Decentralized Edge Intelligence via Dynamic Federated Learning for Predictive Maintenance in Distributed Smart Supply Chains

• Ada John

  ladoke Akintola university of technology

  Author

Abstract

Predictive maintenance in distributed smart supply chains is constrained by data silos, privacy concerns, and the latency of centralized cloud architectures. While federated learning (FL) enables collaborative model training without raw data sharing, conventional FL frameworks rely on static aggregation strategies that fail to adapt to heterogeneous edge conditions and non-IID data distributions characteristic of supply chain environments. This study designs and validates a Dynamic Federated Learning (DFL) framework for decentralized edge intelligence in distributed smart supply chains, integrating adaptive aggregation weighting, trust-based node participation, and hybrid machine learning models deployed at the network edge. Using a design-based research methodology combining retrospective analysis of supply chain sensor data (n=47,000 operational records from 23 edge nodes) and prospective simulation, the proposed framework achieved 89.4% predictive accuracy for equipment failure classification, outperforming static FedAvg (82.1%) and centralized LSTM baselines (85.7%). The DFL framework reduced communication overhead by 37.2% and maintained prediction stability under simulated node dropout conditions. Trust-weighted aggregation proved critical for mitigating performance degradation from heterogeneous data quality across supply chain partners. The findings establish a replicable architecture for privacy-preserving, latency-sensitive predictive maintenance in distributed logistics networks, with implications for supply chain resilience and cross-enterprise collaboration.

• pdf

Copyright (c) 2026 Ada John (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.