Use of support vector machines, neural networks and genetic algorithms to characterize rubber blends by means of the classification of the carbon black particles used as reinforcing agent

In carbon black reinforced rubbers, the shape of the carbon black aggregates has a very significant influence on the final properties of the material. Accurately classifying these particles by shape has proven to be difficult, but the results of the classification would allow to model the final mechanical properties of the material. In this work, 21 features are measured from 7714 isolated filler images obtained from TEM images and used for the classification. Support vector machines and artificial neural network techniques are used to classify the aggregates using a methodology to tune the algorithm parameters to improve the performance of the models. Also, genetic algorithms are applied to make a feature selection in order to get most robust and accurate models. It is demonstrated that the combination of genetic algorithms with support vector machines and artificial neural network improves the classification results and minimizes the complexity of the resulting model.