硅酸盐学报

基于物理信息机器学习的玻璃杨氏模量预测模型构建与机理分析

PDF下载

钱兆青 ¹ , 曹毅 ² , 康泽玉 ¹ , 卢亚东 ¹ , 高文凯 ¹ , 康俊峰 ¹^,³

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): 3593-3603

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) : 3593 -3603

基于物理信息机器学习的玻璃杨氏模量预测模型构建与机理分析

全屏

钱兆青¹, 曹毅², 康泽玉¹, 卢亚东¹, 高文凯¹, 康俊峰¹^,³

作者信息

^1.济南大学材料科学与工程学院，济南　250022

^2.济南大学信息科学与工程学院，济南　250022

^3.山东省高性能玻璃纤维重点实验室，山东　泰安　271000

通讯作者:

康俊峰（1988-），男，博士，副教授。

作者简介:

钱兆青（2001-），男，硕士研究生。

Construction and Mechanism Analysis of Glass Young's Modulus Prediction Model Using Physics-Informed Machine Learning

Zhaoqing QIAN¹, Yi CAO², Zeyu KANG¹, Yadong LU¹, Wenkai GAO¹, Junfeng KANG¹^,³

Affiliations

^1.School of Materials Science and Engineering, University of Jinan, Jinan 250022, China

^2.School of Information Science and Engineering, University of Jinan, Jinan 250022, China

^3.Shandong Key Laboratory of High Performance Glass Fiber, Taian 271000, Shandong, China

Corresponding Author:

KANG Junfeng (1988-), male, Ph.D., Associate Professor. E-mail: mse_kangjf@ujn.edu.cn

QIAN Zhaoqing (2001-), male, Master candidate. E-mail: 864897665@qq.com

出版时间: 2026-07-29 doi: 10.14062/j.issn.0454-5648.20260315

文章导航

摘要

收起

针对玻璃杨氏模量公开数据库数据一致性差、传统清洗难以剔除异常样本、纯数据驱动模型泛化与外推能力受限等问题，提出一种融合物理先验与机器学习的玻璃杨氏模量预测方法。将Makishima-Mackenzie（MM）半经验模型嵌入梯度提升机（GBM）建模流程，依据实测值大于MM计算值的物理规律完成数据清洗，构建受物理约束的高质量数据集；以MM模型计算残差为预测目标，通过残差修正与MM计算值加和实现杨氏模量精准预测。结合SHAP可解释性分析与物理化学信息描述符建模，揭示MM模型系统误差的产生机制。结果表明，所构建的物理信息机器学习（PIML）模型测试集决定系数R²达0.962，中位绝对误差MedAE为0.77，均方误差MSE为7.04，预测精度、鲁棒性与泛化能力显著优于纯数据驱动模型；外推测试验证模型在训练集域外样本仍保持稳定低误差。该方法为高模量玻璃配方快速开发提供了兼具高精度、可解释性与强外推能力的性能预测新途径。

关键词

物理信息机器学习 / Makishima-Mackenzie模型 / 高模量玻璃 / SHAP分析

Abstract

收起

Introduction

As a critical reinforcing material, high-modulus glass has experienced a rapidly growing demand in aerospace, wind power and other high-tech industries. This trend makes the development of novel high-performance, high-modulus glass formulations an urgent and significant task for materials science.

Machine learning has emerged as a transformative new research paradigm following traditional trial-and-error experiments and computer simulations. It offers remarkable advantages in efficiency, accuracy, and low energy consumption, and has been widely used to predict various glass properties.

However, purely data-driven machine learning models face severe performance bottlenecks when predicting glass Young's modulus (YM), showing significantly lower accuracy than models for other glass properties. This limitation mainly stems from two factors: poor data consistency in public databases due to the sensitivity of YM to preparation-related thermal history and testing conditions, and the lack of effective strategies to eliminate abnormal samples, which leads to model overfitting and limited generalization and extrapolation capabilities.

Liu and co-workers showed that pure data-driven models face three major contradictions in materials science, and proposed using data governance and domain knowledge embedding to improve model performance. Data quality is decisive for model performance, and combining anomaly detection with a divide-and-conquer strategy can yield accurate and interpretable models.

The Makishima-Mackenzie (MM) model is a classic semi-empirical model for predicting the YM of oxide glass with clear physical significance, but it exhibits a well-documented systematic underestimation of actual YM values. Residual modeling has been proven to be an effective approach to address the deficiencies of physical models. Therefore, this study aims to develop a physics-informed machine learning (PIML) model by integrating the MM model with machine learning to overcome the limitations of pure data-driven models.

Methods

The original dataset was constructed from the SciGlass database, which contained 2861 samples, with 15 core oxides as features and YM and density as target variables. Basic data cleaning was first performed to remove samples with incomplete compositions, missing values, zero values, negative values, and duplicate formulations.

Five pure data-driven models, namely Gradient Boosting Machine (GBM), Random Forest (RF), eXtreme Gradient Boosting (XGB), Artificial Neural Network (ANN) and Support Vector Regression (SVR), were trained on the original dataset. All models were optimized using Bayesian optimization and five-fold cross-validation, with the dataset split into training and test sets in an 8:2 ratio.

Subsequently, the MM model was used to calculate the theoretical YM values for all the samples. Based on the physical principle that the MM model systematically underestimates glass YM, samples with negative residuals (measured YM minus MM-calculated YM) were eliminated. This procedure yielded a high-quality dataset of 2463 samples constrained by physical principles.

The GBM model, which exhibited the best performance among the three pure data-driven models, was selected for constructing the PIML model. Instead of directly predicting YM, the PIML model used oxide compositions as input features and the MM model residuals as the prediction target, and was trained on the high-quality dataset. The final YM prediction was obtained by adding the predicted residual to the MM-calculated value.

Model performance was evaluated using three comprehensive metrics: the coefficient of determination (R²), mean squared error (MSE), and median absolute error (MedAE). SHapley Additive exPlanations (SHAP) analysis and physicochemical-descriptor modeling were employed to reveal the error mechanism of the MM model. Extrapolation capability was tested using 29 samples with Li₂O, Y₂O₃, La₂O₃ and ZrO₂ contents outside the training set range.

Results and discussion

The pure data-driven GBM model achieved a test-set R² of 0.910 on the original dataset, representing the best performance among the three models. However, its performance was still substantially limited by the abnormal samples present in the dataset.

In contrast, the developed GBM-PIML model trained on the physically constrained high-quality dataset achieved a markedly improved test set R² of 0.962. Compared with the pure GBM model, its MedAE was reduced by approximately 40% and its MSE was reduced by approximately 70%. These results demonstrated that the physical-constraint-based data cleaning effectively removed misleading abnormal samples, and the residual learning strategy reduced the learning difficulty of the model by allowing the model to learn the systematic error pattern of the MM model rather than the entire complex composition-YM relation.

SHAP analysis revealed that increases in Y₂O₃, La₂O₃, MgO, and Li₂O contents significantly increased the MM model residuals. This result was attributed to the fact that the MM model treated glass as a homogeneous system and ignored two key effects: the accumulation of non-bridging oxygen by high-field-strength modifiers and the interaction between rare earth metals and aluminum ions.

A physicochemical descriptor model was further constructed using four descriptors (Ma, ACFS, Re/Al, IMO/Si) that characterize internal glass structure and oxide interactions and achieved a test set R² of 0.932. SHAP analysis of this model confirmed that a lower degree of network polymerization, higher average cation field strength, higher Re/Al ratio and higher IMO/Si ratio all contributed to larger MM model residuals, which was consistent with the underlying physical mechanism.

Extrapolation tests showed that the GBM-PIML model exhibited significantly lower prediction errors than the pure GBM model in the high-modulus region. Most importantly, its error did not increase sharply as YM increased, indicating strong extrapolation stability.

Conclusions

This study constructs a residual-learning-based PIML model by integrating the MM physics-based model with the GBM machine-learning algorithm. The model achieves a test set R² of 0.962, a MedAE of 0.77, and an MSE of 7.04, and demonstrates superior prediction accuracy, robustness and generalization capability compared to conventional purely data-driven models trained on the same original dataset.

The combination of SHAP analysis and physicochemical descriptor modeling provides deep insight into the error mechanism of the MM model. The results reveal that the systematic error of the MM model mainly originates from its neglect of the effects of high-field-strength ions and oxide interactions on glass YM.

Extrapolation tests confirm that the incorporation of physical prior knowledge significantly enhances the model's extrapolation ability, especially in the high-modulus region. The proposed PIML model provides a new paradigm for the rapid development of high-modulus glass formulations with high prediction accuracy, interpretability, and strong extrapolation capability.

Key words

physics-informed machine learning / Makishima-Mackenzie model / high-modulus glass / SHAP analysis

引用本文

钱兆青, 曹毅, 康泽玉, 卢亚东, 高文凯, 康俊峰. 基于物理信息机器学习的玻璃杨氏模量预测模型构建与机理分析. 硅酸盐学报, 2026 , 54 (10) : 3593 -3603 . DOI: 10.14062/j.issn.0454-5648.20260315

Zhaoqing QIAN, Yi CAO, Zeyu KANG, Yadong LU, Wenkai GAO, Junfeng KANG. Construction and Mechanism Analysis of Glass Young's Modulus Prediction Model Using Physics-Informed Machine Learning[J]. Journal of the Chinese Ceramic Society, 2026 , 54 (10) : 3593 -3603 . DOI: 10.14062/j.issn.0454-5648.20260315

基金

收起

山东省自然科学基金面上项目(ZR2025MS774)
国家自然科学基金面上项目(52172019)
泰安市岱岳区科技创新项目(CXXM2025002)

参考文献

收起

文献

收起

参考文献引证文献

排序方式：

[1]

LIU G, ZHANG X F, CHEN X L, et al. Additive manufacturing of structural materials[J]. Mater Sci Eng R Rep, 2021, 145: 100596.

[2]

WAN J T, GAN B, LI C, et al. A sustainable, eugenol-derived epoxy resin with high biobased content, modulus, hardness and low flammability: Synthesis, curing kinetics and structure-property relationship[J]. Chem Eng J, 2016, 284: 1080-1093.

[3]

KANG Z Y, WANG L J, LI X Y, et al. Interpretable machine learning accelerates development of high-specific modulus glass[J]. Comput Mater Sci, 2025, 246: 113482.

[4]

BISHNOI S, SINGH S, RAVINDER R, et al. Predicting Young's modulus of oxide glasses with sparse datasets using machine learning[J]. J Non Cryst Solids, 2019, 524: 119643.

[5]

HU Y J, ZHAO G, ZHANG M F, et al. Predicting densities and elastic moduli of SiO₂-based glasses by machine learning[J]. npj Comput Mater, 2020, 6: 25.

[6]

QIN J C, ZHANG F Q, MA M S, et al. Multiobjective optimization of dielectric, thermal, and mechanical properties of inorganic glasses utilizing explainable machine learning and genetic algorithm[J]. Mater Genome Eng Adv, 2025, 3(2): e70005.

[7]

LIU Y, YANG Z W, ZOU X X, et al. A general framework to govern machine learning oriented materials data quality[J]. Mater Sci Eng R Rep, 2025, 166: 101050.

[8]

LIU Y, GUO B R, ZOU X X, et al. Machine learning assisted materials design and discovery for rechargeable batteries[J]. Energy Storage Mater, 2020, 31: 434-450.

[9]

LIU Y, WU J M, AVDEEV M, et al. Multi-layer feature selection incorporating weighted score-based expert knowledge toward modeling materials with targeted properties[J]. Adv Theory Simul, 2020, 3(2): 1900215.

[10]

LIU Y, ZOU X X, MA S C, et al. Feature selection method reducing correlations among features by embedding domain knowledge[J]. Acta Mater, 2022, 238: 118195.

[11]

刘悦, 邹欣欣, 杨正伟, 等. 材料领域知识嵌入的机器学习[J]. 硅酸盐学报, 2022, 50(3): 863-876.

LIU Yue, ZOU Xinxin, YANG Zhengwei, et al. J Chin Ceram Soc, 2022, 50(3): 863-876.

[12]

刘悦, 马舒畅, 杨正伟, 等. 面向材料领域机器学习的数据质量治理[J]. 硅酸盐学报, 2023, 51(2): 427-437.

LIU Yue, MA Shuchang, YANG Zhengwei, et al. J Chin Ceram Soc, 2023, 51(2): 427-437.

[13]

LIU Y, YANG Z W, ZOU X X, et al. Data quantity governance for machine learning in materials science[J]. Natl Sci Rev, 2023, 10(7): 229-245.

[14]

LIU Y, MA S C, YANG Z W, et al. Domain knowledge-assisted materials data anomaly detection towards constructing high-performance machine learning models[J]. J Materiomics, 2025, 11(6): 101066.

[15]

SHI S Q, SUN S Y, MA S C, et al. Detection method on data accuracy incorporating materials domain knowledge[J]. J Inorg Mater, 2022, 37(12): 1311.

[16]

杨正伟, 刘子图, 张王颖, 等. 材料领域知识嵌入的分而治之机器学习[J]. 硅酸盐学报, 2026, 54(1): 14-24.

YANG Zhengwei, LIU Zitu, ZHANG Wangying, et al. J Chin Ceram Soc, 2026, 54(1): 14-24.

[17]

LIU Y, WANG S Y, YANG Z W, et al. Auto-MatRegressor: Liberating machine learning alchemists[J]. Sci Bull, 2023, 68(12): 1259-1270.

[18]

LIU Y, WU L H, YANG Z W, et al. Descriptors divide-and-conquer enables multifaceted and interpretable materials structure-activity relationship analysis[J]. Adv Funct Mater, 2025, 35(26): 2421621.

[19]

LIU Y, WU J M, WANG Z C, et al. Predicting creep rupture life of Ni-based single crystal superalloys using divide-and-conquer approach based machine learning[J]. Acta Mater, 2020, 195: 454-467.

[20]

LIU Y, WU J M, YANG G, et al. Predicting the onset temperature (T_g) of Ge_xSe_1-_x glass transition: A feature selection based two-stage support vector regression method[J]. Sci Bull, 2019, 64(16): 1195-1203.

[21]

LIU Y, ZHAO T L, YANG G, et al. The onset temperature (T_g) of As_xSe_1-_x glasses transition prediction: A comparison of topological and regression analysis methods[J]. Comput Mater Sci, 2017, 140: 315-321.

[22]

WANG X M, KAN Q H, PETRU M, et al. Study on the composition-property relationships of basalt fibers based on symbolic regression and physics-informed neural network[J]. Compos Part A Appl Sci Manuf, 2024, 185: 108324.

[23]

MAKISHIMA A, MACKENZIE J D. Direct calculation of Young's moidulus of glass[J]. J Non Cryst Solids, 1973, 12(1): 35-45.

[24]

SHI Y, TANDIA A, DENG B H, et al. Revisiting the Makishima-Mackenzie model for predicting the Young's modulus of oxide glasses[J]. Acta Mater, 2020, 195: 252-262.

[25]

WILLARD J, JIA X W, XU S M, et al. Integrating scientific knowledge with machine learning for engineering and environmental systems[J]. ACM Comput Surv, 2023, 55(4): 1-37.

[26]

CASSAR D R. GlassNet: A multitask deep neural network for predicting many glass properties[J]. Ceram Int, 2023, 49(22): 36013-36024.

[27]

PEDREGOSA F, VAROQUAUX G, GRAMFORT A, et al. Scikit-learn: Machine Learning in Python[J]. J Mach Learn Res, 2011, 12: 2825-2830.

[28]

CHEN T Q, GUESTRIN C. XGBoost: A scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco California USA. ACM, 2016: 785-794.

[29]

BERGSTRA J, YAMINS D, COX D D. Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures[C]//International Conference on Machine Learning., 2013

[30]

CHICCO D, WARRENS M J, JURMAN G. The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation[J]. Peerj Comput Sci, 2021, 7: e623.

[31]

JĒKABSONS G, LAVENDELS J, SITIKOVS V. Model evaluation and selection in multiple nonlinear regression analysis[J]. Math Model Anal, 2007, 12(1): 81-90.

[32]

BHATTOO R, BISHNOI S, ZAKI M, et al. Understanding the compositional control on electrical, mechanical, optical, and physical properties of inorganic glasses with interpretable machine learning[J]. Acta Mater, 2023, 242: 118439.

[33]

LUNDBERG S M, ERION G, CHEN H, et al. From local explanations to global understanding with explainable AI for trees[J]. Nat Mach Intell, 2020, 2(1): 56-67.

[34]

FLORIAN P, SADIKI N, MASSIOT D, et al. ²⁷Al NMR study of the structure of lanthanum- and yttrium-based aluminosilicate glasses and melts[J]. J Phys Chem B, 2007, 111(33): 9747-9757.

[35]

SEN S, STEBBINS J F, KROEKER S, et al. Evidence for mixed Mg coordination environments in silicate glasses: Results from ²⁵Mg NMR spectroscopy at 35.2 T[J]. J Phys Chem B, 2023, 127(49): 10659-10666.

[36]

GANISETTI S, GADDAM A, KUMAR R, et al. Elucidating the formation of Al-NBO bonds, Al-O-Al linkages and clusters in alkaline-earth aluminosilicate glasses based on molecular dynamics simulations[J]. Phys Chem Chem Phys, 2019, 21(43): 23966-23977.

[37]

LEIGHT C J, YTSMA C, MCCANTA M C, et al. Compositional characterization of glassy volcanic material from VSWIR and MIR spectra using partial least squares regression models[J]. Earth Space Sci, 2024, 11(6): e2023EA003439.

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.0454-5648.20260315

接收时间：2026-05-13
首发时间：2026-10-22
出版时间：2026-07-29

补充材料

相关文章

文章信息

作者

出版历史

收稿日期：2026-05-13
修回日期：2026-06-08

基金

山东省自然科学基金面上项目(ZR2025MS774)

国家自然科学基金面上项目(52172019)

泰安市岱岳区科技创新项目(CXXM2025002)

作者信息

^1.济南大学材料科学与工程学院，济南　250022

^2.济南大学信息科学与工程学院，济南　250022

^3.山东省高性能玻璃纤维重点实验室，山东　泰安　271000

通讯作者:

康俊峰（1988-），男，博士，副教授。

参考文献

分享链接

https://castjournals.cast.org.cn/joweb/gsyxb/CN/10.14062/j.issn.0454-5648.20260315

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