Data-driven integration of synthetic representative volume elements and machine learning for improved microstructure-property linkage and material performance in ceramics

Abstract

Ceramic materials, characterized by their heat resistance, dielectric properties, and mechanical performance, are often compromised by brittleness due to microstructural inclusions. These inclusions, such as defects, secondary phases, and pores, are critically analyzed for their impact on material strength and fracture properties. In this study, the linkage between microstructure and properties in ceramic materials is explored through a methodological approach that combines experimental observations with physics-based and machine learning models. A data-driven approach has been employed, utilizing synthetic Representative Volume Elements (RVEs) derived from X-ray computed tomography scans of ceramics. The methodology involves an automated finite element (FE) simulation process for progressive failure analysis under uniaxial compression and tension. The analysis is conducted using statistical, data-driven, and machine learning techniques, including principal component analysis and k-means clustering, to assess the microstructural features' impact on material performance. The study focuses on the use of RVEs for accurately capturing essential microstructural characteristics and addresses the challenges in developing synthetic models and the limitations of simulation capabilities. The study's findings reveal that less uniform inclusion distribution and a higher standard deviation in inclusion size correlate to lower mechanical performance. By applying data-driven methods, this research contributes to the optimization of material performance and the establishment of structure-property relationships, with a particular emphasis on the influence of inclusions and defects on the mechanical behavior of ceramic materials.