IIoT Based Hierarchical Data Distribution Strategy over Edge and End Device

In this paper, we propose a novel strategy of Industrial Internet of Things (IIoT) based hierarchical data distribution over edge and end device. Machine learning tasks such as workpiece classification based on image recognition often need to be executed at edges and end devices in an IIoT setting, where end devices have limited computing resources which are abundant in edge server. Thus pre-trained shallow neural network (SNN) and deep neural network (DNN) can be deployed at end devices and edge respectively. Traditional data distribution strategy allocates all data to SNN model deployed at end device while data that fails to classify at end device is transmitted to the edge server. However, this traditional strategy will lead to the edge server being idle and the data processing time being too long, which is difficult to meet the requirements of time-sensitive tasks. Therefore, each end device needs to make appropriate data distribution decisions to shorten the data processing time. In this paper, considering the complexity and the error rate of the machine learning model, the computing capability of the end device and the edge server, an optimization problem is formulated to minimize the data processing time. The optimal solution is obtained by analysis and calculation, and the optimal data allocation strategy is proposed. The simulation experiment is designed to deeply study the effects of crucial parameters on the allocation strategy and demonstrate the effectiveness of the proposed strategy.