A IoT-Enabled Predictive Intelligence for Real-Time Failure and Damage Evolution Monitoring of Polymer Composites
Damage assessment of carbon-fibre-reinforced polymer composites is still challenging since the damage occurs as a combination of matrix cracking, interfacial debonding, delamination and fibre fracture.
Damage assessment of carbon-fibre-reinforced polymer composites is still challenging since the damage occurs as a combination of matrix cracking, interfacial debonding, delamination and fibre fracture. The present work proposes a framework for predictive-intelligence based on IoT for multiaxial fatigue and compression-after-impact (CAI) CFRP experiments, employing publicly available acoustic-emission (AE) datasets. A causal CNN–GRU attention model is developed by integrating time-domain, spectral, wavelet, loading-history and trend features to estimate the damage evolution, imminent failure and normalized remaining life. To avoid leakage and to test robustness, specimen-level splitting, training-only normalization, five-seed evaluation and independent cross domain validation were used. The framework was evaluated on held-out fatigue specimens, and yielded a damage-estimation R² of 0.946, RMSE of 0.052, failure-prediction AUROC of 0.967, F1-score of 0.918, and remaining-life MAE of 0.071. External validation showed R² of 0.889 and AUROC of 0.912 and real-time replay had mean latency of 28.6 ms. The predicted critical transition also remained close to the independently reported unstable damage region, indicating useful transfer across distinct loading conditions. The results show that temporally resolved AE monitoring can support early warning and continuous structural-health assessment of polymer composites, providing a practical bridge between sensing, prognosis, and IoT-enabled condition-based maintenance.