Evaluating the effect of curing conditions on the glass transition of the structural adhesive using conditional tabular generative adversarial networks

Owing to its structural advantages, adhesively bonding fibre-reinforced polymers have been a promising solution for strengthening constructions. However, the effectiveness of this technique is significantly influenced by the material properties of the adhesive layer, which are largely determined by its curing condition. A comprehensive analysis of the effect of curing conditions on structural adhesives is hampered by the lack of sufficient experimental data. To mitigate such a limitation, this present study utilises a deep machine learning (ML) tool, the conditional tabular generative adversarial networks (CTGAN), to generate plausible synthetic dataset for developing a robust data-driven model. An artificial neural network (ANN) was trained on synthetic data and tested on real data, following the "Train on Synthetic - Test on Real" philosophy. The ultimately developed CTGAN-ANN model was validated by newly conducted experiments and several published studies (R2 >= 0.95), which demonstrated the ability to provide accurate estimates of the glass transition temperature values of the polymer adhesive. A comprehensive evaluation of the effect of each curing condition variable on the adhesive was performed, which revealed the underlying relationships, indicating that curing temperature and curing time have a positive effect, but that curing humidity has a negative effect. The ML model developed could inform the practical use of the structural adhesive in civil engineering.