GAN-SAE based fault diagnosis method for electrically driven feed pumps.

Area of Science:

• Engineering
• Artificial Intelligence
• Power Systems

Background:

• High-speed electrically driven feed pumps are critical for power plant safety and economic benefits.
• Traditional fault diagnosis methods struggle with imbalanced datasets where fault data is rare.
• Deep learning approaches face challenges due to the scarcity of fault instances in operational data.

Purpose of the Study:

• To develop an intelligent condition monitoring and fault diagnosis method for electrically driven feed pumps.
• To address the challenge of imbalanced datasets in fault diagnosis.
• To improve the accuracy and reliability of fault detection in power plant feed pumps.

Main Methods:

• A novel Generative Adversarial Network-Stacked Auto Encoder (GAN-SAE) method is proposed.
• Generative Adversarial Network (GAN) is used for sample data compensation to address data imbalance.
• Stacked Auto Encoder (SAE) is employed for effective signal feature extraction.

Main Results:

• The GAN-SAE method demonstrated superior feature extraction capabilities compared to SAE, BP, and MNN.
• The proposed method significantly improved the accuracy of fault diagnosis for electrically driven feed pumps.
• Achieved a high fault diagnosis accuracy rate of 98.89%.

Conclusions:

• The GAN-SAE method effectively overcomes data imbalance issues in fault diagnosis.
• This approach enhances the capability of extracting internal fault features from time series data.
• The developed fault diagnosis program offers a reliable solution for power plant feed pump monitoring.