In addition to thrust, steering and ground conditioning, TBM operators must digest and respond to real-time information about surface and subsurface settlement, vibrations, cutter-head speed and rotation, and screw-conveyor activity. Those are among the outputs collected by the machine’s central control system, frequently a programmable logic controller (PLC).

However, studies indicate that a typical operator can’t optimally control more than three to four of the more than two dozen independent parameters that TBMs generate, says Michael Mooney, professor of civil engineering and the Grewcock chair of underground construction and tunneling at Colorado School of Mines’ (CSM) Center of Underground Construction and Tunneling in Golden, Colo.

Complexity is amplified in urban settings, “where there is little margin for error due to allowable settlements of just ¼ to ½ inch for overlying utilities, streets and buildings, and for projects involving tunnel excavations of up to 700 inches in diameter,” Mooney says.

“You’re operating a system where conditions—ground, torque, thrust, vibrations—are changing all the time,” concurs Randy Essex, executive vice president with London-based engineer and tunneling specialist Mott MacDonald Group. He is also an executive council member with the International Tunneling and Space Association (ITSA). “It’s too much for humans to handle, so technology is stepping in to assist us,” Essex says.

Multiple Partners

Many entities—members of academia, CSM included—are working among themselves and with industry leaders to advance automation in TBMs. The goals: improve tunneling performance and productivity. In particular, CSM is deploying artificial intelligence—pattern recognition in data from operator-generated inputs and performance-related outputs—to understand the physics of TBM ground interactions.

To do so, researchers gather data from thousands of sensors placed throughout the TBM, and on ground surfaces and other targets, then analyze that information via CSM-generated algorithms to detect or “tease out” patterns in data otherwise difficult to distinguish.

“It’s really about identifying relationships between the inputs and outputs—what works or doesn’t as well in actions between the two,” Mooney says. “To achieve a better output—say in advanced speed—you need to specify a better input. Through pattern recognition, AI may suggest the input changes to advance the speed.”

All told, “we’re ideally identifying the best combination of parameters to optimize the TBM’s performance or keep abreast of when conditions change,” says Mooney. Assuming that patterns are identified early and updated as the TBM progresses, tunnelers can use that to improve performance throughout the project.

“To some extent, most TBMs already implement automation via PLCs that respond to a given output with a given input,” says Richard McLane, chief mechanical engineer with Evansville, Ind.-based Traylor Bros. The contracting firm specializes in large tunneling projects, including the West Side Subway Extension in Los Angeles, a 3.92-mile undertaking. Like other large firms, Traylor has dabbled in automation-related boring, including elements of ground conditioning.

The advances come at a time when infrastructure providers, particularly those in urban areas, find they have only one direction left to go. In 2016, worldwide tunneling expenditures totaled $100 billion and projects achieved 3,200 miles, according to ITSA. In addition, activity is growing at an average annual rate of 7%, twice that of the global construction market.

As a result, the time is ripe for CSM to move its AI-related activity lab to the field. If all goes as planned, the CSM team will introduce AI as a project element for the upcoming excavation of the $600-million Northeast Boundary Tunnel (NEBT) in Washington, D.C. It is the largest portion of city’s multibillion-dollar Clean Rivers Project, says Daniele Nebbia, project manager with Lane Construction Corp. The Cheshire, Conn.-based contractor is partnering on NEBT with Milan, Italy-based controlling contractor Salini Impregilo.

Identifying Inefficiencies

Having done tunneling for a previous phase of Clean Rivers, the two firms provided CSM with TBM-generated data from the project. That may help to maximize AI on the 27,000-ft-long, 23-in.-dia NEBT, where depths will extend 50 ft to 160 ft below ground. “Ideally, we’d be riding along virtually, also analyzing TBM data generated from excavating NEBT,” says Mooney.

“We’re attempting to identify the inefficiencies in our parameters and minimize downtime,” adds Nebbia. “As contractors, we want to be as efficient and productive as possible and, in this instance, manage settlement at the same time.”

However, the road to complete TBM automation may prove to be slow. For the foreseeable future, “industry will likely rely on systems in which AI contributes to operator-assisted decision making,” Mooney says. “In time, given the reliability of intelligence, the operator may allow 5% AI-generated decisions, then 25% or 50%, and, finally, 100%. Think of it like the development and acceptance of a self-driving car,” he says.

Meanwhile, the emphasis is on improving imaging ahead of TBMs. Although project teams may bore 6-in.-dia test holes every 100 to 200 yd, those don’t result in a complete map of below-grade conditions.

To better illuminate conditions, CSM is working with radar that reflects water, fracturing and other conditions before the TBM encounters them. CSM recently introduced the technique for a 700- to 900-ft-deep bypass to the leaking Rondout-West-Branch Tunnel. For the project, located in New York’s Orange and Dutchess counties, “joint venture contractors Kiewit and Shea are drilling 2.5-inch-diameter horizontal probe holes through about 200 to 300 feet of ground, then routing a radar-emitting probe through the boreholes,” Mooney says. To date, the technology has proved useful, though it remains to be seen how it performs in difficult terrain with groundwater, faults, fractures, solution cavities, etc. in the limestone.

The CSM lab also is researching imaging from the injection of low levels of electricity into the ground ahead of a TBM’s cutter head. The technology is key for urban settings, given that electrical currents can identify spatial contrast in conductivity, not only from changes in rock and soil types, but also from pipes, foundations, abandoned wells and other anomalies, Mooney says.

However, moving the technology from lab to field may prove problematic. “We’re working on creating more robust instrumentation that can withstand the rigors of a destructive environment,” he says.