Development of nanomodified self-healing mortar and a U-Net model based on semantic segmentation for crack detection and evaluation

Abstract

In this study, a cement-based self-healing (SH) mortar mixed with multiwalled carbon nanotubes (MWCNTs) was developed, and the effect of MWCNT addition on the electromechanical properties of mortar was studied. In addition, a deep learning based U-shaped Network (U-Net) model was adopted for the crack healing evaluation in SH mortar specimens. The trained U-net model is capable to detect the cracks with the accuracy of 99 %. The MWCNT content in the mortar was varied from 0.05 % to 0.10 % by weight of the cement. A minerals-based SH admixture was incorporated at concentrations of 2.5 %, 5 %, and 7.5 % by weight of cement. Compression and flexural strength tests were performed to investigate the mechanical performance of the nanomodified mortar. Piezoresistive sensing properties of MWCNT/cement composite were monitored during flexural test. For crack healing evaluation, cracked mortar prisms were exposed to water for 28 days. Mechanical and piezoresistive sensing evaluation demonstrated improved mechanical performance of SH mortar with appreciable self-sensing properties. Specimens with 2.5 % admixture exhibited a crack sealing efficiency of approximately 90 %, while the cracks in specimens with 5 % and 7.5 % SH admixture content healed 100 % after 28 days of water exposure.