Constituents phase reconstruction through applied machine learning in nanoindentation mapping data of mortar surface

Abstract

In the present study, data generated from nanoindentation were used in order to reconstruct the surface constituent phases of mortar grids through machine learning algorithms. Specifically, the K-Means algorithm (unsupervised learning) was applied to two 49 measurement (7 × 7) datasets with information about the modulus (E) and hardness (H) in order to discover the underlying structure of the data. The resulting clusters from K-Means were then evaluated and values range assigned so as to signify the various constituent phases of the mortar. Furthermore, another dataset from nanoindentation containing information about E, H, and the surface colour of the measured area (obtained from an optical microscope) was used as the training set in order to develop a random forests model (supervised learning), which predicts the surface colour from the E and H values. Colour predictions on the two 7 × 7 mortar grids were made and then possible correlations between the clusters, signifying constituent phases, and the predicted colours were examined. The groupings of data in the clusters (phases) corresponded to a unique surface colour. Finally, the constituent phases of the mortar grids were reconstructed in contour plots by assigning the corresponding cluster of the K-Means algorithm to each measurement (position in the grid).