Prediction of blast-induced ground vibration using artificial neural networks

• Autor(a) principal: Zorzal, Caroline Belisário
• Data de Publicação: 2022
• Outros Autores: Santos, Francisco Lledo dos, Silva, José Margarida da, Souza, Rafael de Freitas
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/34020
• topic: Velocidad máxima de vibración de partículas; Vibraciones sísmicas; Voladura de rocas con explosivos; Redes neuronales artificiales; Ecuaciones empíricas.; Velocidade de vibração de pico de partícula; Vibrações sísmicas; Desmonte de rochas por explosivos; Redes neurais artificiais; Equações empíricas.; Peak particle velocity; Blast-induced ground vibration; Rock blasting; Artificial neural network; Empirical equations.

The peak particle velocity is the most commonly utilized metric in the mineral sector for quantifying and evaluating the damage potential of blast-induced ground vibration (PPV). Over time, initiatives have been conducted with the goal of measuring PPV levels. Intelligent systems are potential methods for estimating rock blasting results, due to significant improvements in computer technology. In this regard, the goal of this research is to use artificial neural networks to evaluate seismic vibrations caused by rock blasting with explosives in a mine in Quadrilátero Ferrífero. The database obtained in the field was separated into training (70%) and test (30%) samples. Different groups of variables were examined considering the necessity of selecting appropriate input variables for neural network training. The distance between the monitoring and detonation points, as well as the maximum charge per delay, were input variables in the network that performed best. The same database was used to compare the performance of neural networks with the performance of empirical and multiple regression models. Finally, in terms of coefficient of determination (R2) and root mean square error (RMSE) for measured and predicted data, the neural network model outperformed empirical equations and multiple regression. Furthermore, the importance of choosing the right input variables when using neural networks to estimate PPV was demonstrated.