Real-time fault detection in industrial rotating machinery requires both accurate machine learning models and software frameworks capable of handling continuous sensor streams. This study introduces FD-REST, an open-source, Dockerized platform that enables the deployment, execution, and real-time visualization of multi-sensor fault diagnosis models. The system integrates vibration, ultrasound, and temperature features and employs a Deep Neural Network (DNN) to generate continuous fault similarity scores across eight mechanical conditions. All predictions and raw signals are streamed to the frontend via WebSockets and stored in a lightweight SQLite database for reproducibility, session replay, and report generation. The embedded DNN model was validated on a real-world multi-modal dataset and achieved strong predictive performance, including a Mean Squared Error (MSE) of 0.00253, an score of 0.8436, and approximately 93% threshold-based classification accuracy. These results demonstrate both the numerical reliability of the model and the effectiveness of FD-REST as a streaming-oriented benchmarking environment. By providing a modular, reproducible, and on-premises-ready framework, FD-REST bridges the gap between offline algorithm development and real-time industrial deployment, offering a practical tool for researchers, engineers, and practitioners in predictive maintenance.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号