上海航天（中英文）2025，Vol.42Issue(6)：26-35,56,11.DOI:10.19328/j.cnki.2096-8655.2025.06.003
基于时频表示与ViT的航天器姿态系统智能故障诊断方法
Intelligent Fault Diagnosis Method for Spacecraft Attitude Systems Based on Time-frequency Representations and Vision Transformer
摘要
Abstract
To address the limitations of traditional deep learning methods in feature extraction and fault diagnosis accuracy in complex space missions,this paper proposes a fault diagnosis method based on time-frequency representations and vision transformer(TFViT).First,the raw signals are transformed into multi-channel time-frequency representations(TFRs)to comprehensively preserve the time-frequency features of the signals.Then,the input layer and feature extraction module of the TFViT model are optimized,further improving the model's capability to capture global dependencies within the TFRs,thereby enabling in-depth exploration of features from different time-frequency regions.With an experimental dataset constructed from a semi-physical simulation platform for spacecraft,systematic experiments are conducted to determine the optimal hyperparameter configuration of the TFViT model.The experimental results demonstrate that the TFViT model exhibits outstanding performance in fault diagnosis missions.The comparative analyses with several state-of-the-art deep learning methods fully validate the significant advantages of the proposed approach in diagnostic accuracy and robustness.关键词
故障诊断/航天器姿态系统/特征提取/时频表示(TFR)/Vision Transformer(ViT)Key words
fault diagnosis/spacecraft attitude system/feature extraction/time-frequency representation(TFR)/Vision Transformer(ViT)分类
航空航天引用本文复制引用
LI Ting,GAO Sheng,ZHANG Wei,ZHANG Rongpeng..基于时频表示与ViT的航天器姿态系统智能故障诊断方法[J].上海航天（中英文）,2025,42(6):26-35,56,11.基金项目
辽宁省自然科学基金面上资助项目(2024-MSBA-80) （2024-MSBA-80）
机器人与智能系统全国重点实验室开放基金资助课题(2025-Z01-05,2025-Z14) （2025-Z01-05,2025-Z14）
辽宁省兴辽英才资助项目(XLYC2402033) （XLYC2402033）
中国科学院沈阳自动化研究所基础研究计划资助项目(2022JC3K03,2023JC2G01) （2022JC3K03,2023JC2G01）
