EnergyTransportationTunnel BoringWater

EPB shield tunnel-boring machine automation using the autonomous-vehicle framework

Advances made in self-driving vehicles and automated drilling during the past decade suggest that tunnel-boring machine (TBM) tunneling is headed toward autonomous operation. Automated drilling, primarily adopted in blast-hole drilling for mining, involves driverless surface drill rigs that self-navigate to a predetermined X-Y location and drill at a preset angle (typically vertical) to a target depth. Remote operators oversee the drill rigs and perform some of the operations (de Wardt et al., 2012; de Wardt et al., 2016; Rogers et al., 2019). In current practice, autonomy is limited to autopositioning of drill rigs, drill-rod handling and drilling fault detection. The main motivation behind automated drilling has been worker safety and workforce shortage in remote area operations. Improvements in production and drilling accuracy have been reported (Kinik et al., 2014; Jacobs, 2015; Lopes et al., 2018). To the authors’ knowledge, the drilling process itself has not become intelligent in terms of learning how to drill more efficiently.

Self-driving or autonomous vehicles (AVs) provide a compelling roadmap for autonomous TBM tunneling given the complexity of the technology development, the partial or subsystem adoption demonstrated, the human factor and the likelihood that full autonomy will be realized within the next decade. There are six levels of autonomous driving, levels 0-5 are summarized in Fig. 1. At the time of this writing, AVs on the road perform at level three via a variety of autonomous functions including adaptive cruise control, anticollision self-braking, traffic-jam assist, self-parking and lane-centering assist. Fully AVs, where a driver is not required, and defined by levels four and five, are commercially used in less complex environments (e.g., local shuttle services). However, fully AVs are predicted to occur within the next decade, and no fewer than 1,400 fully AV vehicles are currently in testing by more than 80 companies across 36 states in the United States (Etherington, 2019).

The main motivation behind AV technology is safety. There are approximately 6.5 million motor-vehicle crashes and 37,000 traffic deaths in the United States annually (Insurance Information Institute, 2019). The underlying premise is that AV technology is better at driving than humans are. Data show that collision avoidance technology on current vehicles has improved safety, and predictions are that fully AVs will dramatically improve safety. Another motivation is improved efficiency. With road congestion growing and limited funding to build new transportation capacity, transportation agencies are counting on significant improvements in traffic flow resulting from AVs (Kockelman et al., 2017).

TBM tunneling has advanced considerably over the past two decades, driven by the key performance indicators realized (e.g., production, ground deformation control) and by technology (e.g., sensing, information, electromechanization). Automation employed by today’s TBMs includes basic subsystem functions including cutterhead rotation speed control (akin to vehicle cruise control), foam injection ratio control and numerous hydraulic and mechanical subsystems. Recently developed and emerging autonomous functions include robotic tool changes (Camus and Moubarak, 2015), ring erection (Martin, 1990; Wu et al., 2011), annulus grouting (Shirlaw et al., 2004) and steering (Shimz Corp., 2018; Schwob et al., 2019). Further, remote operation of TBMs is routinely used in Japan through surface control centers.

Like AVs, the motivation for autonomous shield tunneling is multifaceted and includes safety (of workers and of overlying structures and their inhabitants) and optimization for improved performance (e.g., less downtime, higher production). Cost efficiency is a considerable barrier to growth in underground construction. A key question is what role automation can play in reducing cost.

This article examines autonomous earth pressure balance (EPB) shield tunneling using the development of AVs as a motivating technology. The basic technology of AVs is presented using a sensing–planning–action framework, with noted parallels to TBM tunneling. The development of autonomous EPB shield tunneling subsystems is presented within the sensing–planning– action framework. Subsystems addressed include groundconditioning optimization, deformation control, chamber pressure control, drive-performance optimization, annulus grouting, steering and ring assembly. The potential benefits of EPB shield automation are discussed, as are the challenges, both technical and nontechnical, including liability and risk.

Autonomous-vehicle framework

A sensing (perception)–planning–action (control) framework is adopted to describe AVs and to translate to autonomous EPB shield tunneling.

Six levels of autonomous vehicles as defined by the Society of Automotive Engineers (Estl, 2018).

Fig.1 -Six levels of autonomous vehicles as defined by the Society of Automotive Engineers (Estl, 2018).

AV sensing. AV technology requires an incredible amount of information that is both gathered from a suite of sensors embedded in the AV and gleaned from dynamic maps hosted on cloud servers. Sensing is the process of identifying all relevant objects and conditions within the relevant field of view that an AV requires. Sensing includes identifying objects (animal, baby stroller, traffic signs/signals, pedestrians, other vehicles) as well as their speeds and trajectories (Yurtsever et al., 2019). Sensing also involves internet of things (IoT)-enabled information (from vehicle to vehicle and from smart signage).

An array of remote sensors outfitted on the AV include optical red, green blue depth (RGBD) cameras, radar, ultrasound and light detecting and ranging (LiDAR) (Rosique et al., 2019). Infrared sensing and groundpenetrating radar (GPR) are also gaining acceptance. This suite of remote-sensing technologies provides redundancy and ideally complete coverage day or night and in allweather environments (day, night, fog, rain, snow, smoke). A critical component of sensing is identifying objects from remotely sensed information. A radar image, for example, can convey the size, shape, speed and trajectory of an object, but it does not indicate what the object is (Fig. 2). Machine learning, embedded within the technology of computer vision, plays a significant role in object identification.

AV technology also uses high-resolution maps as a quasi-baseline condition for route and motion planning. Such maps include road configurations and current information about traffic, weather-impacted road conditions, construction conditions, etc. These maps are highly dynamic based on real-time measurements of vehicle speed and congestion for traffic forecasting as well as temperature, precipitation, vehicle braking, traction control and windshield wiper use for road condition forecasting (Galanis et al., 2018). Global positioning system (GPS) and inertial measurement units using gyroscopes and accelerometers characterize the position and trajectory of the AV (i.e., AV localization (Fig. 2)).

There are some important sensing similarities and differences with TBM tunneling. The biggest difference lies in remote-sensing technologies. While AVs can use optical, LiDAR, radar and ultrasonic to image the desired field (tens of meters) with the requisite spatial resolution, tunnel-project remote-sensing options are very limited. Geophysical seismic, GPR and electrical resistivity methods have been used with only marginal efficacy to characterize the ground ahead of TBMs (Mooney et al., 2012). We posit, however, that remote sensing is much more important to AVs than it is to autonomous TBMs. Regarding similarities, AVs need to operate at night and in the most adverse weather conditions (rain, snow, fog). Similar objects that are sensed can change (e.g., pedestrian layers of clothing, size of animal, other vehicle model and age). This introduces significant ambiguity and uncertainty. In shield tunneling, the ground conditions also vary spatially.

Despite tremendous advances, AV sensing is not fully capable of enabling level-five AVs. Sensing techniques struggle to identify some objects (e.g., a ball in the road that has led to AVs stopping). Sensing will also be the limiting factor in autonomous EPB shield tunneling. We cannot reliably image the ground ahead of the TBM, and we do not know what is happening in the critical tool gap, cutterhead openings and excavation chamber. These zones of critical operation need to be better sensed.

Left: Object sensing using camera-measured image (Aryal, 2018). Right: Localization of AV (courtesy of ZME Science,

Fig.2-Left: Object sensing using camera-measured image (Aryal, 2018). Right: Localization of AV (courtesy of ZME Science,

AV planning. The process of planning is the intelligent brain behind AV technology. This is the process undertaken to perform path planning (route optimization/ selection) and motion planning (the longitudinal acceleration/braking and lateral steering commands) for the AV. Considerable machine learning is used within the planning process. Path planning determines the route from the starting position to a desired destination. To find the optimal path from all potential routes, traditionally graphbased methods are used, which consider the optimization of some objective value over all paths (e.g., minimal distance). Also, model predictive control can be used to plan for a limited path ahead. Deep learning methods have been recently used for path planning as well. One representative method is imitation learning, which uses the recorded driving experience to learn how human drivers pick driving paths from the camera-observed images. In that case, the model input is convolutional neural network (CNN)-processed camera images and the output is the human-chosen trajectory. Considering TBM automation, route planning is literally not relevant because the design tunnel alignment is prescribed. However, the process of route planning can be applied to other TBM subsystems.

Motion planning to generate the longitudinal acceleration/braking and lateral steering commands for the AV uses physics-based kinematic and dynamics models to determine the action required for the given scenario. In complex cases, machine-learning methods are used to learn the system dynamics, where the past vehicle states are model inputs and the vehicle observations are the outputs. Either physics-based or machine-learning-based models can be used within strategies such as iterative learning control (ILC) and model predictive control (MPC). ILC is simple and computationally lightweight, and is suitable for controlling systems that work in a repetitive mode, such as path tracking, or automatic parking. MPC is a more robust and capable controlling strategy. The central idea of MPC is to calculate the vehicle actions at each time step by minimizing a cost function (error between desired and predicted performance) over a short time horizon, while considering observations, input-output constraints and the system’s dynamics given by a process model. The cost function may consider multiple performance goals including static safety (collision avoidance of fixed objects), dynamic safety (avoiding other moving traffic objects), energy efficiency, ride smoothness, etc. The input constraints may include the range limit of steering, brake/ acceleration, etc., while the output constraints consist of vehicle speed limit, vehicle cross-track error (deviation) limit, orientation limit and lateral acceleration limit.

AV-motion planning is very relevant and transferrable to TBM automation. In TBM automation, the objective (cost) function also consists of multiple goals (higher advance rate, minimal ground disturbance and deformation, minimal tool wear, minimal alignment deviations). Furthermore, the inputs and outputs, like AVs, are subject to various limitations (thrust force limitation, deviation limitation, torque limitation, chamber pressure limitation, etc.). The TBM dynamics can be modeled based on physics to some extent, but this is uncertain. The ground condition is uncertain due to geotechnical variability, similar to the uncertain traffic environment caused by random driver/pedestrian behavior.

AV action. AV action is concerned with executing the planned operational parameters from the planning stage. The most common feedback controller used in AVs is the proportional-integral-derivative (PID) controller. This seems to also be the case for TBMs. The technology to implement the action phase is straightforward and already developed. As such, little further discussion on action is included in this paper.

Subsystem autonomous operations for EPB shield TBM (image portion courtesy of Herrenknecht).

Fig.3-Subsystem autonomous operations for EPB shield TBM (image portion courtesy of Herrenknecht).

Autonomous EPB shield tunneling

The development of autonomy in subsystem functions has been the mode of operation in AVs and drilling. The same is likely true for EPB shield tunneling. This section summarizes the various subsystem functions in EPB shield tunneling and how they have or will become more autonomous. Seven subsystem autonomous functions are illustrated in Fig. 3. None of these subsystems are mutually exclusive; they are all interrelated. As such, the optimization of parameters within a subsystem is constrained and a system optimization framework is required. Of the subsystems illustrated in Fig. 3, annulus grouting, steering and ring erection have become semi- or fully automated and will not be described due to page limitations. The development of ground conditioning optimization and drive-performance optimization subsystems are detailed using the sensing–planning–action autonomous framework.

Ground-conditioning optimization. Ground conditioning (Fig. 4) is applied to help transform the in situ ground (soil, rock) into a workable medium possessing a consistency that can be used to smoothly balance the water and earth pressure at the face, dissipate that pressure through the screw conveyor, minimize wear and energy required to process, and make for efficient muck passage through the cutterhead, excavation chamber, screw and disposal system (conveyor belt, buckets, truck transport, etc.). There are considerable nuances within this broader goal that depend on ground type. For example, clays and shale present a clogging risk that conditioning must mitigate. Cohesionless sands present ground-water inflow risk and abrasivity that conditioning should mitigate. To this end, there is a suite of ideal conditioned soil parameters that can be defined depending on the ground type. These include slump, permeability, shear strength, compressibility and viscosity for cohesionless sands, and consistency index, viscosity and adhesion for cohesive clays. However, there are numerous optimal conditioning inputs that vary ambiguously across soil types, including the type of conditioning agent, liquid quantity, foam injection ratio, foam expansion ratio and concentration. In addition, the configuration and size of the cutterhead, excavation chamber and screw conveyor play a significant role in the conditioned muck properties.

Foam generation and delivery system on a metro-size EPBM.

Fig.4-Foam generation and delivery system on a metro-size EPBM.

Sensing. Ideally, we wish to know the properties of the conditioned state in the tool gap, cutterhead windows, excavation chamber and screw conveyor. These properties include shear strength, compressibility and adhesion. In current EPBM tunneling, these properties are not directly measured in the desired locations. Instead, we rely on indirect measurements, including cutterhead thrust force and torque, chamber pressure fluctuation and gradient from crown to invert, screw conveyor torque and pressure gradient and measured properties from belt samples (stickiness, slump, density, shear strength). Sensing within future EPBM may involve multiphysics belt muck scanning, direct sensors (e.g., rheometer, embedded in the tool gap, chamber and screw conveyor).

Planning. Measurable criteria are needed, including:

  • Chamber pressure fluctuation (dp/dx where p = pressure and x = longitudinal/advance position), chamber pressure gradient (dp/dz where z is the vertical position), and screw pressure gradient (dp/ dL where L is length along screw conveyor).
  • Belt muck properties (density, adhesion, shear strength under a desired pressure, slump).
  • Cutterhead torque, cutterhead thrust force.

The relationships between desired outcomes and measured outcomes must be developed as a function of conditioning variables (e.g., form of conditioning, location of conditioning, foam injection ratio (FIR), foam expansion ratio (FER), water injection ratio, bentonite injection ratio). Such relationships can be developed from statistical models, machine learning models or physical models. With measurements and desired outcomes, the recommended adjustments to the conditioning variables to move from current outcomes to desired outcomes can be computed using the established relationships.

The mathematical framework for computing recommended adjustments can come from physics- and mechanics-based models (e.g., Yu et al., 2017) or from machine learning-based models (Mooney et al., 2018). With the former, the chamber pressure response is predicted based on TBM operations and conditioning inputs. Given the complexity of the EPBM/ground/ conditioning interaction, the modeled and measured outcomes are likely different. To this end, the model must be calibrated on the fly. Machine-learning models, by design, learn while doing and therefore will continuously expand knowledge of conditioning as data are collected. Where there are co-dependent relationships that sometimes may conflict, a weighting system is implemented. Examples of this include a desire to reduce cutterhead torque to some set point and a desire to maintain a set-point chamber pressure gradient. These two outcomes may invoke different conditioning actions.

Action. The recommended adjustments are implemented via the EPB shield’s PLC that sends analog signals to the system actuators to adjust flow rates as prescribed (e.g., flow controllers in Fig. 4). The array of actions include different recommendations for cutterhead, excavation chamber and screw conveyor conditioning. A feedback control loop is used to implement the continuous loop of sensing, planning and action. The loop is typically completed on a time interval of 5-10 minutes for soil conditioning given the latent response to action. Autonomous soil-conditioning optimization (McLane, 2014) can also incorporate geotechnical information (e.g., both from the ground model built from geotechnical site investigation data and from EPBM prediction of the ground type using deep learning).

Drive-performance optimization. Tunneling performance is determined by multiple factors, such as machine advance rate, alignment deviations, ground settlement/disturbance, material consumption and tool wear. Depending on the specific scenario, their priorities vary. For example, when tunneling in a deformation sensitive area, achieving minimal alignment deviation and ground disturbance is critical. Conversely, a project that has experienced significant delays may prioritize high advance rate. When the performance metric is specified, optimizing drive performance is an optimal control problem, in which the machine operations are adjusted to achieve the best expected performance. To this end, models relating operations with performance (Mokhtari, et al., 2020) are needed, either derived from physics or via statistical learning. In addition, given the limited machine capacity, as well as other coupled reactions that are subject to limitations, feasible machine operations are constrained. Improving drive performance therefore involves solving a constrained, nonconvex optimization problem that is solvable with a heuristic method.

Sensing. Many machine operations and reactions are measured on modern EPB shield machines. For operations, these include the individual thrust and articulation jack forces, cutterhead and screw conveyor rotation speeds, and the material injection rates of foam, bentonite, water and polymer. For machine reactions, advance rate, alignment deviations, chamber pressure and its distribution, cutterhead torque, screw conveyor torque and pressure are all measured. However, there are still performance metrics that are unrecorded. A major limitation is the lack of sensing in the critical tool gap, cutterhead openings and excavation chamber areas. Further, continuous cutting tool wear is rarely measured on EPB shield machines, making it hard to quantify wear rates under different operations. This can be solved by adding wired or wireless sensors to the cutting bits. Another common issue is the lack of proper sensor calibration (e.g., thrust jacks) and inadequate sampling frequency (e.g., torque measurement), both impacting the data quality and their usage for downstream processing.

Planning. To perform optimal control, both the performance objective and all coupled machine reactions are modeled. Take the advance rate maximization as an example, its optimal control formulation is given as follows:

subject to

subject to

where EPB shield operations X include the individual thrust jack forces Fi, cutterhead rotation speed ω, screw conveyor rotation speed ωs, as well as foam air and solution injection rates Qair, Qsol. is the estimated advance rate to maximize and is a function of Fi, ω, estimated cutterhead torque T̑ and estimated chamber pressure p̑. Given the challenge of physics-based modeling, can be obtained with machine learning using data recorded on previous rings as well as geotechnical information (Mooney et al., 2018). In addition, T̑ and p̑ are explicitly modeled and considered as optimization constraints. For example, the estimated value of chamber pressure p̑ should be within the tunneling face pressure bounds [pmin, pmax]. Estimated cutterhead torque T̑ must not exceed the capacity of the torque motors. Cutterhead torque can be estimated using physics-based models (Godinez et al., 2015). A similar situation applies to the FIR and FER. Finally, individual thrust jack force Fi, cutterhead rotation speed ω, and screw conveyor rotation speed ωs all have their own mechanical limits set by interlocks and are all treated as constraints. The objective function and the constraints are nonconvex; therefore, solving the optimization above requires heuristic strategies such as particle swarm optimization.

Action. Formulations like the one above can be deployed for EPBM shield optimal control, though issues such as computation efficiency need to be addressed properly to yield the control in real time. Figure 5 presents one example where a predictive model, = f(Fi, ω, p̑, T̑) was trained with a support vector regression model using the previous 20 rings worth of EPB shield data. The model performance is good, with R2 = 0.75 and RMSE = 5.92 mm/min. The model, in this example, was then used to predict performance over the subsequent 20 rings using the EPB shield inputs actually adopted. As shown in Fig. 5, its performance is still acceptable, with an R2 = 0.65 and RMSE = 7.06 mm/min. In addition to this example, the model provides as output a suite of input parameters to maximize AR. The input parameters are implemented via PID control on the EPB TBM.

Diagram showing the performance of the AR model, learned using support vector regression. The training was conducted on the first 20 rings (#1001-1020) and was then tested on the following 20 rings.

Fig.5-Diagram showing the performance of the AR model, learned using support vector regression. The training was conducted on the first 20 rings (#1001-1020) and was then tested on the following 20 rings.

Schematic layout of the environment perception framework for autonomous TBM tunneling.

Fig.6-Schematic layout of the environment perception framework for autonomous TBM tunneling.

It is worth noting that there are myriad supporting operations as well as logistics that must be adequately functioning to enable autonomous operations. These include material batching (e.g., grout, bentonite, conditioners, consumables resupply, umbilical extensions, segment transport and mucking operations, to name a few). Some, if not all, of these can be made autonomous. Ideally, these operations are taken off the tunneling critical path.

Geotech environment perception

The ground plays a significant role in autonomous TBM tunneling. This environment is akin to the road network and spatial obstacle array in the area around an AV. A major concern and constraint for autonomous TBM tunneling is the fact that the spatial characterization of ground conditions along and above a tunnel project alignment is highly uncertain. A related uncertainty lies in the behavior of the ground (e.g., uncertainty in shear strength, deformation and excess pore water-pressure generation). Efforts to advance autonomous TBM tunneling, therefore, must implement ways to correctly capture spatial geotechnical uncertainty and use data to continuously update geotechnical conditions. This updating of geotechnical conditions is similar to updating road conditions performed in concert with AV technology (described earlier).

Updating the ground condition can be realized by using geostatistical models, machine-learning models or a combination of the two. Geostatistical models enable a spatial estimation of the geological-geotechnical conditions at unsampled locations and help quantify spatial uncertainty. A 3D rendering of the geological-geotechnical environment, developed using the geotechnical site investigation and associated laboratory testing results, will be fed into the local area network (LAN) for the autonomous TBM to use. TBM operation parameters can be used to estimate the as-encountered ground conditions using machine-learning models. Additionally, instrumentation and monitoring (I&M) data can be back-analyzed using suitable machine-learning algorithms and physics models to estimate the ground conditions within the zone of influence. Integrating the prior-to-construction estimates of geological-geotechnical conditions from geostatistical modeling and preposterior estimates from TBM and I&M data can aid in continuously updating the geological-geotechnical environment along the tunnel alignment. The process can enable TBMs to gain updated knowledge of the excavation environment. Figure 6 presents a general environment perception framework for autonomous TBM tunneling.


Autonomous vehicles (AVs) provide a compelling road map for autonomous TBM advancement. With a well-thoughtout structure for autonomy and AVs currently at level 3 (of 6), AV technology development has been remarkable, and fully autonomous AVs are anticipated within the next decade. Using the AV framework of sensing (perception)–planning– action (control), this paper has described the various EPM TBM subsystems that are currently partially automated or can become autonomous in the future. As with AVs, the biggest hurdle to autonomous TBM tunneling is sensing, particularly in the critical areas immediately ahead of the cutterhead, at the cutterhead and through the excavation chamber and screw conveyor. Advances in the sensing of these areas is required to drive autonomous TBM technology.

Nontechnical aspects of AVs and autonomous TBM tunneling include human trust, liability and regulation. Volvo, for example, has stated it will accept all liability with its AVs. The liability associated with autonomous TBM tunneling will be a significant barrier (e.g., Who will accept this risk and how will this liability be underwritten?). Owner acceptance and regulation will also require development (e.g., Under what conditions will owners permit autonomous TBM operations in urban environments?).


Aryal, M., 2018. Object Detection, Classification, and Tracking for Autonomous Vehicle. Camus, T., Moubarak, S., 2015. Maintenance robotics in TBM tunnelling, ISARC. Proceedings of the International Symposium on Automation and Robotics in Construction. IAARC Publications, p. 1.

de Wardt, J., Chapman, C.D., Behounek, M., 2012. Well Construction Automation- Preparing for the Big Jump.

de Wardt, J.P., Sheridan, T.B., DiFiore, A., 2016. Human Systems Integration: Key Enabler for Improved Driller Performance and Successful Automation Application, IADC/SPE Drilling Conference and Exhibition. Society of Petroleum Engineers. Estl, H., 2018. How ADAS Is Paving the Way for Autonomous Driving (https:// Etherington, D,. 2019. Over 1,400 self-driving vehicles are now in testing by 80+ companies across the US. Retrieved from 1400-self-driving-vehicles-are-now-in-testing-by-80-companies-across-the-u-s/ Galanis, I., Gurunathan, P., Burkard, D., Anagnostopoulos, I., 2018. Weather-based road condition estimation in the era of Internet-of-Vehicles (IoV), 2018 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, pp. 1-5.

Godinez, R., Yu, H., Mooney, M., Gharahbagh, E.A., Frank, G., 2015. Earth Pressure Balance Machine Cutterhead Torque Modeling: Learning from Machine Data, Rapid excavation and tunneling conference, New Orleans.

Insurance Information Institute, 2019. Facts + Statistics: Highway safety (https://www.

Jacobs, T., 2015. Automated Drilling Technologies Showing Promise. SPE-0615- 0050-JPT 67, 50-55.

Kinik, K., Gumus, F., Osayande, N., 2014. A case study: first field application of fully automated kick detection and control by MPD system in western Canada, SPE/IADC Managed Pressure Drilling & Underbalanced Operations Conference & Exhibition. Society of Petroleum Engineers.

Kockelman, et al., 2017. An assessment of autonomous vehicles: traffic impacts and infrastructure needs – final report. Report FHWA/TX-17/0-6847-1, 182 p.

Lopes, L., Miklovicz, T., Bakker, E., Milosevic, Z., 2018. The benefits and challenges of robotics in the mineral raw materials sector-an overview, 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, pp. 1507-1512. Martin, D., 1990. JAPANESE ROBOT CAN ERECT SEGMENTS. Tunnels & Tunnelling International 22.

McLane, R., 2014. Automatic Soil Conditioning Through Clay. North American Tunneling: 2014 Proceedings.

Mokhtari, S., Navidi, W., and Mooney, M.A. White Box Regression (Elastic Net) Modeling of Earth Pressure Balance Shield Machine Advance Rate, Automation in Construction, 2020, 115(103208), Mooney, M., Yu, H., Mokhtari, S., Zhang, X., Zhou, X., Alavi, E., Smiley, L., Hodder, W., 2018. EPB TBM Performance Prediction on the University Link U230 Project. North American Tunneling 2018 Proceedings


Mooney, M.A., Walter, B., Frenzel, C., 2012. Real-time tunnel boring machine monitoring: A state of the art review. North American Tunnelling, 2012 Proceedings, 73-81. Rogers, W.P., Kahraman, M.M., Drews, F.A., Powell, K., Haight, J.M., Wang, Y., Baxla, K., Sobalkar, M., 2019. Automation in the Mining Industry: Review of Technology, Systems, Human Factors, and Political Risk. Mining, Metallurgy & Exploration, 1-25. Rosique, F., Navarro, P.J., Fernández, C., Padilla, A., 2019. A systematic review of perception system and simulators for autonomous vehicles research. Sensors 19, 648. Schwob, A., Guedon, F., Combe, B., Lockhart, T., 2019. Tuen Mun–Chek Lap Kok Link in Hong Kong–Innovative solutions for construction of an outstanding Subsea Tunnel, Tunnels and Underground Cities. Engineering and Innovation Meet Archaeology, Architecture and Art. CRC Press, pp. 3086-3096.

Shimz Corporation, 2018. AI-based Support System Developed for Shield Excavation Planning.

Shirlaw, J.N., Richards, D.P., Ramond, P., Longchamp, P., 2004. Recent experience in automatic tail void grouting with soft ground tunnel boring machines, Proceedings of the ITA-AITES World Tunnel Congress, Singapore, pp. 22-27.

Wu, C., Liu, X.-J., Wang, L.-P., Wang, J., 2011. Dimension optimization of an orientation fine-tuning manipulator for segment assembly robots in shield tunneling machines. Automation in Construction 20, 353-359.

Yu, H., Mooney, M., Bezuijen, A., 2017. A Simplified Chamber Pressure Model for EPB TBM Tunneling in Granular Soil, in: Jr., A.N., Jr., M.O.C. (Eds.), 9th International Symposium on Geotechnical Aspects of Underground Construction in Soft Ground, Sao Paulo.

Yurtsever, E., Lambert, J., Carballo, A., Takeda, K., 2019. A Survey of Autonomous Driving: Common Practices and Emerging Technologies. arXiv preprint arXiv:1906.05113.

Related Articles

Back to top button