Power Equipment Fault Diagnosis Method Based on Energy Spectrogram and Deep Learning.

Area of Science:

• Industrial Manufacturing Intelligence
• Mechanical Engineering
• Artificial Intelligence

Background:

• Rotating machinery is crucial in industrial production, but its complex environment hinders effective fault detection.
• Existing deep learning models struggle with lightweight efficiency and extracting fault features from limited computing power.

Purpose of the Study:

• To propose an efficient and accurate fault detection method for power equipment, specifically rolling bearings.
• To address the limitations of current deep learning models in terms of computational efficiency and feature extraction.

Main Methods:

• Developed a two-dimensional time-frequency feature representation and energy spectrum feature map using wavelet packet transform for multi-resolution analysis.
• Proposed a lightweight residual dense convolutional neural network (LR-DenseNet) model, combining residual learning and dense connections.
• Introduced an LR-DenseSENet model incorporating transfer learning and an attention mechanism for enhanced feature fusion.

Main Results:

• The proposed energy spectrum feature map effectively extracts fault signal information and accelerates model convergence.
• The LR-DenseSENet model achieved a detection accuracy of up to 99.4%.
• The LR-DenseSENet model's parameter calculation was reduced to one-fifth of that of VGG, demonstrating significant computational efficiency.

Conclusions:

• The combined approach of energy spectrum feature maps and the LR-DenseSENet model provides a satisfactory fault detection effect for power equipment.
• This method offers a lightweight and highly accurate solution for fault diagnosis in industrial rotating machinery.