Characterization and Prediction of Epoxy Resin Curing Behavior Using a Data-Driven Kinetic Model Based on Dual Kriging
- 1School of Materials Science and Engineering, Wuhan University of Technology, Wuhan 430070, China
- 2Xi’an Aerospace Composites Research Institute, Xi’an 710025, China
- 3State Key Laboratory of Advanced Technology for Material Synthesis and Processing, Wuhan University of Technology, Wuhan 430070, China
Abstract
A data-driven kinetic framework based on Dual Kriging is developed to represent epoxy curing behavior and supply a local cure-rate function for finite element (FE) analysis. Error-controlled compression reduces 7200 isothermal observations to 187 points, with maximum and mean normalized reconstruction errors of 0.45% and 0.069%, respectively. The compressed interpolant reproduces the E-51 conversion histories measured by differential scanning calorimetry (DSC), including their late-stage plateaus. A separable 245-node rate representation is implemented in a sequential thermo-chemo-mechanical analysis, combining E-51 experimental data at 55–70 °C with auxiliary Kamal–Sourour data at 30–50 °C. The two kinetics models produce similar temperature and strain histories during staged curing; both underestimate the measured temperature peak and late-stage contraction. Temperature interpolation is evaluated separately using independent DSC measurements for another epoxy system, with the rate model constructed entirely from experimental data. With the complete 70 °C curve withheld and six temperatures used for training, the peak-normalized rate root mean square error (RMSE) is 1.11%, and the apparent-conversion RMSE is 0.00271. Withholding each of the five interior temperatures in turn gives rate errors of 1.11–5.58%. Within this dataset, expanding the training-temperature set improves prediction more than increasing the node count along existing curves.
Related
- Thermal History Effects on DSC Kinetic Analysis of DGEBA–DICY Epoxy Curing
- Cross-Linkable Hydrocarbon Resins Bearing β-Styryl Moieties and Allyl Groups: Synthesis, Curing Behavior and Performance for Copper-Clad Laminate Applications
- Recent Advances in Lignin-Based Wood Adhesives: Curing Behavior and Water Resistance
- Cyclic CO2 capture behavior of slag-derived Li4SiO4: A kinetic analysis of CO2 desorption
- Reignition Behavior Evolution and Microscopic Kinetic Mechanism of Lignite at Various Preoxidation Temperatures
BibTeX
@article{yang2026characterization,
title = {{Characterization and Prediction of Epoxy Resin Curing Behavior Using a Data-Driven Kinetic Model Based on Dual Kriging}},
author = {Chao Yang and Jinxin Huang and Xiaolong Xing and Yueming Zhou and Bin Yang and Aiqing Ni and Jihui Wang},
journal = {Polymers},
year = {2026},
month = oct,
doi = {10.3390/polym18212643},
url = {https://www.mdpi.com/2073-4360/18/21/2643},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Characterization and Prediction of Epoxy Resin Curing Behavior Using a Data-Driven Kinetic Model Based on Dual Kriging
AU - Chao Yang
AU - Jinxin Huang
AU - Xiaolong Xing
AU - Yueming Zhou
AU - Bin Yang
AU - Aiqing Ni
AU - Jihui Wang
PY - 2026
DA - 2026/10/25
JO - Polymers
DO - 10.3390/polym18212643
UR - https://www.mdpi.com/2073-4360/18/21/2643
AB - A data-driven kinetic framework based on Dual Kriging is developed to represent epoxy curing behavior and supply a local cure-rate function for finite element (FE) analysis. Error-controlled compression reduces 7200 isothermal observations to 187 points, with maximum and mean normalized reconstruction errors of 0.45% and 0.069%, respectively. The compressed interpolant reproduces the E-51 conversion histories measured by differential scanning calorimetry (DSC), including their late-stage plateaus. A separable 245-node rate representation is implemented in a sequential thermo-chemo-mechanical analysis, combining E-51 experimental data at 55–70 °C with auxiliary Kamal–Sourour data at 30–50 °C. The two kinetics models produce similar temperature and strain histories during staged curing; both underestimate the measured temperature peak and late-stage contraction. Temperature interpolation is evaluated separately using independent DSC measurements for another epoxy system, with the rate model constructed entirely from experimental data. With the complete 70 °C curve withheld and six temperatures used for training, the peak-normalized rate root mean square error (RMSE) is 1.11%, and the apparent-conversion RMSE is 0.00271. Withholding each of the five interior temperatures in turn gives rate errors of 1.11–5.58%. Within this dataset, expanding the training-temperature set improves prediction more than increasing the node count along existing curves.
SN - 2073-4360
ER -