Machine learning-based automatic detection and prediction of cracks and corrosion using spatiotemporal measurements from distributed fiber optic sensors

Monitoring and predicting damages of civil infrastructure are essential for safe and efficient operation and maintenance. This paper presents a digital twin-based approach for automatic detection and prediction of cracks and corrosion utilizing spatiotemporal measurements of strains from distributed fiber optic sensors. Generative machine learning techniques are used to improve the quantity and quality of datasets used to develop damage detection and prediction models. The performance of the approach was evaluated using laboratory experiments through case studies on reinforced concrete beams and steel pipes. Results demonstrated that cracks and corrosion were detected accurately (accuracy>0.98) and efficiently (latency = 0.17 ms). Predictions of strain distributions were performed 7 min ahead for cracks and 21 h ahead for corrosion. The effects of sensing parameters on performance were investigated, enabling sensor configuration optimization. The presented approach advances the ability to monitor and predict damages based on advanced machine learning and distributed fiber optic sensing techniques.