Image Processing and Machine Learning to investigate fibre distribution on fibre-reinforced shotcrete Round Determinate Panels

Mirko Manca, Ali Karrech, Phil Dight, Daniela Ciancio

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

Fibres in fibre reinforced concrete (FRC) and fibre reinforced shotcrete (FRS) are known to increase the toughness of the composite material thanks to their ability to transfer tensile stresses even after cracks have opened. However, fibres are deemed responsible for uncertainty in mechanical performance of FRC and FRS. This uncertainty is ascribed to how the fibres are physically located with respect to the crack. In this study a novel approach is presented to measure the distribution of fibres from digital images. A computer code that features Image Processing and Machine Learning algorithms has been developed to extract: i) the 2D location, ii) the mode of failure, iii) the orientation, iv) the pull-out length for each single fibre and v) the total number of fibres bridging the crack. The Machine Learning Algorithm is trained with a database that is attached to this article. This methodology, apart from giving an understanding of how fibres are distributed over the crack, provides the main input of several numerical models that simulate the behaviour of FRC/FRS taking into account the position and orientation of each single fibre. The main advantage of this approach over existing methods is that the hardware required to carry out the analysis consists of a simple smartphone camera while an output with errors within a few thousands parts of the actual measures. The algorithm is employed to analyse the fibres’ distribution for 9 Round Determinate Panels (RDP) made of wet-mix shotcrete. The spatial 2D-location of fibres in the crack is tested for randomness with a Monte Carlo procedure and the fibres’ distribution is found to follow an Independent Random Process. For fibres that pulled-out from one side of the crack, a further 3D investigation has been carried out. The analysis on the fibres’ orientation confirmed the conclusion found by other authors that fibres tend to align perpendicular to the direction of spraying. Furthermore, it is unlikely to find fibres perpendicular to the crack surface. This last conclusion, however, is specific of the properties of fibres tested in this study, and in general, of macro synthetic fibres that tend to snap when their orientation is closer to the direction of the main stress. Finally, in the attempt of examining the representativeness of smaller version round determinate panels, the fibres’ distribution of 3 Mini RDP has been assessed and is found to be comparable to that of full size RDP. This suggests that a similar fibre's distribution can be achieved even in a panel that is half the weight of the full size RDP.

Original languageEnglish
Pages (from-to)870-880
Number of pages11
JournalConstruction and Building Materials
Volume190
DOIs
Publication statusPublished - 30 Nov 2018

Fingerprint Dive into the research topics of 'Image Processing and Machine Learning to investigate fibre distribution on fibre-reinforced shotcrete Round Determinate Panels'. Together they form a unique fingerprint.

Cite this