Clustering analysis of grid nanoindentation data for cementitious materials

Nanoindentation technology is an advanced method to explore the microscopic mechanical properties of layer or block materials. Grid nanoindentation coupled with statistical analysis is effective to give an insight into multiphase materials just like cement paste, mortar and concrete. However, traditional statistical methods, such as deconvolution analysis and Gaussian Mixture Model (GMM), are limited by the nondeterminacy of normal distribution assumption, computational instability due to random selection of initial values, and some problems induced by large amount of calculation. In this paper, clustering analysis, an advanced analysis method for mixed data and widely used in machine learning field, is developed to deal with grid nanoindentation test data. Calculation results suggested that K-medoid clustering is suitable and highly efficient to explain grid nanoindentation tests. Furtherly, clustering method is more robust than deconvolution analysis and GMM when data size is reduced. In addition, normal distribution assumption is not always available for the mechanical properties of some mineral phases in cement pastes. This work offers a new optional mathematical tool to interpret and understand the multiphase properties of cementitious materials probed by grid nanoindentation technology.