Cluster Analysis of Risk Factors from Near-Miss and Accident Reports in Tunneling Excavation

Abstract

Many inherently risky industries improve their safety management by learning from near-miss incidents. The construction industry is starting to manage near-miss incidents for improving safety, and several studies have been performed to introduce a system to manage near-miss incidents during construction. However, an analytic framework to technically investigate near-miss events remains missing. In this research, a structuralized analysis of safety events (including both near misses and accidents) in metro tunnel construction was presented. A number of incidents (57 accidents and 186 near-miss events) were collected, and these crude reports were compiled through qualitative analysis into a database of safety events. The database, with both categories and variable definitions incorporated, served as the basis for quantitative analysis. Groups of events were mined to figure out whether they were similar or identical through cluster analysis. The entries in the database were divided into clusters by the iterative self-organization data analysis (ISODATA) algorithm based on the variables defined. For each level of outcome severity, the risk potential of each cluster was compared with that of other clusters and the whole database; thus, the magnitude of the risk potential of the cluster under consideration was quantified. The analysis showed that the biggest risk factors in metro tunneling excavation were in (1) improper soil reinforcement and drainage at the launching or arrival portal and (2) soil instability of the tunneling face. The developed approach in this research can be used as a decision tool to provide insights for better interpreting characteristics and patterns of the identified clusters (within different levels of risk potential) mined from the historical near-miss and accident reports in tunnel construction.