Leveraging Lightweight AI for Anomaly Detection in Mechanical Power Transmission At The Edge

The integration of IoT and TinyML platforms has revolutionized fault detection and condition monitoring in electromechanical systems, enabling real-time data acquisition and analysis in resource-constrained environments. This study presents an implementation of TinyML model for mechanical power transmission fault detection on DC electric motors. The proposed model is deployed on 32-bit ARM Cortex-M4F microcontroller. A built-in 3-axis IMU (Inertial measurement Unit) fusion sensor is used to get the acceleration, angular rate and orientation. A custom-developed setup installed on the top end of the DC motor was utilized to collect real time data. This dataset comprised of eleven operation conditions, encompassing idle, noisy, loose, and misaligned states. Time-domain and frequency-domain features were extracted and used to train a Multi-Layer Perceptron (MLP) model, achieving an overall classification accuracy of 97.1%. The model’s deployment on the Arduino edge demonstrated real-time fault detection capabilities with 73.19% accuracy on unseen data, leveraging minimal computational resources. The results showcase the potential of TinyML-based systems for predictive maintenance, providing a cost-effective and scalable solution for industrial applications.