Rolling bearing remaining useful life prediction using deep learning based on high-quality representation.

Area of Science:

• Engineering
• Computer Science

Background:

• Accurate remaining useful life (RUL) prediction is crucial for intelligent manufacturing and rotating machinery reliability.
• Challenges exist in representing bearing degradation states and capturing temporal dependencies for RUL prediction.

Purpose of the Study:

• To propose a novel deep learning approach for enhanced RUL prediction in rolling bearings.
• To address limitations in feature extraction and temporal dependency modeling for degradation assessment.

Main Methods:

• A one-dimensional deep convolutional autoencoder (1D-DCAE) was employed for high-quality feature extraction from vibration signals.
• A multilevel bidirectional long short-term memory (Bi-LSTM) network integrated with a temporal pattern attention (TPA) mechanism was utilized to capture temporal dependencies.
• Extracted health indicators (HIs) and self-labelled data were used as inputs for the prediction model.

Main Results:

• The proposed method demonstrated effective signal feature extraction, outperforming traditional labelling techniques.
• Experimental results on the PHM2012 bearing dataset showed higher prediction accuracy and robustness compared to existing methods.
• The model achieved superior performance in predicting the remaining useful life of rolling bearings.

Conclusions:

• The integrated 1D-DCAE and Bi-LSTM+TPA model offers a robust solution for RUL prediction in intelligent manufacturing.
• The approach effectively represents degradation states and captures complex temporal patterns in bearing data.
• The model's generalizability and transferability across diverse operating conditions highlight its practical applicability.