Advancements in Predictive Modeling for Peak Particle Velocity in Rock Blasting: Tracing the Evolution from Empirical Models to Artificial Intelligence Techniques

Abstract The prediction of peak particle velocity (PPV) constitutes a fundamental element for operational safety and efficiency in rock blasting activities using explosives. This review article systematically examines the methodological evolution of predictive PPV models, from pioneering empirical approaches to contemporary artificial intelligence algorithms. Traditional empirical models, exemplified by the U.S. Bureau of Mines (USBM) and Langefors-Kihlstrom formulations, established the foundations of vibration prediction through simplified mathematical relations based on scaled distance. However, their inherent limitations, particularly the dependence on site-specific constants and the inability to model complex non-linear relationships, motivated the transition to more sophisticated approaches. The incorporation of artificial intelligence techniques (Artificial Neural Networks (ANN), Support Vector Machines (SVM), ensemble methods, and deep learning architectures) has improved reported predictive accuracy, although this review shows that advantage to be conditional, not universal. This review analyzes the advantages, limitations, and practical applications of each model category through a structured, quantitative synthesis of the literature. Its contributions are a critical comparison of reported performance across studies, a balanced appraisal of the recurring limitations of data-driven models (data dependency, interpretability, generalization, and deployment), a practitioner-oriented framework for model selection based on data availability and site conditions, and a research-gap roadmap of underexplored directions (hybrid physics-and-data models, uncertainty quantification, explainable AI, and IoT-enabled monitoring). Future trends point towards the integration of Internet of Things (IoT) technologies, the development of hybrid models combining physical knowledge with machine learning, and the implementation of uncertainty quantification techniques, aiming for more robust and reliable control of the environmental impacts of rock blasting.