Federated Learning for Cost Optimized Offloading in Edge-enabled Industrial Internet of Things

Industrial Edge Computing (IEC) is a viable solution for realizing tasks generated by the Industrial Internet of Things (IIoT) at the network's edge. Although resource optimization can be done at IEC, it is also crucial to reduce the overall execution cost and maintain the trade-off between energy consumption and service delay as these metrics are critical in the current industrial automation process and subsequent upcoming industrial revolutions. Considering these requirements, in this work, we design a cost-optimized framework for efficient task offloading in IIoT, specifically focusing on task classification, device selection, and task offloading issues. Our proposed framework named FedOff aims to concurrently manage industrial dynamic service requests and IEC resources simultaneously. To achieve this, we formulate an objective function as an energy-delay cost optimization problem for industrial networks. At first, we develop a heuristic task categorization technique called De-Sparse to select and execute delay-sensitive industrial tasks on networks' edge. Subsequently, we leverage federated learning to identify the most suitable IEC servers for task offloading by analyzing the historical non-Independent and Identically Distributed (IID) dataset generated by the IIoT devices. Experimental results show that our proposed FedOff strategy is cost-efficient and exhibits shorter delays in IIoT networks compared to the existing methods.