Decentralized adaptive work package learning for personalized and privacy-preserving occupational health and safety monitoring in construction

Abstract

<p>Precision construction occupational health and safety (COHS) is a prerequisite for project success. Work package-based distributed monitoring shows a high capability for this purpose. However, a theoretical dilemma exists between larger work packages with greater technical efficiency and smaller ones with greater data privacy. This paper develops a <em>decentralized adaptive work package</em> (DAWP) learning model and blockchain for personalized COHS monitoring. The DAWP learning model is first formulated to form adaptive topologies to concatenate and share model parameters of work packages with their neighbors. DAWP learning can compute graphs using mixing weights and similarity to improve personalization. Then, studying <em>blockchain</em> can transform DAWP into a decentralized collaboration. Lastly, blockchain-DAWP (BC-DAWP) is evaluated by controlled experiments of multiple monitoring tasks. The results indicated that the BC-DAWP with lightweight models outperforms the proposed baselines in a personalized and privacy-preserving manner, which paves the way for the next-generation decentralized COHS monitoring.</p>