Fatigue detection and optimisation method for human–robot collaborative assembly based on digital twin

Fatigue detection in human-robot collaborative (HRC) assembly exemplifies the ‘Human-centered manufacturing’ principle in Industry 5.0. However, most current fatigue detection methods depend on a single data source, which hinders comprehensive quantitative assessment and scheme optimisation for different assembly workers. Therefore, this study proposes a digital twin-based method for detecting and regulating physical fatigue in human–robot collaborative assembly. First, an information model for HRC assembly is developed, capturing basic data among personnel, equipment, and processes. Then, high-load workstations are identified using Stochastic Petri Nets (SPN) and their isomorphic Markov chains. Next, for these high-load workstations, techniques like machine vision and deep learning are employed to perceive multi-dimensional fatigue states from three data types: eye closure data, joint load data, and surface electromyography (sEMG) fatigue data of workers. Using this data, the Analytic Hierarchy Process (AHP) and Fuzzy Comprehensive Evaluation (FCE) are applied to integrate and quantitatively assess the multi-source fatigue indices. Finally, a digital twin system is constructed, and experimental verification is performed.