A Novel Hybrid Neural Network with Attentive Feature Selection for Degradation Status Identification of Aircraft Self-locking Nuts

The self-locking nuts are widely used to connect structures in aerospace assembly lines. As these parts are usually installed in regions that suffered from heavy shocks and vibrations, the status of such essential parts closely relates to the safety and reliability of the aircraft. To enable precise sensing of these nuts and improve the system reliability, this paper proposes a hybrid neural network with a novel feature selection module for the identification of its degradation status. Specifically, the temporal tendency information and spatial fault patterns of monitored degradation torques are captured through a long-short-term memory (LSTM) network and a stacked Convolutional neural network (CNN) respectively. Besides, to effectively integrate these dual networks, a novel attention module absorbing the temporal features is proposed to reweight the spatial convolutional features. In particular, to explore fault information in the presence of multiple monitored torques, a regularized multi-task classifier is introduced to learn diverse representations. Experiments based on an industrial self-locking dataset proved that the proposed method possesses an accurate identification capability of degradation status than conventional neural networks.