Transfer-Learning-Based Estimation of the Remaining Useful Life of Heterogeneous Bearing Types Using Low-Frequency...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Estimating Remaining Useful Life (RUL) is crucial for predictive maintenance of mechanical components like bearings.
• Traditional methods often struggle with small datasets and low sampling rates, limiting practical applications.
• Deep learning offers potential but requires robust feature extraction and effective transfer learning strategies.

Purpose of the Study:

• To propose and evaluate a novel transfer learning-based deep learning approach for RUL estimation in bearings.
• To address challenges of small datasets and low sampling rates in bearing prognostics.
• To enable RUL estimation using low-cost sensors operating in a low-frequency range.

Main Methods:

• A transfer learning approach utilizing an intermediate domain based on bearing fault frequencies.
• Feature abstraction using convolutional layers followed by RUL estimation with a Long Short-Term Memory (LSTM) network.
• Fixed-feature extraction for efficient transfer learning across different bearing types.

Main Results:

• The proposed deep learning approach successfully processes low-frequency sensor data.
• Validation against the IEEE PHM 2012 Data Challenge demonstrated superior performance compared to the winning approach.
• The method shows significant effectiveness in transfer learning between diverse bearing types.

Conclusions:

• The developed deep learning approach is well-suited for RUL estimation using low-frequency sensor data.
• It offers an efficient and effective solution for transfer learning in bearing prognostics.
• The methodology supports the use of low-cost sensors for condition monitoring.