Anomaly Detection and Remaining Useful Life Estimation for the Health and Usage Monitoring Systems 2023 Data...

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Gear fault detection and remaining useful life (RUL) estimation are critical for rotating machinery health monitoring.
• Existing datasets may not fully capture the complexities of endurance gear vibrations.
• Anomaly detection algorithms are essential for identifying early signs of gear failure.

Purpose of the Study:

• To introduce a new benchmark dataset for endurance gear vibration signals.
• To evaluate the performance of traditional signal processing versus deep learning for anomaly detection.
• To present a novel digital twin for RUL estimation.

Main Methods:

• Development and release of a new endurance gear vibration signal dataset.
• Comparative analysis of traditional signal processing and deep learning anomaly detection algorithms.
• Implementation and validation of a new signal processing-based anomaly detection algorithm.
• Introduction of a digital twin for RUL estimation.

Main Results:

• Traditional signal processing techniques outperformed deep learning algorithms on the new benchmark dataset.
• The proposed signal processing anomaly detection algorithm surpassed a standard deep learning approach.
• The digital twin effectively estimated the RUL of gears from the benchmark dataset.
• Out of 11 participating groups in the HUMS 2023 challenge, traditional methods yielded superior results.

Conclusions:

• Traditional signal processing offers a robust and effective approach for gear fault detection and RUL estimation.
• The new benchmark dataset provides a valuable resource for advancing anomaly detection research in rotating machinery.
• The developed digital twin shows promise for real-time RUL prediction and proactive maintenance strategies.