Learning-based Co-Design of Distributed Edge Sensing and Transmission for Industrial Cyber-Physical Systems

Industrial cyber-physical systems (ICPS) refer to an emerging generation of intelligent systems, where distributed data acquisition is of great importance and is influenced by data transmission. In the improvement of the overall performance of sensing accuracy and energy efficiency, sensing and transmission are tightly coupled. Due to the unknown transmission channel states in the harsh industrial field environment, intelligently performing sensor scheduling for distributed sensing is challenging. In this paper, edge computing technology is utilized to enhance the level of intelligence at the edge side and deploy advanced scheduling algorithms. We propose a learning-based distributed edge sensing-transmission co-design (LEST) algorithm under the coordination of the sensors and the edge computing unit (ECU). Deep reinforcement learning is applied to perform real-time sensor scheduling under unknown channel states. The conditions for the existence of feasible scheduling policies are analyzed. The proposed algorithm is applied to estimate the slab temperature in the hot rolling process, which is a typical ICPS. The simulation results demonstrate that the overall performance of LEST is better than other suboptimal algorithms.