Deep learning based magneto failure detection in Diamond DA20-C1 aircraft

In this study, a deep learning-based system was developed to detect magneto failures that may occur during flight in the Continental IO-240-B32B engine of the Diamond DA20-C1 aircraft. Features were extracted from engine sounds using Mel- Frequency Cepstral Coefficients and classified via Convolutional Neural Networks. The system is capable of distinguishing between dual-magneto and single-magneto operating conditions. Within the scope of the study, 30-second audio recordings were collected for every 100 revolutions per minute increment within the 1000 to 2300 revolutions per minute range, resulting in a total of 39 raw data samples. To address class imbalance, data augmentation techniques were applied to the dual-magneto sound class. The developed model achieved an accuracy rate of 98.23% on the test dataset and 84.62% in an independent trial. While the model performed particularly well in the single-magneto class, it encountered difficulties at specific revolutions per minute values within the dualmagneto class. This system has the potential to provide pilots with real-time magneto status information during flight, offering an early warning and serving as a significant alternative to existing procedures limited to ground tests.