6): 103. 1283 https://doi.org/10.3390/infrastructures8060103. 1284 Sheil, B. 2017. “Numerical simulations of the reuse of piled raft foundations in clay.” Acta 1285 Geotech., 12 (5): 1047–1059. https://doi.org/10.1007/s11440-017-0522-8. 1286 Shi, C., and Y. Wang. 2021a. “Development of Subsurface Geological Cross-Section from 1287 Limited Site-Specific Boreholes and Prior Geological Knowledge Using Iterative 1288 Convolution XGBoost.” J. Geotech. Geoenvironmental Eng., 147 (9): 04021082. 1289 https://doi.org/10.1061/(ASCE)GT.1943-5606.0002583. 1290 Shi, C., and Y. Wang. 2021b. “Development of Subsurface Geological Cross-Section from 1291 Limited Site-Specific Boreholes and Prior Geological Knowledge Using Iterative 1292 Convolution XGBoost.” J. Geotech. Geoenvironmental Eng., 147 (9): 04021082. 1293 https://doi.org/10.1061/(ASCE)GT.1943-5606.0002583. 1294 Shi, C., and Y. Wang. 2023. “Data-driven spatio-temporal analysis of consolidation for rapid 1295 reclamation.” Géotechnique, 1–21. https://doi.org/10.1680/jgeot.22.00016. 1296 “Singapore-ETH Centre.” 2023. Accessed May 30, 2024. https://sec.ethz.ch/research/digital-1297 underground.html. 1298 Soga, K., and L. Luo. 2018. “Distributed fiber optics sensors for civil engineering infrastructure 1299 sensing.” J. Struct. Integr. Maint., 3 (1): 1–21. 1300 https://doi.org/10.1080/24705314.2018.1426138. 1301 Spross, J. 2014. “A Critical review of observational method.” Licentiate Thesis. Stockholm: 1302 KTH Royal Institute of Technology. 1303 Spross, J., and F. Johansson. 2017. “When is the observational method in geotechnical 1304 engineering favourable?” Struct. Saf., 66: 17–26. 1305 https://doi.org/10.1016/j.strusafe.2017.01.006. 1306 Suhail, S. A. 2017. “Monitoring Pile Performance using Fiber Optic Sensor Technology – An 1307 Account on Recent Developments.” Int. Conf. New Chall. Geotechncial Eng. Lahore 1308 Pak. 1309 Sun, Z., H. Li, Y. Bao, X. Meng, and D. Zhang. 2023. “Intelligent Risk Prognosis and Control 1310 of Foundation Pit Excavation Based on Digital Twin.” Buildings, 13 (1): 247. 1311 https://doi.org/10.3390/buildings13010247. 1312 48 Svensson, M. 2017. “GeoBIM - a tool for optimal geotechnical design.” Geotech. Eng., 1781. 1313 Seoul. 1314 Tao, F., F. Sui, A. Liu, Q. Qi, M. Zhang, B. Song, Z. Guo, S. C.-Y. Lu, and A. Y. C. Nee. 1315 2019a. “Digital twin-driven product design framework.” Int. J. Prod. Res., 57 (12): 1316 3935–3953. https://doi.org/10.1080/00207543.2018.1443229. 1317 Tao, F., H. Zhang, A. Liu, and A. Y. C. Nee. 2019b. “Digital Twin in Industry: State-of-the-1318 Art.” IEEE Trans. Ind. Inform., 15 (4): 2405–2415. 1319 https://doi.org/10.1109/TII.2018.2873186. 1320 Tayler, H. 2020. “A short guide to reusing foundations.” Struct. Eng., 2020. 1321 Teizer, J., K. W. Johansen, and C. Schultz. 2022. “The Concept of Digital Twin for 1322 Construction Safety.” Constr. Res. Congr. 2022, 1156–1165. Arlington, Virginia: 1323 American Society of Civil Engineers. 1324 Templeman, J., and B. Sheil. 2024. “Development of a new soil-structure contact stress sensor 1325 for underground construction applications.” ICE Publishing. 1326 https://doi.org/10.17863/CAM.105660. 1327 Torzoni, M., M. Tezzele, S. Mariani, A. Manzoni, and K. E. Willcox. 2024. “A digital twin 1328 framework for civil engineering structures.” Comput. Methods Appl. Mech. Eng., 418: 1329 116584. https://doi.org/10.1016/j.cma.2023.116584. 1330 Tschuchnigg, F., and C. Lederhilger. 2020. “BIM in Geotechnics - Application to Road and 1331 Railway Construction.” Inf. Technol. Geo-Eng., Springer Series in Geomechanics and 1332 Geoengineering, A. G. Correia, J. Tinoco, P. Cortez, and L. Lamas, eds., 471–482. 1333 Cham: Springer International Publishing. 1334 Uhlenkamp, J.-F., J. B. Hauge, E. Broda, M. Lutjen, M. Freitag, and K.-D. Thoben. 2022. 1335 “Digital Twins: A Maturity Model for Their Classification and Evaluation.” IEEE 1336 Access, 10: 69605–69635. https://doi.org/10.1109/ACCESS.2022.3186353. 1337 Valinejadshoubi, M., A. Bagchi, and O. Moselhi. 2019. “Development of a BIM-Based Data 1338 Management System for Structural Health Monitoring with Application to Modular 1339 Buildings: Case Study.” J. Comput. Civ. Eng., 33 (3): 05019003. 1340 https://doi.org/10.1061/(ASCE)CP.1943-5487.0000826. 1341 Voyagaki, E., J. J. Crispin, C. E. L. Gilder, K. Ntassiou, N. O’Riordan, P. Nowak, T. Sadek, 1342 D. Patel, G. Mylonakis, and P. J. Vardanega. 2022. “The DINGO database of axial pile 1343 load tests for the UK: settlement prediction in fine-grained soils.” Georisk Assess. 1344 Manag. Risk Eng. Syst. Geohazards, 16 (4): 640–661. 1345 https://doi.org/10.1080/17499518.2021.1971249. 1346 Wan, Z., J. Qin, and Z. Wei. 2023. “Overview of technical research on safety monitoring, early 1347 warning, and risk assessment for underground structural engineering construction.” 1348 Emerg. Manag. Sci. Technol., 3 (1): 0–0. https://doi.org/10.48130/EMST-2023-0018. 1349 Wang, M., and X. Yin. 2022. “Construction and maintenance of urban underground 1350 infrastructure with digital technologies.” Autom. Constr., 141: 104464. 1351 https://doi.org/10.1016/j.autcon.2022.104464. 1352 Wang, X., C.-J. Liang, C. C. Menassa, and V. R. Kamat. 2021. “Interactive and Immersive 1353 Process-Level Digital Twin for Collaborative Human–Robot Construction Work.” J. 1354 Comput. Civ. Eng., 35 (6): 04021023. https://doi.org/10.1061/(ASCE)CP.1943-1355 5487.0000988. 1356 Wang, Y. J., R. Li, Z. Yang, Z. Tan, and Z. Xu. 2022a. “Exploration on 3D Geological 1357 Modeling Technology Based on BIM Secondary Development—Taking Liangshuijing 1358 Tunnel as an Example.” 2022 Int. Conf. Comput. Infrastruct. Urban Plan., 26–35. 1359 Nanchang China: ACM. 1360 Wang, Y., C. Shi, and X. Li. 2022b. “Machine learning of geological details from borehole 1361 logs for development of high-resolution subsurface geological cross-section and 1362 49 geotechnical analysis.” Georisk Assess. Manag. Risk Eng. Syst. Geohazards, 16 (1): 2–1363 20. https://doi.org/10.1080/17499518.2021.1971254. 1364 Wang, Y., and H.-M. Tian. 2023. “Digital geotechnics: from data-driven site characterisation 1365 towards digital transformation and intelligence in geotechnical engineering.” Georisk 1366 Assess. Manag. Risk Eng. Syst. Geohazards, 1–25. 1367 https://doi.org/10.1080/17499518.2023.2278136. 1368 Wei, L., S. Liu, Q. Wei, and Y. Wang. 2020. “Concept, Method and Application of 1369 Computational BIM.” Adv. Intell. Syst. Interact. Appl., Advances in Intelligent Systems 1370 and Computing, F. Xhafa, S. Patnaik, and M. Tavana, eds., 392–398. Cham: Springer 1371 International Publishing. 1372 Wu, J., L. Dai, G. Xue, and J. Chen. 2022a. “Theory and Technology of Digital Twin Model 1373 for Geotechnical Engineering.” Proc. 8th Int. Conf. Civ. Eng., Lecture Notes in Civil 1374 Engineering, G. Feng, ed., 403–411. Singapore: Springer Singapore. 1375 Wu, S., L. Hou, G. (Kevin) Zhang, and H. Chen. 2022b. “Real-time mixed reality-based visual 1376 warning for construction workforce safety.” Autom. Constr., 139: 104252. 1377 https://doi.org/10.1016/j.autcon.2022.104252. 1378 Xie, P., K. Chen, M. J. Skibniewski, J. Wang, and H. Luo. 2023. “Parametric geological model 1379 update and probabilistic analysis of shield tunnel excavation: A borehole-based 1380 conditional random fields approach.” Comput. Geotech., 157: 105349. 1381 https://doi.org/10.1016/j.compgeo.2023.105349. 1382 Xie, P., K. Chen, Z. Yin, Y. Zhu, H. Luo, and Q.-B. Zhang. 2024. “A BIM-based multi-model 1383 framework for advancing TBM performance − part 1: Real-time prediction of thrust 1384 force.” Tunn. Undergr. Space Technol., 151: 105856. 1385 https://doi.org/10.1016/j.tust.2024.105856. 1386 Yeow, H.-C., D. Nicholson, C.-L. Man, A. Ringer, P. Glass, and M. Black. 2014. “Application 1387 of observational method at Crossrail Tottenham Court Road station, UK.” Proc. Inst. 1388 Civ. Eng. - Geotech. Eng., 167 (2): 182–193. https://doi.org/10.1680/geng.13.00035. 1389 Yiu, W. N., H. J. Burd, and C. M. Martin. 2017. “Finite-element modelling for the assessment 1390 of tunnel-induced damage to a masonry building.” Géotechnique, 67 (9): 780–794. 1391 https://doi.org/10.1680/jgeot.sip17.P.249. 1392 Yoshida, I., Y. Tomizawa, and Y. Otake. 2021. “Estimation of trend and random components 1393 of conditional random field using Gaussian process regression.” Comput. Geotech., 1394 136: 104179. https://doi.org/10.1016/j.compgeo.2021.104179. 1395 Yu, G., Y. Wang, Z. Mao, M. Hu, V. Sugumaran, and Y. K. Wang. 2021. “A digital twin-based 1396 decision analysis framework for operation and maintenance of tunnels.” Tunn. 1397 Undergr. Space Technol., 116: 104125. https://doi.org/10.1016/j.tust.2021.104125. 1398 Yu, S., and Y. Geng. 2019. “Influence Analysis of Underground Excavation on the Adjacent 1399 Buildings and Surrounding Soil Based on Scale Model Test.” Adv. Civ. Eng., 2019: 1–1400 15. https://doi.org/10.1155/2019/6527175. 1401 Zeng, C.-F., H.-B. Chen, H. Liao, X.-L. Xue, Q.-N. Chen, and Y. Diao. 2023. “Behaviours of 1402 groundwater and strata during dewatering of large-scale excavations with a nearby 1403 underground barrier.” J. Hydrol., 620: 129400. 1404 https://doi.org/10.1016/j.jhydrol.2023.129400. 1405 Zhang, N., R. Bahsoon, N. Tziritas, and G. Theodoropoulos. 2022. “Explainable Human-in-1406 the-loop Dynamic Data-Driven Digital Twins.” Dyn. Data Driven Appl. Syst. DDDAS 1407 2022 Lect. Notes Comput. Sci., 13984. https://doi.org/10.1007/978-3-031-52670-1_23. 1408 Zhou, X., P. Shi, and B. Sheil. 2024a. “Knowledge-based multiple point statistics for soil 1409 stratigraphy simulation.” Tunn. Undergr. Space Technol., 143: 105475. 1410 https://doi.org/10.1016/j.tust.2023.105475. 1411 50 Zhou, X., P. Shi, B. Sheil, and S. Suryasentana. 2024b. “Knowledge-based U-Net and transfer 1412 learning for automatic boundary segmentation.” Adv. Eng. Inform., 59: 102243. 1413 https://doi.org/10.1016/j.aei.2023.102243. 1414 Zhu, H., M. Huang, and Q.-B. Zhang. 2024. “TunGPR: Enhancing data-driven maintenance 1415 for tunnel linings through synthetic datasets, deep learning and BIM.” Tunn. Undergr. 1416 Space Technol., 145: 105568. https://doi.org/10.1016/j.tust.2023.105568. 1417 1418