Local Outlier Factor와 의사결정나무 알고리즘을 활용한 이상사고 주요원인 분석

Abstract

As industrial technology becomes more sophisticated and production process becomes more precise and more complicated, it is becoming more difficult to analyze the process of occurrence, causes of accidents. Above all, incidents that occur within an acceptable range, such as frequent occurrences of accidents, are managed appropriately, whereas incidents and methods of managing incidents that do not occur frequently are not being performed well. In particular, the management of abnormal accidents using numerical data such as temperature, vibration, and use time of mechanical equipment or process is relatively much studied, but abnormal accident management using qualitative textual information of accident report documents written by the safety manager is not attempted. In this paper, we propose a Local Outlier Factor (LOF) and a Decision Tree (DT) algorithm to analyze the main causes of abnormal accident scientifically using text data included in the accident report document. Specifically, we construct an accident document - keyword matrix by applying text mining based on natural language processing to the accident report document, and find abnormal accident through LOF. Then, we propose an analytical method to derive the main cause of abnormal accidents by using DT. LOF algorithm is a density-based outlier detection technique that utilizes the concept of determining the abnormal value differently according to density in multi-dimensional, and is used as a technique to analyze the abnormal value in detail. Also, DT algorithm is an analytical technique that classifies decision rules into tree form, and is used as an analytical model that can easily analyze analytical factors systematically and hierarchically. To do this, a case study is the chemical industry, which deals with a variety of hazardous substances and has complex, technologically intensive processes. It analyzes systematically the main causes of abnormal accidents in chemical process and classifies the causes of the accidents according to process safety indicators. This will contribute to providing a guideline for the safety manager to manage the main causes of abnormal accidents.