MVB fault diagnosis based on time-frequency analysis and convolutional neural networks.

Area of Science:

• Automotive Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Traditional multifunction vehicle bus (MVB) fault diagnosis relies on feature extraction and classification, demanding expert knowledge and often resulting in low accuracy.
• Existing methods struggle with the complexity and variability of MVB fault data.

Purpose of the Study:

• To develop an automated and accurate fault diagnosis method for MVB systems.
• To leverage deep learning for enhanced MVB fault detection capabilities.

Main Methods:

• Utilized short-time Fourier transform (STFT) to transform MVB vibration signals into time-frequency images.
• Developed a specific convolutional neural network (CNN) model, STCNN, for deep spatial feature learning on these images.
• Employed a Softmax classifier for final fault classification.

Main Results:

• The STCNN model achieved a remarkable fault detection accuracy of 99.68% on a diverse MVB network dataset.
• Demonstrated significantly superior performance compared to existing fault diagnosis methods.
• Validated the model's effectiveness under various operating conditions on a test bench.

Conclusions:

• The proposed STCNN method offers a highly accurate and efficient solution for MVB fault diagnosis.
• Deep learning, specifically CNNs, provides a powerful approach to overcome limitations of traditional methods.
• This technique holds significant potential for improving the reliability and safety of vehicle networks.