Intelligent Fault Diagnosis of Satellite Communication Antenna via a Novel Meta-learning Network Combining with Attention Mechanism

Shipborne satellite communication antenna which is used for remote control plays an irreplaceable role in ships, it is necessary to monitor its operation state. However, obtaining sufficient fault information in mass monitoring data is particularly difficult, which greatly degrades performance of existing intelligent algorithms. In this paper, a novel meta-learning network is proposed to realize state recognition of shipborne antenna under small samples prerequisite. The network is constructed to improve generalization even though inputs collected under different operating conditions. Meta-learning network consists of sampler, feature extractor, auxiliary classifier and discriminator. It trains an adaptive pseudo-distance to evaluate the degree of correlation between different data, then realize classification task. Feasibility and effectiveness of the network are verified by three bearing datasets. Results show that the proposed method uses few samples to successfully classify mechanical data of shipborne antenna even with different rotating speed and random noise.