Intelligent Fragmentation: A Systematic Review of Imaging and AI for Blast Optimization in Open-Pit Mining

Blast-induced rock fragmentation in open-pit mining constitutes a fundamental component of the mine-to-mill value chain, directly influencing downstream energy consumption, production cost, operational safety, and processing efficiency. This chapter provides a systematic review of studies published between 2000 and 2025, focusing on the application of image processing and artificial intelligence (AI) techniques for the analysis, prediction, and optimization of blast-induced fragmentation. The reviewed literature is structured around three interconnected themes: (i) classical and advanced empirical fragmentation models, including Kuz–Ram, Rosin–Rammler, and Swebrec formulations; (ii) image-processing-based methods for particle size distribution assessment, with emphasis on scale calibration, illumination correction, and the limitations of two-dimensional observations; and (iii) intelligent data-driven approaches based on machine learning and deep learning, as well as their integration with multi-objective optimization. Only peer-reviewed studies reporting quantitative field or laboratory data and explicit performance metrics were considered, with particular emphasis on works employing spatially independent validation. The analysis shows that empirical models remain useful for preliminary blast design due to their simplicity and interpretability, although their accuracy is highly sensitive to site-specific calibration and geological variability. Image-based fragmentation analysis offers a rapid and cost-effective monitoring tool when rigorous acquisition and calibration protocols are applied; however, particle overlap and the inherently two-dimensional nature of images may introduce systematic bias. In contrast, AI-based models generally demonstrate superior predictive accuracy for key fragmentation indices such as P50 and P80, provided that high-quality datasets and robust validation schemes are employed. Nevertheless, challenges related to data standardization, computational cost, and limited interpretability continue to restrict their large-scale industrial adoption. Recent studies increasingly support hybrid, physics-informed frameworks that combine calibrated image-derived features with data-driven models under explicit physical constraints. Such integrated approaches, particularly when coupled with multi-objective optimization, show strong potential for improving fragmentation control, reducing energy consumption, and enhancing operational stability in open-pit mining systems.