Comparison of intelligence science techniques and empirical methods for prediction of blasting vibrations

Masjed-Soleiman dam is one of the national projects in Iran, having the most complexity and a lot of underground excavations in its scale. The damage of blast induced vibrations in the excavations of this project results in decreasing the safety of the newly constructed structures. Therefore, prediction and control of the vibrations is a crucial task in the Masjed-Soleiman project. To predict the vibrations in this specific area, three approaches were used and the results were interpreted and compared. The vibrations were first predicted using several widely used empirical methods. Then, two intelligence science techniques namely general regression neural network (GRNN) and support vector machine (SVM) were used for prediction as well. In this study, predictions of blast induced ground vibration were performed by taking into consideration of maximum charge per delay and distance between blast face to monitoring point. Obtained results indicated that average correlation coefficient between measured and predicted PPV of SVM is 0.946 compared with 0.92 of GRNN and 0.658 of the best empirical approach in testing dataset. In addition, relative root mean square error (RMSE) and associated running time of SVM are of the main reasons proving the strength and robustness of this machine learning methodology. Hence, it can be concluded that the SVM technique is a faster and more precise than the GRNN and empirical methods in prediction of PPV comparatively.

► Blast induced vibrations decrease safety of structures.
► Peak particle velocity (PPV) is used for structural responses to ground vibrations.
► PPV is estimated based upon maximum charge per delay and distance.
► Support vector machine is used for the prediction of blast induced ground vibration.
► There is a strong correlation between predicted and measured data.