Avionics Module Fault Diagnosis Algorithm Based on Hybrid Attention Adaptive Multi-Scale Temporal Convolution...

Area of Science:

• Aerospace Engineering
• Computer Science
• Artificial Intelligence

Background:

• The reliability of avionics modules is critical for aircraft safety, necessitating effective fault diagnosis and health management (PHM).
• Existing deep learning PHM methods struggle with inefficient feature extraction, limited generalization, and a lack of specific avionics fault data.

Purpose of the Study:

• To address the limitations of current deep learning PHM methods for avionics modules.
• To develop a novel, accurate, and generalizable fault diagnosis method for avionics integrated functional circuits.

Main Methods:

• Fault injection was used to simulate diverse avionics module faults, followed by data enhancement to create a specialized dataset.
• A Hybrid Attention Adaptive Multi-scale Temporal Convolution Network (HAAMTCN) was proposed, featuring adaptive convolutional kernel sizing for efficient feature extraction.
• The HAAMTCN integrates Interaction Channel Attention (ICA) and Hierarchical Block Temporal Attention (HBTA) to focus on critical channel and temporal information.

Main Results:

• The constructed P2020 communications processor fault dataset enabled robust model training.
• The HAAMTCN demonstrated superior performance in extracting features from high information entropy avionics fault signals.
• The proposed HAAMTCN achieved a high accuracy of 99.64% in avionics module fault classification.

Conclusions:

• The HAAMTCN method significantly improves fault diagnosis accuracy and efficiency for avionics modules compared to existing approaches.
• The study successfully addresses the challenges of data scarcity and feature extraction limitations in deep learning-based avionics PHM.
• The developed method offers a promising solution for enhancing the safety and reliability of aircraft through advanced fault management.