A Cotraining-Based Semisupervised Approach for Remaining-Useful-Life Prediction of Bearings

Area of Science:

• Mechanical Engineering
• Data Science
• Artificial Intelligence

Background:

• Bearing failures critically impact equipment safety and operational continuity.
• Deep learning excels in Remaining Useful Life (RUL) prediction due to its scalability and nonlinear fitting capabilities.
• Supervised deep learning demands extensive labeled data, which is often costly and time-consuming to acquire.

Purpose of the Study:

• To develop an innovative cotraining-based method for RUL prediction that minimizes the requirement for labeled data.
• To leverage unlabeled data effectively to enhance the accuracy of RUL predictions.
• To validate the proposed approach against existing state-of-the-art methods.

Main Methods:

• A cotraining framework was established, integrating a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network.
• The CNN and LSTM were trained collaboratively on a large volume of unlabeled data to generate a health indicator (HI).
• The derived HI was utilized with monitoring data for accurate RUL prediction.

Main Results:

• The proposed cotraining approach demonstrated superior performance compared to individual CNN and LSTM models in RUL prediction.
• The Root Mean Square Error (RMSE) achieved was 54.72, significantly lower than the SAE+LSTM (137.12) baseline.
• The RMSE value was comparable to the CNN+LSTM (49.36) method, indicating strong competitive performance.

Conclusions:

• The cotraining-based RUL prediction method effectively utilizes unlabeled data, reducing labeling costs and time.
• The approach shows significant potential for practical application, as evidenced by successful testing on a real-world task.
• This method offers a valuable alternative for accurate equipment health monitoring and RUL prediction.