Automated measurement system for detecting carbonation depth: Image-processing based technique applied to concrete sprayed with phenolphthalein.

• An automated system for standard carbonation depth measurements is proposed. • Processing algorithm provides average and maximum carbonation depths. • Computer vision, image processing, and Convolutional Neural Networks are used. • Convolutional Neural Networks separate carbonated aggregates from paste. • The measurement system is challenged in terms of repeatability and reproducibility. This paper aims at discussing an automated measurement system for detecting carbonation depth in concrete sprayed with phenolphthalein. Image processing and Convolutional Neural Networks strategies are exploited to accurately separate the carbonated and non-carbonated areas and to remove those aggregates on the carbonation front that could bring to a wrong evaluation of the carbonation depth. Very strong correlation (R2 = 0.96) is found between results provided by the proposed approach and the method suggested by the EN 13295 standard. The expanded uncertainty (coverage factor k = 2) of this novel approach is 0.08 mm. ANOVA analysis performed in multi-operator tests proved that the highest source of uncertainty is the measurement system, which, on the other hand, is robust to changes in the operator performing the measurement. [ABSTRACT FROM AUTHOR]