Research on bearing fault diagnosis based on a multimodal method

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Bearing fault diagnosis is critical for mechanical system safety.
• Vibration data from bearings often presents complex non-stationary and nonlinear characteristics.
• Existing methods struggle with the intricate nature of bearing vibration signals.

Purpose of the Study:

• To develop a novel model for enhanced bearing fault diagnosis.
• To address the challenges posed by non-stationary and nonlinear vibration data.
• To improve the accuracy and reliability of bearing condition monitoring.

Main Methods:

• A multi-scale time-frequency and statistical features fusion model (MTSF-FM) was developed.
• Continuous wavelet transform generated time-frequency images for feature extraction.
• Visual geometry group and convolutional neural networks were employed for deep feature extraction from images and time-frequency data.

Main Results:

• The MTSF-FM model achieved high diagnostic accuracies of 98.5% and 95.1% on two public datasets.
• The fusion of multi-scale time-frequency and statistical features proved effective.
• The model successfully captured intricate signal features for improved fault detection.

Conclusions:

• The MTSF-FM model offers a robust and effective approach for bearing fault diagnosis.
• This method demonstrates significant potential for real-world applications in predictive maintenance.
• The study highlights the value of combining diverse feature extraction techniques for complex signal analysis.