Automated Design of Fault Diagnosis CNN Network for Satellite Attitude Control Systems

Area of Science:

• Spacecraft Engineering
• Artificial Intelligence
• Fault Diagnosis

Background:

• Unsupervised and self-supervised methods dominate satellite fault diagnosis.
• Supervised methods offer advantages in sensitivity and deployment for subsystems like ACSs.
• Existing CNN designs face challenges in accuracy and over-design for satellite ACS fault diagnosis.

Purpose of the Study:

• To introduce a novel CNN architecture for satellite attitude control system (ACS) fault diagnosis.
• To address over-design and accuracy issues in current CNN-based fault diagnosis methods.
• To develop an optimized CNN model for high-sensitivity and lightweight fault detection.

Main Methods:

• Developed the Advanced Convolution Blocks-based CNN (ACBCNN) with efficient design elements and dense connections.
• Introduced the Modified Particle Swarm Optimization-ACBCNN (MPSO-ACBCNN) algorithm for optimizing the ACBCNN architecture.
• Implemented fitness design balancing parameter quantity and training effectiveness, and methods for feasible solutions within MPSO.

Main Results:

• The ACBCNN architecture enhances feature extraction from power spectral density (PSD) graphs.
• The MPSO-ACBCNN algorithm effectively optimizes the CNN for ACS fault diagnosis.
• Numerical experiments confirmed the effectiveness and superiority of the MPSO-ACBCNN method.

Conclusions:

• The MPSO-ACBCNN method provides a superior approach for satellite ACS fault diagnosis.
• This method achieves high-sensitivity detection and lightweight deployment for critical subsystems.
• The proposed approach overcomes limitations of existing CNN designs in satellite fault diagnosis.