A Simulation-Based Framework for Fatigue-Aware Task Reallocation in Human–Robot Collaboration: Software Verification With 1000-Episode Digital Twin

Abstract

Collaborative robotics in Industry 4.0 manufacturing faces persistent challenges in balancing productivity with operator safety, particularly when human fatigue accumulates during extended work shifts. While physical validation remains the gold standard, computational simulation offers advantages for systematic algorithm development: perfect ground truth unavailable from self-reported fatigue measures, parameter sweeps across physiological ranges infeasible with human subjects, and controlled testing of extreme scenarios unsafe for physical trials. Unlike existing digital twin frameworks for human-robot collaboration (HRC) that treat operator capability as static or model it through coarse binary states, this work presents a simulation framework that integrates continuous physiological fatigue dynamics into the digital twin control loop, enabling fatigue-aware task reallocation algorithm development before deployment on actual collaborative robots. The framework implements a tricolor semaphore system (Green/Orange/Red thresholds at 30%/40% fatigue) with dynamic sampling rates (5.0 min baseline, 0.1 min critical monitoring) and automated robotic intervention when operator fatigue exceeds safe thresholds. Software verification across 1000 simulated episodes demonstrates 99.30% collision-free operation—approaching but not yet achieving the ISO/TS 15066:2016 safety target of 99.85%, with a 0.55 percentage point gap whose causes are characterized: 57% of collisions concentrate at episode termination (t= 45 min, fatigue \geq 68 %), indicating that extreme cumulative fatigue overwhelms intervention effectiveness. Friedman analysis confirms statistically significant fatigue progression (chi-squared= 3000.0, p<0.001) from 28.06% at baseline to 68.04% at 45 minutes, with all pairwise comparisons significant (p<0.001, Cohen’s d= 1.18–2.34). The framework achieves 100% intervention triggering when needed, with robots active during 50.28% of episode duration. Analysis reveals strong temporal coupling between skill and fatigue (r = +0.970): operators advance from Intermediate to Advanced competence while simultaneously accumulating physiological depletion. This correlation arises naturally from the model structure where both variables increase with time, demonstrating that growing expertise does not offset fatigue-related safety risks. Four targeted refinement strategies—predictive trajectory monitoring, graduated intervention intensity, forced micro-recovery protocols, and fatigue-velocity-based dynamic thresholds—are proposed to close the ISO/TS 15066 gap. The framework’s modular architecture separates physiological modeling from algorithmic decision logic, so that replacing the linear fatigue model with nonlinear alternatives requires only parameter-level changes, not system redesign. Results establish computational feasibility and provide baseline metrics for physical implementation with actual collaborative robots.