Intelligent Drone-Assisted Fault Diagnosis for B5G-Enabled Space-Air-Ground-Space Networks

• Published in: IEEE transactions on network science and engineering Vol. 8; no. 4; pp. 2849 - 2860
• Main Authors: Wang, Xiaoding, Lin, Hui, Zhang, Hongyan, Miao, Dejun, Miao, Qinyang, Liu, Wenxin
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.10.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Beyond 5G; Buoys; Circuit faults; Cloud computing; Data collection; Deep learning; deep reinforcement learning; Drones; Edge computing; Energy costs; Fault detection; Fault diagnosis; Kalman filters; Machine learning; multiple-access edge computing; Network reliability; Neural networks; Quality of service; Reinforcement learning; Satellites; Servers; space-air-ground-sea networks
• ISSN: 2327-4697; 2334-329X
• DOI: 10.1109/TNSE.2020.3043624

The ubiquitous network services provided by the Beyond 5G enabled space-air-ground-sea networks (B5G-SAGS) depends on the reliability of each intelligent device within. However, the QoS of B5G-SAGS could be compromised if there exists faults on individual network. That suggests the significance of fault-diagnosis in the B5G-SAGS design. Previous works on fault diagnosis were designed without extra information to improve diagnosis accuracy. In this paper, we propose an Intelligent Drone-assisted Fault-diagnosis Algorithm (IDFA) utilizing B5G-enabled Multiple-access Edge Computing/Cloud (B5G-MEC) services to detect faulty buoys. Specifically, IFDA first employs a Cubature Kalman Filter based Radial Bias Function Neural Network (CKF-RBFNN) for each fault-diagnosis center to perform preliminary fault detection based on the data provided by both buoys and drones. The data collection path is planned utilizing the deep reinforcement learning algorithm, Deep Deterministic Policy Gradient (DDPG), on B5G-MEC servers for energy efficiency. Eventually, the collective decision made by all fault-diagnosis centers determines the faulty status of each buoy. The theoretical analysis and validation experiments show that: (i) the IDFA has a better diagnosis accuracy in both single fault detection and multi-fault classification while compared with contemporary algorithms; (ii) the IDFA obtains a high aggregation ratio and a low energy cost.