.png){kind=icon}
.png){kind=icon}
Return
Multi-source heterogeneous data integration for incident likelihood analysis
.png){kind=icon}
PRE
AI
DOI：10.1016/j.compchemeng.2024.108677.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Structured data, such as sensor data, can provide valuable insights to safety practitioners for developing prevention and mitigation strategies. However, relying on a single data source can introduce biases. In this era of safety 4.0, a methodology that can leverage insights from multiple sources (incident databases and physical observations) is required. This study proposes an approach based on natural language processing (NLP) to learn lessons from past incidents and combine them with contemporary data to predict adverse events. The model is based on feature extraction using a co-occurrence network on the loss of containment (LOC)/release of hazardous substance accidents from 2002 to 2021, sourced from the Chemical Safety and Hazard Investigation Board (CSB) database. Coupled with the operational parameters, it provides a robust likelihood model. Scenario-based model verification is performed by simulated scenarios based on past incidents of LOC to assess model efficacy in predicting similar incidents. Sensitivity analysis shows inadequate written procedures resulting from management and organizational failure have the highest sensitivity towards LOC incidents. This work assists practitioners in monitoring sensor data and lessons learned from past incidents by utilizing multi-source heterogeneous data sources. Thus, the current research work serves as an important tool to enhance data-driven prediction as a part of safety 4.0.
Keywords:
Safety 4.0
Natural language processing (NLP) Chemical
Safety and Hazard Investigation Board (CSB)
Data-driven prediction
Journal
C
IF:
3.9
Papers:
8.1K
Citations:
1.7W