Fault Diagnosis Method for Main Pump Motor Shielding Sleeve Based on Attention Mechanism and Multi-Source Data Fusion

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Condition Monitoring

Background:

• Shielding sleeves in main pump motors face complex operating environments leading to failures like bulging and cracking.
• Limited space hinders sensor installation for effective condition monitoring.
• Existing fault diagnosis methods struggle with multi-scale feature extraction and multi-source data fusion.

Purpose of the Study:

• To propose an advanced fault diagnosis method for shielding sleeve failures.
• To address limitations in current methods by integrating attention mechanisms and multi-source data fusion.
• To enable accurate diagnosis even when single data sources have weak fault characteristics.

Main Methods:

• Developed an attention-based multi-scale convolutional neural network (AM-MSCNN) for rich feature extraction.
• Proposed an attention-based convolutional neural network for multi-scale and multi-source data fusion (AM-MSMDF-CNN) to integrate torque, speed, voltage, and current data.
• Utilized BP algorithm and cross-entropy loss for final fault classification.

Main Results:

• The AM-MSMDF-CNN method demonstrated superior performance in fault diagnosis.
• Achieved 5-10% higher accuracy on simulation data and 10-15% higher accuracy on experimental data compared to 1D-CNN, Bagging, Random Forest, and SVM.
• Validated effectiveness through finite element simulation and small-scale prototype experiments.

Conclusions:

• The proposed AM-MSMDF-CNN method offers a robust and accurate solution for shielding sleeve fault diagnosis.
• The integration of attention mechanisms and multi-source data fusion is crucial for handling complex fault characteristics.
• This approach enhances equipment reliability in challenging operating conditions.