Visualizing what’s missing: Using deep learning and Bow-Tie diagrams to identify and visualize missing leading indicators in industrial construction

Abstract

Introduction: In the construction industry, where safety is paramount, the frequency and severity of workplace incidents remain critical concerns. Therefore, site safety inspections have become essential for health and safety programs. While incident data is frequently used to identify gaps in the safety management system, inspection reports are rarely analyzed to identify unsafe patterns on site and reveal measures for safety enhancement. This limitation can reduce the effectiveness of safety inspections, and therefore, this study aims to identify what safety leading indicators do not capture hazards during inspections. Methods: Natural language processing (NLP), text mining, and deep learning techniques such as sentence bidirectional encoder representations from transformers (SBERT) are used to generate embeddings and compute the similarity between 633 incidents and 9,681 inspection descriptions of a construction project from 2015 to 2018 in Canada. Root cause analysis is conducted on selected incidents with the slightest similarity with inspection descriptions using a customized human and organizational framework. Bow-tie and Sankey’s diagrams illustrate and visualize what leading indicators miss capturing hazards during inspections that lead to incidents. In addition, N-gram models are used for validation and co-occurrence networks to extract meaningful information and identify patterns from incident and inspection reports. Results: The results demonstrate that the indicators that cause incidents with the most severe consequences and are inadequately captured during inspections are: working at heights (81%), equipment handling/storage (17%), and ergonomics (0.4%). Conclusion and practical application: The findings provide insights for decision-makers on the strategies needed to enhance risk management, facilitating predictive and proactive approaches. By embracing a transdisciplinary approach, the research techniques applied in this study can be effectively used and transferred across various other industries.