A Self-Adaptive 1D Convolutional Neural Network for Flight-State Identification

The vibration of a wing structure in the air reflects coupled aerodynamic⁻mechanical responses under varying flight states that are defined by the angle of attack and airspeed. It is of great challenge to identify the flight state from the complex vibration signals. In this paper, a novel one-dimension convolutional neural network (CNN) is developed, which is able to automatically extract useful features from the structural vibration of a recently fabricated self-sensing wing through wind-tunnel experiments. The obtained signals are firstly decomposed into various subsignals with different frequency bands via dual-tree complex-wavelet packet transformation. Then, the reconstructed subsignals are selected to form the best combination for multichannel inputs of the CNN. A swarm-based evolutionary algorithm called grey-wolf optimizer is utilized to optimize a set of key parameters of the CNN, which saves considerable human efforts. Two case studies demonstrate the high identification accuracy and robustness of the proposed method over standard deep-learning methods in flight-state identification, thus providing new perspectives in self-awareness toward the next generation of intelligent air vehicles.