A Diversity-Preserving Multi-Objective PSO Framework for SFC Scheduling in a Digital Twin-Enabled IIoT Architecture Toward Industry 5.0

INTRODUCTION: Under Industry 5.0, discrete manufacturing faces a fundamental dilemma: it must balance human-centric flexibility with sustainable and resilient operations. Resolving this dilemma requires the Industrial Internet of Things (IIoT) to coordinate and integrate the three types of heterogeneous resources—sensing, computing, and communication—found in factory workshops, production facilities, and assembly lines. We present an IIoT architecture centered on the digital twin, integrating these capabilities into a single closed-loop feedback chain to resolve resource contention issues at the source,  that arise when various manufacturing tasks compete for limited computing power and network bandwidth. OBJECTIVES: The architecture is physically divided into four layers—end devices, access, edge, and cloud—with each layer coupled via a global feedback channel based on the digital twin, giving the manufacturing environment end-to-end visibility and adaptive resource allocation. For SFC scheduling, we formulate a multi-objective optimization model that simultaneously considers three metrics—total end-to-end task delay, system-wide energy consumption, and inter-node load variance—while imposing hierarchical resource constraints inherent to multi-tier industrial deployments. METHODS: The algorithm used to solve the above model is ADP-MOPSO, a multi-objective particle swarm optimization method that balances convergence and diversity. It integrates workload-aware heuristic initialization, a dynamic constraint repair operator, crowding-distance-based archive management, and roulette-wheel leader selection within a single PSO framework, whose strategies interact through an externally maintained elite archive. RESULTS: We tested three IIoT scales based on real-world manufacturing deployment parameters (small, medium, and large; 12/32/64 nodes, 4/8/16 edge servers) and compared against five benchmark algorithms. Across all three scales, ADP-MOPSO achieved the best Hypervolume (HV) and Inverted Generational Distance (IGD) values and remained competitive with the strongest baselines on Spacing (SP). Ablation experiments showed that heuristic initialization contributes a 6.8% HV improvement, and the diversity-preserving module adds a further 4.2%, yielding an overall gain of 10.2%. CONCLUSION: This study provides an integrated architecture-algorithm solution for real-time multi-objective resource optimization in Industry 5.0 manufacturing systems. Multi-scale simulations calibrated to industrial production settings confirm the effectiveness and scalability of the proposed approach.