Automatic detection of steel rebar corrosion based on machine learning and light spectrum of fiber optic corrosion sensors

A method for automatic monitoring of steel rebar corrosion by integrating machine learning (ML) with single mode-multimode-single mode (SMS) fiber optic corrosion sensors is proposed in this study. SMS fiber optic corrosion sensors are fabricated in the laboratory and employed for corrosion monitoring of steel rebar in 3.5 wt% NaCl solution. The data of both the transmission spectrum of the SMS fiber optic corrosion sensor and the corrosion-induced mass loss of steel rebar are collected for training ML models. A total of twelve ML algorithms is trained and compared based on the whole and portion of the light spectrum database. Results show that only seven ML algorithms demonstrate good performance based on the whole original transmission spectrum data obtained from the SMS fiber optic corrosion sensors. However, they show poor performance based on portion of the database in corroded chronological order in comparison with those based on the whole database due to the nonlinear relationship between the corrosion-induced mass loss of steel rebar and the shift of the transmission spectrum of the SMS fiber optic corrosion sensors. The limitations of the ML algorithm based on laboratory data in this study are discussed and future work regarding real structure applications are also anticipated.