This repo provides official code of Global Feature Enhancing and Fusion Framework for Strain Gauge Time Series Classification (GFEF, published in WWW 2025 Industry Track). We propose a multi-view global feature enhancement model that leverages hypergraph neural networks to fuse multi-view features, including image data and expert knowledge. This approach significantly enhances time-series representations and improves classification performance.
The overall architecture of GFEF is shown in following figure:
The proposed method is validated on an aircraft wing strain gauge anomaly detection task under static strength experiments in real industrial scenarios, where it demonstrates high accuracy and strong generalization capability across different experimental groups, as shown in the following figure.
Due to confidentiality reasons, we are currently unable to release the experimental strain gauge data related to national aircrafts.
Experts possess extensive business knowledge and experience. Based on this, they transform strain gauge time series into twelve statistical features (global features) to extract the characteristics of the strain gauge data, such as the degree of bending, thereby helping further represent the strain gauge series and benefiting strain gauge states recognition.
The code for constructing the twelve expert-designed global features for strain gauge time-series data is provided in Data_process/Expertise12GlobalFeatures.py.
If this repository and the work are helpful to you, please consider citing it:
@inproceedings{zhang2025global,
title={Global Feature Enhancing and Fusion Framework for Strain Gauge Status Recognition},
author={Zhang, Xu and Wang, Peng and Wang, Chen and Xu, Zhe and Nie, Xiaohua and Wang, Wei},
booktitle={Companion Proceedings of the ACM on Web Conference 2025},
pages={611--620},
year={2025}
}