A Sustainable AI-based Solution for Workplace Injury Classification in Industry 5.0

Industry 5.0 promotes a human-centric and sustainable vision of industrial development, where workplace safety is a key priority alongside technological progress and environmental protection. To design effective safety rules and preventive measures , it is crucial to collect as much information as possible from past accidents. However, accident reports-typically written in multiple languages and according to different regulations-often lack standardized detail, making it difficult to assess severity and extract consistent insights. Relevant information may appear in one report but be absent in another, requiring experts to manually review and integrate data to build a comprehensive overview. This process is time-consuming, prone to subjectivity, and limits scalability, highlighting the need for more efficient approaches to safety information integration. To address this challenge, we defined a lightweight approach for analyzing and processing accident reports that combines large language models with machine learning. The system was tested on the specific task of injury severity classification and integrated into an Active Learning scheme, which selects the most informative samples for model training, making the approach data-efficient and more environmentally sustainable. A real-world case study demonstrated the effectiveness of the proposed method, highlighting its potential to support accurate, scalable, and green safety analysis.