AI-Driven Fault Detection and Health Monitoring in Integrated Modular Avionics for Unmanned Aerial Systems

In recent years, the focus has increased to integrate the artificial intelligence (AI) to Unmanned Aerial Systems (UASs) for the improvement of health monitoring and fault detection system. The paper presents a hybrid deep learning architecture constituted of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks trained upon the synthetic sensor data like accelerometer signals, thermal features, voltage patterns, and communication anomalies for the application of fault detection. The aim of the research work is to develop a modular fault detection system that can be modeled in the software-defined avionics environment so that it can be deployed and experimented. To simulate real-world subsystem anomalies, various fault injection events were introduced to the simulation, including thermal spikes, voltage drops, and increased lost packets. The dataset was produced as a time-series windowed dataset, and the CNN-LSTM model was trained and cross-validated upon a 70:30 train-test split, achieving a 98.5% accuracy upon testing through a corresponding area under the receiver operating characteristic curve (ROC curve) of 0.99, showing strong classification capability. The confusion matrix reported a true positive of 96% for fault detection, showing the model’s resilience in detecting abnormal activity. This work introduces a new viewpoint of using interpretable AI in avionics systems and suggests that the simulation-based framework proposed can enhance the accuracy of the diagnostics as well as result in more dependable and explainable modular avionics architectures.