A Review of Soft Computing Techniques in Predicting Overbreak Induced by Tunnel Blasting

Abstract

Blasting is an economical and practical construction technique for excavating underground structures. However, there are some undesirable environmental issues caused by tunnel blasting. One of the detrimental effects induced by tunnel blasting is overbreak. To alleviate the influence of overbreak, some techniques like empirical, statistical, and numerical have been proposed to solve this problem. However, considering the limitations of the aforementioned methods in analyzing the relationship between the target tunneling overbreak and influential input parameters, soft computing (SC) and artificial intelligence (AI) techniques were developed and employed to evaluate the impact of each factor related to the formation of overbreak. The researchers still face difficulties in making an optimal choice of the best SC and AI techniques to solve complex problems related to overbreak. Hence, this review paper aims to present the state of the art about the application of SC/AI techniques for predicting tunnel overbreak and discuss the advantages and disadvantages of the used SC/AI techniques. First, the background and reasons for the formation of overbreak are introduced. Then, the mechanism of influential parameters governing the overbreak is explained. These parameters can be divided into three categories: controllable parameters such as tunnel blasting design parameters, uncontrollable parameters such as rock mass properties and geological characteristics, and semi-controllable parameters such as tunnel geometry and size. Then, the SC/AI techniques used for estimating overbreak are summarized, while these techniques include single and hybrid models. The details of these models for predicting overbreak are reviewed. Subsequently, performance assessments about the advantages and disadvantages of the used SC/AI algorithms have been described, respectively, to compare the accuracy level of these algorithms in estimating overbreak and then introducing the most powerful AI/SC techniques to be used in this area. Finally, a new perspective that uses physics-based machine learning techniques to estimate overbreak is proposed.