This paper presents an embedded machine learning fault detection system for electric fan drives. The system uses a Convolutional Neural Network (CNN) to monitor fan vibrations via an accelerometer and detect anomalies in real-time. The CNN model was trained on vibration data from three fan states: Fan-On, Fan-Off, and Fan-Fault. The model achieved high accuracy (99.8%, 99.9%, and 100.0% for the respective states) on a validation dataset, with real-time testing showing accuracy ranging from 90% to 100%. The CNN model was optimized for efficiency, achieving optimal accuracy within 30 epochs and requiring no manual feature extraction. It was deployed on an embedded system, demonstrating its effectiveness in real-time fault detection. The system addresses challenges such as algorithm selection, real-time deployment on embedded systems, hyperparameter tuning, sensor integration, and energy efficiency. The CNN model outperformed traditional machine learning algorithms like SVM, K-Nearest Neighbors, Random Forest, and Gradient Boosting in terms of accuracy and efficiency. The model's ability to automatically extract features from raw vibration data made it more effective than models requiring manual feature engineering. The system was tested on a real-world dataset collected from a small DC fan, with data collected for 10 minutes per class. The dataset was preprocessed, normalized, and split into training and test sets. The CNN model was trained on this data and deployed on an embedded device, demonstrating its potential for broader applications in industrial and smart sensing systems. The model's performance was evaluated using accuracy, precision, recall, and F1-score, with the CNN model achieving the highest results. The study highlights the effectiveness of CNNs for vibration signal analysis and their potential for real-time fault detection in industrial settings. The model's cost-effective design and adaptability to different fan configurations make it a promising solution for industrial fan health monitoring.This paper presents an embedded machine learning fault detection system for electric fan drives. The system uses a Convolutional Neural Network (CNN) to monitor fan vibrations via an accelerometer and detect anomalies in real-time. The CNN model was trained on vibration data from three fan states: Fan-On, Fan-Off, and Fan-Fault. The model achieved high accuracy (99.8%, 99.9%, and 100.0% for the respective states) on a validation dataset, with real-time testing showing accuracy ranging from 90% to 100%. The CNN model was optimized for efficiency, achieving optimal accuracy within 30 epochs and requiring no manual feature extraction. It was deployed on an embedded system, demonstrating its effectiveness in real-time fault detection. The system addresses challenges such as algorithm selection, real-time deployment on embedded systems, hyperparameter tuning, sensor integration, and energy efficiency. The CNN model outperformed traditional machine learning algorithms like SVM, K-Nearest Neighbors, Random Forest, and Gradient Boosting in terms of accuracy and efficiency. The model's ability to automatically extract features from raw vibration data made it more effective than models requiring manual feature engineering. The system was tested on a real-world dataset collected from a small DC fan, with data collected for 10 minutes per class. The dataset was preprocessed, normalized, and split into training and test sets. The CNN model was trained on this data and deployed on an embedded device, demonstrating its potential for broader applications in industrial and smart sensing systems. The model's performance was evaluated using accuracy, precision, recall, and F1-score, with the CNN model achieving the highest results. The study highlights the effectiveness of CNNs for vibration signal analysis and their potential for real-time fault detection in industrial settings. The model's cost-effective design and adaptability to different fan configurations make it a promising solution for industrial fan health monitoring.