Low-code analysis of glass transition temperatures of structural strengthening adhesives

Structural strengthening adhesives are commonly used for bonding fibre-reinforced composites in the rehabilitation of civil engineering constructions. These adhesives are polymers whose performance is significantly influenced by their glass transition temperatures (Tg). A lower Tg can result in reduced stiffness and strength, particularly in warm temperature conditions. The Tg of adhesives is closely related to the types of adhesives and curing scenarios, yet predictions remain challenging. This study presents a low-code paradigm to examine the Tg values of strengthening adhesives. Multiple mainstream regression models were trained in parallel using 404 experimental data points and 606 synthetic data points generated by an advanced conditional tabular generative adversarial network. The final fine-tuned Light Gradient Boosting Machine (LightGBM) model demonstrated high accuracy in predicting Tg based on the curing conditions and type of adhesive, achieving a coefficient of determination of 0.890 for the test set. Furthermore, the low-code approach was employed to evaluate and interpret the performance of the LightGBM black-box model, and a corresponding web-based Tg prediction application was developed using a light-code platform called Streamlit, providing practical insights for engineers and researchers. The analysis of feature variations in the prediction results offers guidance for achieving optimal adhesive selection and curing strategies.

结构加固胶黏剂（structural strengthening adhesives）常用于土木工程结构修复中纤维增强复合材料的粘接作业。此类胶黏剂属于聚合物，其性能受玻璃化转变温度（glass transition temperature, Tg）的显著影响。较低的Tg会导致胶黏剂刚度与强度下降，在高温环境下该效应尤为突出。胶黏剂的Tg与其类型及固化条件密切相关，但相关预测仍存在较大挑战。本研究提出一种低代码范式，用于探究加固胶黏剂的Tg数值。研究依托404组实验数据点，以及由先进条件式表格生成对抗网络（conditional tabular generative adversarial network）生成的606组合成数据点，并行训练了多款主流回归模型。最终经微调的轻量梯度提升机（Light Gradient Boosting Machine, LightGBM）模型展现出优异的Tg预测精度，可基于胶黏剂类型与固化条件实现Tg预测，其在测试集上的决定系数（coefficient of determination）可达0.890。此外，本研究借助该低代码范式对LightGBM黑盒模型的性能进行了评估与解释，并基于轻量代码平台Streamlit开发了对应的Web端Tg预测应用，可为工程师与科研人员提供实用参考。对预测结果中的特征变量变化开展分析，可为胶黏剂的最优选型与固化策略制定提供指导。