Spectrally Efficient Peer-Peer Networking for Enhanced Distributed Predictive Maintenance

Abstract

Spectrally Efficient Peer-to-Peer Distributed Predictive Maintenance (SEPP-DPM) is a decentralized architecture designed for resilient, scalable system health monitoring in communication-constrained or contested environments. Unlike traditional predictive maintenance frameworks that rely on centralized data aggregation and processing, SEPP-DPM distributes both learning and inference across a network of edge nodes. Each node trains local autoencoder-based health models on sensor data and exchanges compact model updates through spectrally efficient, peer-to-peer communications rather than raw data. Coordination is achieved via the Kademlia Distributed Hash Table (DHT), enabling asynchronous and fault-tolerant model synchronization across heterogeneous nodes. The system integrates a dual-model approach consisting of the Distributed System Health Model (DSHM) and the Distributed Communication Channel Model (DCCM), jointly capturing equipment behavior and RF channel dynamics. A digital twin simulation environment is employed to evaluate performance under non-stationary, regime-dependent degradation conditions, demonstrating that reconstruction loss from decentralized autoencoders reliably tracks system degradation and supports real-time Remaining Useful Life (RUL) estimation. Experimental results show that SEPP-DPM maintains high diagnostic fidelity and operational awareness even under degraded connectivity, providing fleet-level health indicators without centralized coordination. This approach aligns with modern defense and industrial objectives—such as Joint All-Domain Command and Control (JADC2)—by enabling autonomous, edge-based predictive maintenance across distributed assets. SEPP-DPM represents a significant advancement in resilient machine learning architectures, coupling spectrum-efficient communications with decentralized intelligence for predictive maintenance in next-generation tactical and industrial networks.