Maider Iturrondobeitia

A methodology for finding the optimal iteration number of the SIRT algorithm for quantitative Electron Tomography

The SIRT (Simultaneous Iterative Reconstruction Technique) algorithm is commonly used in Electron Tomography to calculate the original volume of the sample from noisy images, but the results provided by this iterative procedure are strongly dependent on the specific implementation of the algorithm, as well as on the number of iterations employed for the reconstruction. In this work, a methodology for selecting the iteration number of the SIRT reconstruction that provides the most accurate segmentation is proposed. The methodology is based on the statistical analysis of the intensity profiles at the edge of the objects in the reconstructed volume. A phantom which resembles a a carbon black aggregate has been created to validate the methodology and the SIRT implementations of two free software packages (TOMOJ and TOMO3D) have been used.

A parameter for the assessment of the segmentation of TEM tomography reconstructed volumes based on mutual information

A method is proposed and verified for selecting the optimum segmentation of a TEM reconstruction among the results of several segmentation algorithms. The selection criterion is the accuracy of the segmentation. To do this selection, a parameter for the comparison of the accuracies of the different segmentations has been defined. It consists of the mutual information value between the acquired TEM images of the sample and the Radon projections of the segmented volumes. In this work, it has been proved that this new mutual information parameter and the Jaccard coefficient between the segmented volume and the ideal one are correlated. In addition, the results of the new parameter are compared to the results obtained from another validated method to select the optimum segmentation.

Methodology to classify the shape of reinforcement fillers: optimization, evaluation, comparison, and selection of models

Abstract This work applies statistical analysis, and classical and advanced machine learning algorithms to classify 7714 aggregates into four categories according to their shape. The aggregates under study are obtained from several grades of carbon black: Vulcan XC 605, Vulcan XC 72, CSX 691, Printex 25, N990, and N762. The classification of the shape is of great significance in order to explain and predict the end-use properties of the composite materials, like mechanical properties. The proposed approach combines transmission electron microscopy and automated image analysis to obtain the dataset of the morphological features that defines the shape of the aggregate, and statistical analysis and machine learning techniques to create the classification models using feature transformation and reduction, parameter tuning, and validation methods in order to achieve robust classification models. The best result is obtained from a classification tree based on evolutionary algorithms with a principal component analysis-based feature reduction that reports an acceptable accuracy, thereby validating both the final chosen model and the methodology.

Modeling of the Mechanical Properties of Carbon-Black Reinforced Rubber Blends by Machine Learning Techniques

Mastering the relationship between the final mechanical properties of carbon black reinforced rubber blends and their composition is a key advantage for an efficient design of the composition of the blend. In this work, some models to predict three relevant physical attributes of rubber blends — modulus at 100% deformation, Shore A hardness, and tensile strength — are built by machine learning methods and subsequently evaluated. Linear regression, artificial neural networks, support vector machine, and regression trees are used to generate the models. The number of used samples and the values for the input variables is determined by a Taguchi design of experiments, and prior to the modeling the uncertainty of the experimental data was analyzed.

Use of Classification Trees and Rule-Based Methods to Predict Shapes of Nano-Aggregates of Reinforcement Fillers

While manufacturing composite materials, reinforcement fillers inevitable collide with each other and subsequently they congregate to aggregates with different shapes. The shape of these nanoparticles aggregates are of great significance for the mechanical material properties and in consequence, knowing the percentage of aggregates of each shape within of a specific group of shapes can give an idea of the final properties of the material. This work classifies aggregates, a new dataset of 5713 carbon black aggregates gathered based on transmission electron microscopy image processing, using several classification trees and rule-based methods. Based on these methods several models are built, trained and tested to perform the classification. And then, the most reliable and accurate model to classify aggregates is selected, obtaining a testing accuracy of the 74.57% according to their shape.