DeepLight-RPL: Context-aware Adaptive RPL with Lightweight Deep Learning for Improving the QoS in Industrial IoT Application Scenarios

Internet of Things (IoT) networks are the corner-stone of Industry 4.0, enabling seamless real-time data exchange through resource-restricted wireless devices or sensors.Although Routing Protocol for Low-Power and Lossy Networks (RPL) is ideal for resource-limited industrial environments,conventional Objective Functions (OF) face critical restrictions, including suboptimal routing, limited resilience, and reduced responsive-ness to dynamic network conditions.Exploiting edge-based QoS optimization solutions is insufficient, as numerous performance disruptions emerge from within the network, affecting local routing behaviors.Deploying lightweight on-node distributed intelligence is crucial to facilitate early, localized adaptation and response without overwhelming computational and energy resources of Industrial IoT (IIoT) nodes.Therefore, this work introduces a DeepLight-RPL, an enhanced context-aware RPL-OF integrated with lightweight Deep Learning (DL) to optimize dynamic QoS-driven decision-making within RPL nodes for IIoT.This work improves the design of context-aware OF by continuously adapting to real-time IIoT conditions, fusing critical application and environmental QoS parameters to enhance learning accuracy and RPL parent selection efficiency.A lightweight hierarchical learning model is utilized to build resource-limited RPL industrial nodes with a wider network perspective, com-bining the strengths of hierarchical BiLSTM and Q-learning.Knowledge distillation optimizes the efficiency of the lightweight BiLSTM model, capturing temporal patterns of the network.Q-learning updates policies based on real-time network patterns adaptively and rewards feedback, guaranteeing efficient QoS-aware forwarding.This approach ensures efficient, adaptive, and dynamic QoS-aware routing without increasing computational overhead, effectively addressing IIoT routing challenges.Finally, experimental results demonstrate notable improvements in RPL routing, showcasing the capability of DeepLight-RPL to address the stringent and dynamic QoS performance demands of RPL-enabled Industry 4.0.For instance, the performance evaluation is validated under various QoS scenarios; the compact DeepLight-RPL improves 6.8% of PDR under a low-latency QoS scenario for a smart factory and 8.4% of PDR under energy-demanding QoS smart warehousing for 25 node density, ensuring optimized real-time low-latency and energy-efficient QoS communication in IIoT.