Inverse Estimation of Defect Distribution in Ceramics using Particle Swarm Optimization

This study presents an automated framework for characterizing defects in ceramics by integrating the YOLO-based computer vision and particle swarm optimization (PSO). A total of 16,215 defects across 886 ceramic samples were analyzed and categorized into surface, crack, and chip types. Bounding box data from the YOLO detection were converted to equivalent crack lengths (5–70 . Thereafter, the Weibull distribution parameters for the crack defects were estimated using PSO, yielding a shape parameter β = 2.188 and scale η = 13.59 indicative of the increasing failure rate of typical brittle materials. In addition, PSO outperformed the traditional maximum likelihood estimation (MLE), with lower Kolmogorov-Smirnov statistics (KS = 0.1091 vs. 0.1092), confirming the superior model fit. Furthermore, the theoretical mean defect size (12.04 ) closely matched the observed data (11.98 ), with a 0.43% error. Hence, validation using probability plots, cumulative distribution analysis, and the Gumbel probability paper demonstrated the robustness of this approach. The framework supports real-time defect characterization and reliability assessment, providing practical relevance for industrial quality control. Overall, this approach enhances inverse defect analysis in ceramics by combining modern computer vision and global optimization for accurate and efficient defect estimation.