A Sources Reconstruction and Auxiliary Training Two-Stage Network for Fault Diagnosis of Antenna Arrays

A novel two-stage network for fault diagnosis of antenna arrays is proposed. The proposed two-stage network integrates two key components: a sources reconstruction network (SR-Net) designed to extract fault-related features from the input far-field radiation patterns and output the excitation amplitudes to identify the faulty element and an auxiliary training network (AT-Net) takes the excitation amplitudes as input and generates the corresponding far-field radiation patterns as output. By employing a two-stage supervised learning strategy, the one-to-many problem in the fault diagnosis network is effectively addressed, leading to improved diagnostic accuracy for antenna arrays. The proposed supervised sources reconstruction and auxiliary training two-stage network (SR-ATTN) not only accelerates the training process but also achieves higher fault diagnosis accuracy compared to the traditional single-network solution, as it introduces a new gradient descent path for improved optimization. Moreover, the concept of balanced datasets is introduced to address the imbalance between fault occurrence probabilities and the distribution of fault categories in the dataset, which significantly enhances the diagnosis accuracy. The proposed supervised SR-ATTN for fault diagnosis of antenna arrays is validated on a $3 \times 5$ dipole antenna array and a $5 \times 5$ microstrip antenna array, demonstrating high accuracy, real-time performance, robust resistance to noise, and strong generalization capabilities.