Link-Aware Energy-Frugal Continual Learning for Fault Detection in IoT Networks

The use of lightweight machine learning (ML) models in internet of things (IoT) networks enables resource constrained IoT devices to perform on-device inference for several critical applications. However, the inference accuracy deteriorates due to the non-stationarity in the IoT environment and limited initial training data. To counteract this, the deployed models can be updated occasionally with new observed data samples. However, this approach consumes additional energy, which is undesirable for energy constrained IoT devices. This letter introduces an event-driven communication framework that strategically integrates continual learning (CL) in IoT networks for energy-efficient fault detection. Our framework enables the IoT device and the edge server (ES) to collaboratively update the lightweight ML model by adapting to the wireless link conditions for communication and the available energy budget. Evaluation on real-world datasets show that the proposed approach can outperform both periodic sampling and non-adaptive CL in terms of inference recall; our proposed approach achieves up to a 42.8% improvement, even under tight energy and bandwidth constraint.

💡 Research Summary

The rapid proliferation of Internet of Things (IoT) networks has driven the demand for lightweight machine learning (ML) models capable of performing on-device inference. This approach is crucial for latency-sensitive and privacy-preserving applications. However, two fundamental challenges hinder the long-term deployment of these models: the non-stationarity of IoT environments, which leads to significant accuracy degradation over time, and the extreme resource constraints of IoT devices, particularly regarding energy consumption. While Continual Learning (CL) offers a way to update models with new data, traditional periodic update strategies are prohibitively expensive in terms of energy and bandwidth for battery-powered devices.

To address these challenges, this paper proposes an innovative “Link-Aware Energy-Frugal Continual Learning” framework designed specifically for efficient fault detection in IoT networks. The core innovation lies in an event-driven communication mechanism that moves away from rigid, periodic updates toward a strategic, adaptive approach. The proposed framework integrates the decision-making process of model updates with the real-time physical constraints of the network, specifically focusing on wireless link conditions and the available energy budget of the IoT device.

The framework operates through a collaborative architecture involving both the IoT device and an Edge Server (ES). By intelligently monitoring the communication bandwidth and the device’s remaining energy, the system determines the optimal moments to trigger model updates. This “link-aware” strategy ensures that data transmission occurs when the network is most efficient, while the “energy-frugal” aspect prevents the depletion of the device’s battery. This collaborative approach allows the heavy lifting of the continual learning process to be distributed, leveraging the edge server’s computational power while minimizing the burden on the resource-constrained IoT node.

Experimental evaluations conducted on real-world datasets demonstrate the superiority of the proposed approach. Compared to conventional periodic sampling and non-adaptive continual learning methods, the proposed framework achieves a significant boost in inference recall, with improvements of up to 42.8%. Remarkably, these performance gains are maintained even under stringent energy and bandwidth constraints. This research provides a scalable and sustainable blueprint for maintaining high-accuracy, intelligent fault detection in dynamic and resource-limited IoT ecosystems, bridging the gap between advanced machine learning requirements and the physical realities of IoT deployment.