Elucidating the Origin of Heterogeneous Stresses in Granular and Microstructured Materials Using Data Science Methods.

다국어 초록 (Multilingual Abstract)

The heterogeneity in microstructured and granular materials due to micro-scale structures causes high stresses when an external load is applied. The high stresses in microstructured and granular materials are seen in the form of peak-stress clusters and force chains, respectively. These high stresses play a significant role in governing materials’ strength; hence it is crucial to know their relationship with the micro-scale properties to design high-strength materials.We use data-science-based methods to understand the relationship between microscale properties and high stresses in these materials. Since both materials are different, the data science approach for finding the relationship between the micro-scale properties and high stresses is also different. A learned feature-based approach is utilized for microstructured materials, and for granular materials, a hand-crafted feature-based approach is used.In the learned feature-based approach, the microstructures’ grain features that are causing peak-stress clusters are detected using a deep-learning-based Convolutional Encoder-Decoder (CED) method. The CED method is first trained to predict linear elastic calculations of von Mises stress fields in synthetically-generated microstructures. The accuracy analysis showed that the proposed method is well-suited to predict the characteristics of the peak-stress clusters. Later, saliency maps are obtained to realize the microstructures’ regions that were causing peak-stresses. The proposed CED method is computationally much faster than existing numerical schemes. With high accuracy, fast predictions, and the ability to detect long-range effects, the CED method can be used for the on-site design of high-strength materials.In granular materials, the effects of particle size and cohesion on the force chains are examined in bi-disperse (with two types of particles) systems using a hand-crafted feature-based approach. The effect of particle size is studied in terms of various disorders induced in the granular materials. Here, the disorders induced due to the difference in the size of the two types of particles are denoted as size-disorders. In the hand-crafted feature-based approach, the features describing the disorders and cohesion are then examined against the graph-based descriptors of force chains using data visualization. The particles in the granular materials with stresses above a threshold and forming long continuous chains are characterized as force chains. It was observed that no force chains are present in ordered materials. Moreover, with the increase in size-disorder, the force chains are observed even in the materials without positional disorders. It was also observed that in the granular materials, irrespective of positional disorder, the force chains do not form with an increase in the cohesion of the particles. This study shows that the size and cohesion of the particles play an important role in causing force chains irrespective of positional disorders.In conclusion, the work in this thesis shows that data science-based methods can be effectively used to understand the relationship between the micro-scale properties and high stresses of microstructured and granular materials.