01852503

Component

https://annualmeeting.myt...nlineProgram/Details/17298

This study proposes a machine learning based prediction model for early prediction of ground collapses during rock tunneling through adverse geologic conditions. The occurrence of sudden collapses while operating a tunnel boring machine (TBM) through weak rock condition can cause accidents, resulting in operational delay and cost overrun. Current field practices incorporate geologic prospecting sensors to the TBM to detect such collapse locations which require extra cost and maintenance. However, the operational parameters collected by the TBM data acquisition system can offer significant benefit by eliminating the need for such additional sensors. In this study, the authors used the operational parameters collected by a gripper TBM used in a water conveyance tunneling project in China experiencing ground collapses to build the proposed prediction model. The TBM data collected from the tunneling cycles were utilized without any data compression to have sufficient data instances for the collapse incidents. Three machine learning classifiers, namely: (1) multilayer perceptron, (2) support vector machine, and (3) random forest, were trained on the processed TBM data and geological survey data. The prediction accuracy of the proposed model reached 98% for training data and 96% for validation data.

This paper was sponsored by TRB committee AKB60 Standing Committee on Tunnels and Underground Structures.

TRBAM-22-00163

English

18p

2022

Transportation Research Board 101st Annual Meeting

Digital/other

Figures; References; Tables

Collapse; Machine learning; Predictive models; Tunnel borers; Water tunnels

China

Bridges and other structures; Geotechnology; Transportation (General)

Transportation Research Board Annual Meeting 2022 Paper #TRBAM-22-00163

TRIS, TRB, ATRI

Jul 19 2022 12:23PM