Key points are not available for this paper at this time.
The rapid adoption of Internet of Things (IoT) devices and Edge computing has increased the demand for efficient processing of high-frequency sensor data. In soft real-time (RT) applications, excessive latency can significantly degrade system performance. While high-frequency data streams provide valuable insights, they also introduce network congestion, storage limitations, and computational overhead. This paper presents a novel IoT architecture wherein an Edge server dynamically down-samples sensor data before transmission to a Fog (central) server equipped with upsampling, feature extraction, machine learning (ML) inference, and network latency analysis services. The system leverages asynchronous messaging protocols (MQTT and ZMQ) and a Docker-based microservices framework to ensure scalable and efficient data processing. A Proportional-Integral-Derivative (PID) controller dynamically adjusts the downsampling rate based on network conditions, optimizing bandwidth while maintaining data integrity and keeping latency below pre-defined threshold. The proposed architecture is first tested with simulated data and then validated on an industrial conveyor belt system, demonstrating its effectiveness in predicting operational states through the integration of a Random Forest-based ML model. By balancing data reduction and analytical accuracy, this approach addresses key challenges in high-frequency data processing for industrial applications.
Barrena et al. (Thu,) studied this question.