International Graduate Research Symposium, İstanbul, Turkey, 1 - 03 June 2022
The aim of this study is the prediction of blast vibrations in an Istanbul Region aggregate quarry. Blast-induced vibrations were measured in the field. Special blast seismographs were used to monitor ground vibrations. The seismographs measure both particle velocity and frequency. The measurement location and instantaneous explosive charge were also measured in each blast. All the blast design parameters were recorded. Gaussian Process Regression (GPR) method and classical scaled distance equation were applied to develop prediction models. Sixty-nine training data and 26 test data were used to construct the forecasting models. The GPR model was created by the Regression Learner application of Matlab software. In the modelling stage, GPR model was trained by using 5 different kernel functions: squared exponential, exponential, rational quadratic, matern 5/2 and matern 3/2. Spacing to burden ratio (S/B), bench height to burden ratio (H/B) and scaled distance (SD) were chosen as input parameters. The predicted output is the peak particle velocity (ppv). The scaled distance equation was developed by univariate regression. Seven error criteria were applied to determine capability of the developed models. Absolute error metrics, percentage errors metrics and correlation coefficients were calculated for each model: As a result, it has been revealed that the GPR model, which is a machine learning method, is more successful than the classical scaled distance equation. The GPR model predicts particle velocity with a mean absolute error lower than 1.5 millimeter per second. In addition, twenty-one cases were forecasted with a mean error lower than 2 mm/s. The GPR model is not complex and suitable for practical applications.