AI-Driven Anomaly Detection in Quadcopters Using ADXL345 Accelerometer Vibration Data and IoT Integration

This study investigates artificial intelligence methods for offline anomaly detection in quadcopters to improve flight safety. Vibration data were collected using ADXL345 accelerometers interfaced with ESP32 modules. Eight time-domain features were extracted from triaxial acceleration signals. Four machine learning classifiers—Random Forest (RF), Support Vector Machine, K-Nearest Neighbors, and Neural Networks—were trained and evaluated on a dataset representing a healthy state and four propeller damage levels (10% to 40% cuts). The RF classifier achieved the highest accuracy of 98% using standard deviation features. The results demonstrate the effectiveness of time-domain features and tree-based models for propeller fault diagnosis. This benchmarking approach enables precise identification and quantification of propeller damage severity, supporting rapid maintenance decisions and proactive flight risk management for UAV platforms.