硅酸盐学报

基于机器学习-大语言模型协同的卫星通信用抗辐射保偏光纤应力区多目标智能设计

PDF下载

杨坤 ¹^,² , 王静 ¹ , 张翼菲 ² , 李羽 ²

硅酸盐学报 | 第二十八届中国科协年会智能建造关键无机非金属材料–先进玻璃材料智能设计与产业前沿专题-研究论文 2026,54(10): 3604-3613

收起

硅酸盐学报 |第二十八届中国科协年会智能建造关键无机非金属材料–先进玻璃材料智能设计与产业前沿专题-研究论文 2026 , 54 (10) : 3604 -3613

基于机器学习-大语言模型协同的卫星通信用抗辐射保偏光纤应力区多目标智能设计

全屏

杨坤¹^,², 王静¹, 张翼菲², 李羽²

作者信息

^1.武汉理工大学材料科学与工程学院，武汉　430070

^2.长飞光纤光缆股份有限公司，武汉　430070

通讯作者:

王静（1988-），女，博士，研究员。

作者简介:

杨坤（1987-），男，博士研究生。

Intelligent Multi-Objective Design of Stress-Applying Part in Polarization-Maintaining Fibers via Machine Learning and Large Language Model Collaboration

Kun YANG¹^,², Jing WANG¹, Yifei ZHANG², Yu LI²

Affiliations

^1.School of Materials Science and Engineering, Wuhan University of Technology, Wuhan 430070, China

^2.Yangtze Optical Fibre and Cable Joint Stock Limited Company, Wuhan 430070, China

Corresponding Author:

WANG Jing (1988-), female, Ph.D., Professor. E-mail: wangj@whut.edu.cn

YANG Kun (1987-), male, Doctoral candidate. E-mail: yangkun_02543@yofc.com

出版时间: 2026-07-30 doi: 10.14062/j.issn.0454-5648.20260279

文章导航

摘要

收起

为实现卫星通信极端环境下光纤兼具高抗辐射性能与高保偏性能，针对PANDA型保偏光纤应力区设计中参数耦合强、性能指标多且传统试错效率低的问题，提出一种机器学习（ML）-大语言模型（LLM）协同的智能设计方法。基于500组系统实验数据，以应力区直径、B₂O₃摩尔分数和应力区边缘-纤芯边缘距离为输入变量，分别建立衰耗、辐致衰耗、拉断力、nd值、消光比、全温消光比变化、拍长和研磨开裂概率8项性能指标的预测模型，并对随机森林（RF）、XGBoost和梯度提升回归（GBR）进行五折交叉验证比较。结果表明：最优模型的决定系数（R²）为0.76~0.99，平均R²为0.906；SHAP分析表明，边缘距离主导衰耗类指标，应力区直径主导力学与偏振性能，而B₂O₃摩尔分数对研磨开裂概率影响最为显著。在此基础上，引入LLM实现自然语言需求解析、约束冲突诊断及最小松弛建议。7组典型设计场景的测试结果表明，与随机试错相比，该方法可将期望实验次数平均降低91%，在四约束场景下最高降低99.8%。当设计需求不可同时满足时，系统能够定位主要冲突约束并给出定量松弛方案。该方法可为保偏光纤应力区的高效逆向设计提供数据驱动的新思路。

关键词

保偏光纤 / 应力区 / 机器学习 / 大语言模型 / 逆向设计

Abstract

收起

Introduction

The design of the stress-applying part (SAP) in polarization-maintaining fibers (PMF) is a typical multi-variable, multi-objective optimization problem. The coupled effects of SAP diameter, B₂O₃ molar fraction and SAP edge-to-core distance determine attenuation, polarization-maintaining capability, mechanical reliability, and radiation resistance. Conventional trial-and-error design is therefore inefficient, especially when multiple performance constraints must be satisfied simultaneously.

Methods

Based on 500 experimental datasets, predictive models were established for eight performance indicators, including attenuation, post-irradiation loss, tensile strength, nd value, extinction ratio, temperature-induced extinction-ratio variation, beat length, and grinding crack probability. Three ensemble learning algorithms, i.e., random forest (RF), XGBoost, and gradient boosting regressor (GBR), were systematically compared via 5-fold cross-validation. SHAP analysis was employed to reveal the dominant effects of design variables. Furthermore, a large language model (LLM) was introduced as a natural-language interface and explanation layer to parse user requirements, diagnose infeasible constraints and provide minimum-relaxation suggestions.

Results and discussion

The optimal models achieved R² values ranging from 0.76 to 0.99 with an average of 0.906, indicating that the nonlinear mapping between SAP parameters and performance metrics can be reliably captured. SHAP analysis revealed that the edge distance dominated attenuation-related indices, the SAP diameter governed mechanical and polarization-related properties, and the B₂O₃ molar fraction exerted the strongest effect on grinding crack probability. For 30 natural-language test cases, the LLM achieved 100% accuracy in indicator recognition and inequality-direction determination, and 85.0% exact-value accuracy. Across seven representative design scenarios, the proposed framework reduced the expected number of trial experiments by 91% on average and by up to 99.8% in the four-constraint scenario. When no feasible solution existed, the system could identify the bottleneck constraint and quantify the minimum relaxation required to recover feasibility.

Conclusions

A collaborative ML-LLM framework was developed for the inverse design of PMF SAPs. The method integrates quantitative prediction, mechanism-oriented interpretation and natural-language-driven interaction, thereby converting conventional trial-and-error design into an efficient data-driven screening process. This approach provides a practical pathway for intelligent design of polarization-maintaining fibers and may also be extended to other inorganic non-metallic materials with strongly coupled process-property relationships.

Key words

polarization-maintaining fiber / stress-applying part / machine learning / large language model / inverse design

引用本文

杨坤, 王静, 张翼菲, 李羽. 基于机器学习-大语言模型协同的卫星通信用抗辐射保偏光纤应力区多目标智能设计. 硅酸盐学报, 2026 , 54 (10) : 3604 -3613 . DOI: 10.14062/j.issn.0454-5648.20260279

Kun YANG, Jing WANG, Yifei ZHANG, Yu LI. Intelligent Multi-Objective Design of Stress-Applying Part in Polarization-Maintaining Fibers via Machine Learning and Large Language Model Collaboration[J]. Journal of the Chinese Ceramic Society, 2026 , 54 (10) : 3604 -3613 . DOI: 10.14062/j.issn.0454-5648.20260279

基金

收起

湖北省自然科学基金杰出青年项目(2023AFA067)

参考文献

收起

文献

收起

参考文献引证文献

排序方式：

[1]

NODA J, OKAMOTO K, SASAKI Y. Polarization-maintaining fibers and their applications[J]. J Light Technol, 1986, 4(8): 1071-1089.

[2]

RASHLEIGH S. Origins and control of polarization effects in single-mode fibers[J]. J Light Technol, 1983, 1(2): 312-331.

[3]

KERSEY A D, DANDRIDGE A. Applications of fiber-optic sensors[J]. IEEE Trans Compon Hybrids Manuf Technol, 1990, 13(1): 137-143.

[4]

KASHAYKIN P F, POSPELOVA E A, SHARONOVA Y O, et al. Temperature and polarization dependence of radiation-induced attenuation in pure-silica-core PANDA optical fiber[J]. Opt Mater, 2022, 131: 112510.

[5]

TOMASHUK A L, KASHAYKIN P F, SEMJONOV S L, et al. Comparison study of radiation-resistant polarization-maintaining PANDA fibers with undoped- and N-doped-silica core[J]. J Light Technol, 2020, 38(20): 5817-5824.

[6]

ERONYAN M A, DEVETYAROV D R, REUTSKIY A A, et al. Polarization-maintaining and radiation resistant optical fiber with germanosilicate elliptical core[J]. Opt Fiber Technol, 2022, 68: 102789.

[7]

ROVERA A, TANCAU A, BOETTI N, et al. Fiber optic sensors for harsh and high radiation environments in aerospace applications[J]. Sensors, 2023, 23(5): 2512.

[8]

CAO L D, ABEDIN S, CUI G Q, et al. Artificial intelligence and machine learning in optical fiber sensors: A review[J]. Sensors, 2025, 25(24): 7442.

[9]

BUTLER K T, DAVIES D W, CARTWRIGHT H, et al. Machine learning for molecular and materials science[J]. Nature, 2018, 559(7715): 547-555.

[10]

WEI J, CHU X, SUN X Y, et al. Machine learning in materials science[J]. InfoMat, 2019, 1(3): 338-358.

[11]

SCHMIDT J, MARQUES M R G, BOTTI S, et al. Recent advances and applications of machine learning in solid-state materials science[J]. npj Comput Mater, 2019, 5: 83.

[12]

ASADIAN M, SEYEDEIN S H, ABOUTALEBI M R, et al. Optimization of the parameters affecting the shape and position of crystal-melt interface in YAG single crystal growth[J]. J Cryst Growth, 2009, 311(2): 342-348.

[13]

KUTSUKAKE K, NAGAI Y T, BANBA H. Virtual experiments of Czochralski growth of silicon using machine learning: Influence of processing parameters on interstitial oxygen concentration[J]. J Cryst Growth, 2022, 584: 126580.

[14]

杨明亮, 王瑞仙, 孙贵花, 等. 机器学习在晶体生长中的应用研究进展[J]. 硅酸盐学报, 2024, 52(7): 2412-2424.

YANG Mingliang, WANG Ruixian, SUN Guihua, et al. J Chin Ceram Soc, 2024, 52(7): 2412-2424.

[15]

CHEN C, ONG S P. A universal graph deep learning interatomic potential for the periodic table[J]. Nat Comput Sci, 2022, 2(11): 718-728.

[16]

MERCHANT A, BATZNER S, SCHOENHOLZ S S, et al. Scaling deep learning for materials discovery[J]. Nature, 2023, 624(7990): 80-85.

[17]

VENKETESWARAN A, LALAM N, WUENSCHELL J, et al. Recent advances in machine learning for fiber optic sensor applications[J]. Adv Intell Syst, 2022, 4: 2100067.

[18]

LI A, WANG T H. Research on radiation damage characteristics of optical fiber materials based on data mining and machine learning[M]//Advanced Hybrid Information Processing. ChamSpringer International Publishing, 2019: 410-418.

[19]

JABLONKA K M, SCHWALLER P, ORTEGA-GUERRERO A, et al. Leveraging large language models for predictive chemistry[J]. Nat Mach Intell, 2024, 6(2): 161-169.

[20]

BOIKO D A, MACKNIGHT R, KLINE B, et al. Autonomous chemical research with large language models[J]. Nature, 2023, 624(7992): 570-578.

[21]

ZHENG Z L, ZHANG O F, BORGS C, et al. ChatGPT chemistry assistant for text mining and the prediction of MOF synthesis[J]. J Am Chem Soc, 2023, 145(32): 18048-18062.

[22]

MIRET S, ANOOP KRISHNAN N M. Are LLMs ready for real-world materials discovery?[EB/OL]. 2024: arXiv: 2402.05200. https://arxiv.org/abs/2402.05200

[23]

BREIMAN L. Random forests[J]. Mach Learn, 2001, 45(1): 5-32.

[24]

LUNDBERG S M, LEE S I. A unified approach to interpreting model predictions[EB/OL]. 2017: arXiv: 1705.07874. https://doi.org/10.48550/arXiv.1705.07874.

[25]

WANG Z B, SUN P, HU Y D, et al. Crack morphology tailoring and permeability prediction of polyvinyl alcohol-steel hybrid fiber engineered cementitious composites[J]. J Clean Prod, 2023, 383: 135335.

2026年第54卷第10期

PDF下载

引用本文

BibTeX

文章信息

doi: 10.14062/j.issn.0454-5648.20260279

接收时间：2026-04-28
首发时间：2026-10-22
出版时间：2026-07-30

补充材料

相关文章

文章信息

作者

出版历史

收稿日期：2026-04-28
修回日期：2026-05-26

基金

湖北省自然科学基金杰出青年项目(2023AFA067)

作者信息

^1.武汉理工大学材料科学与工程学院，武汉　430070

^2.长飞光纤光缆股份有限公司，武汉　430070

通讯作者:

王静（1988-），女，博士，研究员。

参考文献

分享链接

https://castjournals.cast.org.cn/joweb/gsyxb/CN/10.14062/j.issn.0454-5648.20260279

分享至

全文二维码

扫描看全文

引用本文

BibTeX

本文的引用情况

2种不同金属材料的力学参数

科 Family	属数 Number of genus	种数 Number of species	占总种数比例 Percentage of total species (%)	属 Genus	种数 Number of species	占总种数比例 Percentage of total species (%)
鹅膏菌科Amanitaceae	2	11	5.26	鹅膏菌属 Amanita	10	4.78
小菇科 Mycenaceae	2	12	5.74	丝盖伞属 Inocybe	5	2.39
多孔菌科 Polyporaceae	8	14	6.70	蜡蘑属 Laccaria	5	2.39
红菇科 Russulaceae	3	23	11.00	小皮伞属 Marasmius	6	2.87
				小菇属 Mycena	11	5.26
				光柄菇属 Pluteus	5	2.39
				红菇属 Russula	17	8.13
				栓菌属 Trametes	5	2.39

关闭全屏

BibTeX
EndNote
RefWorks
TxT