Use of decision tree models based on evolutionary algorithms for the morphological classification of reinforcing nano-particle aggregates

Abstract

An original method is presented to classify nano-particle aggregates into one of the morphological classes that were previously proposed in the literature. Carbon black was selected as a study case in this work, since it is a nanoreinforcement used massively in many industries, but the proposed method can be applied to other particulate aggregates as well. This methodology consists of three steps: (i) transmission electron microscopy image processing to compute a set of twenty-one morphological characteristics for each aggregate including the new attributes proposed herein, (ii) a multivariate analysis of the dataset to reduce the problem dimensionality and (iii) creation and assessment of decision trees based on evolutionary algorithms to classify the aggregates. The effectiveness of the method was proven for a balanced-per-class sample of forty-eight selected aggregates. The original dimension of the problem is reduced to three principal components, which explain 90% of the total variance. The best model classifies an aggregate into one of the four morphological classes in a simple comparison process. The accuracy of the applied model to classify new aggregates was 75%.