LiReD: A Light-Weight Real-Time Fault Detection System for Edge Computing Using LSTM Recurrent Neural Networks

Area of Science:

• Industrial IoT and Smart Manufacturing
• Machine Learning for Predictive Maintenance
• Edge Computing Architectures

Background:

• Real-time monitoring and fault detection are crucial for smart factories, but massive data processing demands significant resources.
• Cloud-based analysis is viable but less efficient than edge computing for immediate data handling.
• Edge devices offer faster processing, reduced network load, and localized data analysis.

Purpose of the Study:

• To develop an efficient edge computing system for real-time machine fault detection in smart factories.
• To implement a data processing and analysis framework on edge devices.
• To evaluate the performance of a Long Short-Term Memory (LSTM) based fault detection model.

Main Methods:

• Construction of an edge device using a single-board computer and sensors for data collection, processing, storage, and analysis.
• Development of a machine fault detection model utilizing Long Short-Term Memory (LSTM) recurrent neural networks.
• Implementation and testing of the LiReD system on an industrial robot manipulator.

Main Results:

• The edge device successfully collected, processed, stored, and analyzed sensor data.
• The LSTM-based fault detection model demonstrated superior performance compared to six other fault detection models.
• The LiReD system proved effective for real-time fault detection in an industrial setting.

Conclusions:

• Edge computing, implemented via the LiReD system, offers an efficient solution for smart factory monitoring and fault detection.
• LSTM networks are highly effective for developing accurate machine fault detection models in edge computing environments.
• The proposed system enhances operational efficiency and reduces network costs in industrial automation.