Efficient Feature Learning Approach for Raw Industrial Vibration Data Using Two-Stage Learning Framework.

Area of Science:

• Machine Learning
• Data Science
• Industrial IoT

Background:

• Data-driven methods and machine learning are increasingly vital in industry.
• Acquiring large labeled datasets for these methods is costly and challenging.
• Few-shot learning offers a promising alternative for data-scarce scenarios, especially in time series analysis.

Purpose of the Study:

• To develop an efficient two-stage feature learning approach for anomaly detection in machine processes.
• To address the challenge of limited labeled data in time series analysis.
• To enhance the reliability of anomaly detection in industrial settings.

Main Methods:

• A prototype few-shot learning technique requiring minimal labeled samples.
• A two-stage feature learning strategy tailored for time series data.
• Evaluation on a real-world CNC Machining dataset.

Main Results:

• The proposed method achieved a high F1-score of 90.3%, outperforming conventional methods.
• Feature analysis confirmed strong generalization ability and robustness.
• Deep features demonstrated invariance to data shifts across machines and time.

Conclusions:

• The developed few-shot learning approach is effective for anomaly detection with limited data.
• The method offers robust and generalizable deep features for industrial sensory applications.
• This technique provides a reliable solution for anomaly detection in dynamic machine environments.