Life Prediction of Rolling Bearing Based on Optimal Time-Frequency Spectrum and DenseNet-ALSTM

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Rolling bearing life prediction is challenged by noisy temporal signals, hindering fault feature extraction and reducing accuracy.
• Effective fault diagnosis and remaining useful life estimation are critical for industrial machinery maintenance.

Purpose of the Study:

• To develop a robust method for rolling bearing life prediction overcoming noise interference.
• To enhance fault feature extraction and improve prediction accuracy using optimal time-frequency representations.

Main Methods:

• Signal reconstruction using CEEMDAN (Complementary Ensemble Empirical Mode Decomposition) and Teager energy operator to denoise and enhance vibration signals.
• Optimization of Generalized S-transform (GST) time-frequency spectra using the Snake Optimizer (SO) to obtain optimal time-frequency spectra.
• Life prediction using the DenseNet-ALSTM network fed with the optimal time-frequency spectrum set.

Main Results:

• The proposed method demonstrates high prediction accuracy in rolling bearing life assessment.
• Comparison and ablation experiments validate the effectiveness and ideal performance of the developed approach.
• Signal enhancement and optimal time-frequency spectral analysis significantly contribute to improved prediction outcomes.

Conclusions:

• The integration of CEEMDAN, Teager energy operator, SO-optimized GST, and DenseNet-ALSTM offers a superior solution for rolling bearing life prediction.
• This method effectively addresses noise challenges in vibration signals, leading to more reliable machinery health monitoring.
• The study highlights the potential of advanced signal processing and deep learning for accurate predictive maintenance.