Guidance System for Directional Control in Shield Tunneling Using Machine Learning Techniques

Abstract

In shield tunneling, it is essential to control the shield jacks appropriately so that the shield machine follow a planned path, which depends on the position of the shield machine and the geological conditions. However, the quality of the directional control of the shield machine depends on the skill of its operator. Herein, a guidance system that provides a method for controlling shield jacks that is equivalent to the techniques utilized by a skillful operator is described. The developed guidance system uses machine learning models trained by gradient tree boosting with the operational data related to the actions of skilled operators. The models predict the optimal point at which the resultant force of shield jacks should be acted upon to control the propulsive direction of a shield machine. To validate the performance of the guidance system, a shield machine was driven according to the system at a site under construction. Deviation from the planned path and the attitude of the shield machine were within set tolerances. The results show that our guidance system has applicability in real environments, indicating the future possibility of self-driving shield machines.