Enhancing pumping unit diagnosis with similarity splicing data augmentation and wavelet denoising

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Dynamometer card-based fault diagnosis faces challenges with imbalanced data, noise, and limited generalization.
• Existing methods struggle to effectively handle these complex diagnostic scenarios.

Purpose of the Study:

• To develop a novel diagnostic framework addressing data imbalance and noise in dynamometer card analysis.
• To enhance the accuracy and robustness of fault diagnosis in industrial applications.

Main Methods:

• Implemented similarity-guided data augmentation for minority class expansion using kinematic feature recombination.
• Fused diverse operational parameters into standardized dynamometer card representations.
• Integrated Discrete Wavelet Transform (DWT) layers into MobileNet-V2 for noise suppression during feature downsampling.

Main Results:

• The proposed framework demonstrated superior accuracy and noise robustness compared to conventional methods.
• Achieved significant performance improvements while maintaining computational efficiency.
• Validated the framework's effectiveness in real-world oilfield monitoring applications.

Conclusions:

• The novel framework effectively resolves data imbalance and noise contamination in dynamometer card analysis.
• The wavelet-optimized lightweight neural network approach offers a practical and efficient solution for industrial fault diagnosis.
• This research establishes a deployable system for precise and robust fault diagnosis in oilfield operations.