Design for Support Patterns of NATM Tunnel Using Machine Learning

NATM (New Austrian Tunneling Method) known as sequential excavation method, is a construction method of mountain tunnels. Support patterns by NATM in Japan are designed by experienced experts regarding an overall evaluation based on the condition of a cutting face, the condition of an underground surface, compressive strength, weathering, condition of joints, joins set, joint spacing, orientation of joint spring water and deterioration due to water. However, current decision methods on support patterns require expertise, and moreover, there often exists discrepancy in judgements. Therefore, in this research, we try to build expert systems on design of support patterns in tunnel construction using support vector machines (SVM) which is a kind of statistical machine learning. SVM has been widely applied to various pattern classification and regression problems and recognized to be powerful learning method. We will also show through some real data that the proposed expert systems using SVM can suggest reasonable and objective design of tunnel support patterns.