Prediction of TBM Performance in Soft Soils using Artificial Neural Network (ANN) Approach

Abstract

Modern Tunneling projects are often relying on Tunnel Boring Machines (TBMs) rather than traditional methods to increase the tunneling efficiency and decrease the time plan of the projects. TBMs have in general a circular cross-section with diameter conventionally vary from 1.0 m to 19.0 m depending on the tunnel usage which can be underground water and sewer tunnel or big tunnels for roads, railway and subways. These tunnels can be rather short in length or can get extended over miles connecting countries under seas and oceans like Channel Tunnel between France and United Kingdom. Most modern TBMs have more than 400 mounted monitoring sensors that are connected to the TBM management control system to record all data measurements during tunneling. Most of these data measurements are obtained and observed during mining by the TBM operators. Every 10 seconds, the mounted sensors could record the data measurements in a central data collection system to be studied later while some of these measurements may not be visible during the TBM operation. Several performance predictions approaches have been used to study the recorded data measurements during tunneling to estimate and predict the performance of various TBMs. The most critical and affected parameter to evaluate the TBM performance is the penetration rate. The penetration rate unit may differ from one study to another as it can be considered as the advance per rotation (such as mm/rot.) or advance per time unit (such as m/hr.) Most of these models are based on specific ground types that may differ from the regional geological characteristics. While several major tunneling projects are yet to realize regionally in challenging and varying ground, it became important to develop regional-specific prediction models that are particularly suitable for local challenging geological and hydrological features such as soft clay.