Steel corrosion detection with hyperspectral imaging and diffuse spectroscopy: A Feasibility Study

Abstract: Steel is widely used in civil infrastructure construction, such as bridges, pipelines, railroads, etc. Structural steel, i.e. A36, consists of 97-98% iron (Fe), which is highly susceptible to environmental corrosion. Due to this reactive nature, inspections, which are typically initiated in response to visible signs of corrosion, the global cost of corrosion repair is significant, accounting for 3-4% of the GDP. In this study, the feasibility of diffuse spectroscopy, as a noncontact sensing method, is investigated to detect corrosion via changes in the spectral features of steel before and after corrosion. A36 steel samples were exposed to an accelerated alkaline medium for 6 hours, 16 hours, 24 hours, and 21 days. Household bleach at a 1:200 ratio was used as the corrosion medium. The experiment revealed that visual signs of corrosion appeared on samples exposed for 24 hours or longer. The exposure periods were categorized into early (6 and 16 hours) and corroded (24 hours and 21 days) phases. Each sample was scanned with a Visual Near Infrared (VNIR) sensor and X-ray diffraction (XRD) after each exposure period. Corrosion products such as Iron (III) Oxide Hydroxide, Lepidocrocite, Goethite, and Amakinite were confirmed as early as 6 hours using XRD. A maximum 97% change in spectral signatures in the 600-710nm range was observed in the reflectance values recorded by the hyperspectral sensor. Subsequently, an SVM model was trained on 2,214 reflectance values, labeled as early, corroded, and sound. A k-fold cross-validation was used, splitting the data into five folds. For each iteration, four folds were used for training and one for testing. The average true positive rate (TPR) and true negative rate (TNR) were 89.92% and 83.82%, respectively. The novelty of this study lies in the use of a hyperspectral sensor to detect changes in the spectral characteristics of steel due to corrosion, even before visible signs appear.