Research on a Real-Time Pilot-Induced Oscillation Detector Based on a Convolutional Neural Network

This study aimed to develop a real-time Pilot-Induced Oscillation (PIO) detector by utilizing a Convolutional Neural Network (CNN) and applying it to flight test monitoring. For data training, a database of PIO samples is established. CNN has the advantage of lower complexity and automatic feature learning; it is chosen to develop the PIO detector. The results showed that the accuracy of the trained model has reached 93.71%, where the recall rate reaches 94.74%. To better assist the monitoring team, a simple software is designed based on the trained model for PIO detection, and it has been applied in flight test monitoring. In conclusion, this research demonstrates that the CNN Algorithm can be utilized in PIO discrimination and improve the monitoring capacity for flight testing.