From Optical Inversion to AI Vision: an AI-driven SEM Workflow for Empowering Precise Granulometric Analysis

Abstract The particle characteristics of fly ash are critical for its resource utilization. However, traditional laser diffraction analysis, which relies on an “equivalent sphere” model, struggles to accurately represent the complex and variable particle morphology and agglomeration states. This study presents a systematic comparison between laser diffraction analysis and a method combining scanning electron microscopy (SEM) with the Segment Anything Model (SAM), an artificial intelligence model, for characterizing fly ash from an Inner Mongolia power plant. While laser diffraction analysis provided the volume-based particle size distribution, the SEM-SAM approach enabled automated instance segmentation and statistical analysis of over 280 thousand individual particles, with quantitative comparison facilitated by equivalent circular diameter. The results reveal systematic differences between the two methodologies: laser diffraction analysis, influenced by ultrasonic dispersion and its optical inversion model, showed a broader distribution (Span = 2.194) and detected a higher fraction of sub-micron fine particles (D10 = 0.677 μm). In contrast, the SEM-SAM method, based on direct geometric measurement, yielded a narrower distribution (Span = 1.326), with a median diameter D50 (2.091 μm) closer to the true geometric size of the particles, and demonstrated reduced sensitivity to undispersed agglomerates. This study confirms the unique advantages of the SEM-SAM approach in characterizing true particle morphology and providing statistically robust data, offering a novel, high-throughput technical pathway for the precise analysis of fly ash particle properties.