Association Mining for Operation and Maintenance Safety Risks of EMUs Based on Unstructured Event Data

Abstract The safety features of Electric Multiple Units (EMUs) are intricate and redundant, and the associated data is massive, multi-sourced, heterogeneous, and interdisciplinary. Constructing appropriate safety feature quantities by fully and effectively utilizing this data is a prerequisite for establishing a safety prevention and control network for EMUs. This paper proposes a model that matches risks in the operation and maintenance safety of EMUs with associated unsafe events, utilizing regular expression and pattern-matching technologies. The relationship between these risks and unsafe events is thoroughly analyzed and mined based on unsafe event data analysis. The paper presents a data-driven method for risk assessment that effectively tackles the issue of subjective bias in existing studies that rely on expert evaluations. The method automatically extracts key risk information, such as the likelihood and severity of consequences, identifies high-risk elements, and scientifically measures the safety risks of EMUs.