Document Type : Article
Authors
1
School of Mechanical Engineerig, College of Engineering, University of Tehran
2
School of Mechanical Engineering, Faculty of Engineering, University of Tehran
10.24200/j40.2025.65863.1730
Abstract
Manufacturing companies are facing increasing pressure to meet the growing demand for high-quality products while maintaining cost efficiency. A significant portion of production costs—around 60% to 70%—is attributed to the maintenance of industrial machinery. As a result, an effective and real-time method for fault detection has become critical to reducing maintenance costs, extending the lifespan of equipment, and minimizing production downtime. To address these challenges, this article presents a novel, high-performance anomaly detection framework based on edge computing technology for real-time health monitoring of industrial assets. The proposed system allows for the execution of essential tasks, including data acquisition, preprocessing, feature extraction, and algorithm training, directly on the edge node, specifically a Microcontroller Unit (MCU). This design is particularly challenging due to the MCU's limited processing power and memory capabilities. To monitor machine conditions, a 3-axis accelerometer is employed to capture vibration signals, which are then stored in the MCU's Flash memory for subsequent analysis. An autoencoder with three hidden layers is trained on the edge device to model the healthy state of the machinery. The reconstruction error of unseen data is then used to detect anomalies, indicating potential faults. To the best of our knowledge, this is the first study that trains an artificial neural network (ANN) entirely on an MCU and performs end-to-end condition monitoring on the edge, without relying on cloud computing or external servers. The effectiveness of the proposed system was validated through experiments on a centrifugal pump, where the framework successfully detected a variety of pump faults. The results demonstrated outstanding performance, with precision, recall, accuracy, and F1 scores exceeding 99.9%. This approach offers a cost-effective, efficient solution for real-time fault detection in industrial applications, paving the way for smarter, more reliable manufacturing processes. By enabling real-time, on-site monitoring of industrial assets, this solution has the potential to significantly reduce maintenance costs, enhance operational efficiency, and improve equipment reliability across a wide range of industries.
Keywords
Main Subjects
)