| dc.description.abstract | Blasting is the cheapest and most effective way of constructing the tunnel. However, it produces many undesirable consequences from blasting. Some of them are not so important but with increasing, tunneling practice in urban areas blast-induced vibration is a serious problem that has to be taken into consideration while constructing the underground tunnel. There are various rules and regulations that have to be followed to restrict the vibration from blasting but the study of Peak Particle Velocity (PPV) is the main for predicting and limiting the vibration.
In this thesis, a detailed literature study is conducted for both conventional and machine learning techniques used while predicting the ground vibration due to blasting. The summary of the literature review is compiled in tabulated form in the report. Two datasets from E39 Kristiandsand – Mandal tunnel blasting project and the Karlberg tunnel in Moss are used in this study to predict the vibration. Two performance indices, coefficient of correlation (R2) and mean square error (MSE) are used to compare and choose the best result in this study.
Based on the literature study, four empirical relations from the conventional method and 8 models from machine learning techniques are chosen to predict the blast-induced vibration. Using the conventional method, the empirical relation proposed by Ambressys–Hendron shows the best performance in predicting PPV for both datasets. Their performance indices are found to be (R2 = 0.5296, MSE = 9.53), ( R2 = 0.9141, MSE = 27.14) for training and (R2 = 0.2769, MSE = 16.60),( R2 = 0.885, MSE = 37.83) for testing, dataset 1 and 2 respectively. Using the machine learning model, for dataset 1, the lightGBM model performs better and for dataset 2 ExtraTree gives a better prediction with cross-validation scores of R2mean=0.65, MSEmean=6.207 and R2mean=0.885, MSEmean=17.927 for dataset 1 and 2 respectively. Parametric analysis of the input parameters are done and machine learning models are re-run and final results are predicted. For dataset 1, the machine learning model increases the prediction results by 47.04% and 32.71% for the training and testing datasets respectively. Similarly, dataset 2 gives 8.59% and 3% better prediction results for training and testing datasets. Thus, the ML model developed in this thesis work can accurately predict the vibration from tunnel blasting and it can be used in other tunnel blasting sites and datasets to predict the blast-induced vibration. | |