Enhancing Aerospace Fault Diagnosis With Conditioned Multiscale Generative Adversarial Networks

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Aerospace equipment failures present significant safety risks and economic losses.
• Traditional fault diagnosis requires large, labeled datasets, which are difficult to acquire for rare aerospace faults.
• Small-sample fault diagnosis is a critical challenge in the aerospace industry.

Purpose of the Study:

• To propose a novel conditioned multiscale generative adversarial networks (GANs) approach for enhanced fault diagnosis under small-sample conditions.
• To address the limitations of traditional fault diagnosis methods in aerospace applications.
• To improve the accuracy and efficiency of fault diagnosis with limited data.

Main Methods:

• Preprocessing raw vibration signals using short-time Fourier transform to extract time-frequency features.
• Training conditioned multiscale GANs with multiscale convolutional kernels on limited datasets to generate synthetic samples.
• Combining synthetic and original data to train a convolutional neural network for offline and online fault diagnosis.

Main Results:

• The proposed method significantly enhances fault diagnosis accuracy.
• The approach improves the efficiency of fault diagnosis, especially with limited training data.
• Validation on two aerospace datasets confirmed the method's effectiveness.

Conclusions:

• The conditioned multiscale GANs approach effectively overcomes the challenges of small-sample fault diagnosis in aerospace.
• The method provides a viable solution for improving safety and reducing economic losses through reliable fault diagnosis.
• This technique offers a promising direction for future research in intelligent fault diagnosis systems.