A Multitask CNN-Transformer Network for Aircraft Structural Fatigue and Load Prediction Using FBG Sensors

计算机科学 结构工程 结构健康监测 声学 电子工程 材料科学 人工神经网络 汽车工程 信号处理 工程类 加速度计 有限元法 数值模型 噪音(视频) 压力(语言学)
作者
Yiqun Zhang,Mingxin Yu,Yumin Zhang,Du Wang,Chaofan Deng,Lianqing Zhu
出处
期刊:IEEE Sensors Journal [IEEE Sensors Council]
卷期号:26 (16): 24910-24925
标识
DOI:10.1109/jsen.2026.3708876
摘要

Fatigue assessment and load prediction are two core tasks in aircraft structural health monitoring (SHM), and both are commonly constructed from strain measurements. In practice, however, fatigue-state evolution and load response are inherently coupled within the same strain sequence, whereas most existing studies treat them as two independent single-task problems. This separation limits the model’s ability to exploit its physical correlation and often leads to suboptimal performance in both classification and regression. To address this issue, we formulate fatigue assessment and load prediction as a strongly coupled dual-task learning problem driven by the same fiber Bragg grating (FBG) strain signal. We develop a coupling-aware multitask framework in which a shared 1-D convolutional neural network (CNN)-Transformer encoder learns common representations from strain sequences, while two task-specific heads perform fatigue classification and load regression. Furthermore, an epoch-level dynamic task-weight scheduling strategy is introduced to balance the optimization pace of the two objectives, thereby alleviating gradient conflicts and preventing task dominance during joint training. This design aligns the model organization with the fatigue-load interaction mechanism, where local perturbations are informative for fatigue-state discrimination and longer range temporal dependencies are critical for load reconstruction. The proposed framework is validated on a full-life-cycle dataset collected from a 7075-T651 aluminum alloy fatigue platform. Experimental results show that the method achieves a fatigue assessment accuracy of 95.47% and a recall of 94.04%, while reducing the average relative error (ARE) of load prediction to 1.15%, outperforming all baseline models. These results demonstrate that the proposed framework can effectively capture coupled fatigue-load characteristics and provide a promising solution for multi-indicator collaborative modeling in aircraft SHM.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
顾鹏飞发布了新的文献求助10
刚刚
蓝天的应助被何hehe采纳,获得10
1秒前
Eternity完成签到 ,获得积分10
1秒前
2秒前
2秒前
乐乐的应助被比巴卜采纳,获得10
2秒前
ZJK发布了新的文献求助10
2秒前
3秒前
4秒前
闫妍关注了科研通微信公众号
4秒前
水煮电吹风的应助被e00ec采纳,获得10
4秒前
高高冥幽完成签到,获得积分10
4秒前
4秒前
求求完成签到,获得积分10
4秒前
小郭发布了新的文献求助10
5秒前
VInci发布了新的文献求助10
5秒前
maT发布了新的文献求助10
5秒前
5秒前
lsq发布了新的文献求助30
6秒前
川页完成签到,获得积分20
6秒前
yang发布了新的文献求助10
6秒前
鸥羡发布了新的文献求助10
7秒前
7秒前
充电宝的应助被chiweiyoung采纳,获得10
8秒前
领导范儿的应助被11采纳,获得10
8秒前
VInci发布了新的文献求助10
8秒前
8秒前
VInci发布了新的文献求助10
8秒前
9秒前
芒果完成签到,获得积分10
9秒前
科研通AI2S的应助被退休小行星采纳,获得10
9秒前
蓝天的应助被哈基米大魔王采纳,获得10
9秒前
优秀傲松完成签到,获得积分10
9秒前
huanggaga发布了新的文献求助10
10秒前
深情安青的应助被暴躁的酸奶采纳,获得10
10秒前
YYY完成签到,获得积分10
10秒前
毛毛完成签到,获得积分10
11秒前
11秒前
11秒前
小明同学发布了新的文献求助10
12秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
What Else Can I Tell You?: A Pragmatic Study of English Rhetorical Questions as Discursive and Argumentative Acts 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8163922
求助须知:如何正确求助?哪些是违规求助? 9558007
关于积分的说明 21743214
捐赠科研通 7754605
什么是DOI,文献DOI怎么找? 3428426
关于科研通互助平台的介绍 2553053
邀请新用户注册赠送积分活动 2473048