Development of an Artificial Neural Network for Optimization of Tunnel Blasting Design

發展一種人工智能模型用於優化隧道爆破設計

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

  • Tsz Hang LEE

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date15 Jul 2016

Abstract

Drill and blast on tunnel work is very proper and widely use all over world and the choice on drill and blast method or the use of tunnel boring machine depends on the geology and the length of the tunnel. Nowadays, tunnel blasting is very common at developed urban areas however the impact to the structures from the ground vibration become the most concern issue to the engineers.
Blast-induced vibration is the major side-product from blasting and it will cause significant impact to the nearby structures. The recent innovation approach on the use of intelligent model predicts the vibration in more accurate and effective way. The conventional approach by using convention predictors found the predicted blast-induced vibration was approximated 30% from the measured vibration which will highly affect the productivity from the blasting works.
Artificial Neural Network (ANN) model is widely used over the world for analyzing and optimizing difference engineering problems. In this research, the author use multi-layers perception (MLP) to analyze and predict the tunnel blast vibration on making use of the past blasting records. Total 1014 sets of blast monitoring records were collected from 343 blasts of a current railway tunneling project at urban areas and more than 100 sensitive receivers on building structures, utilities services and geotechnical slope features. In the MLP training, it was discovered that using of correct input parameters will achieve a better performance on coefficient-of-correlation. Compared the results obtained from MLP Model to conventional predictors and local predictor and found that the MLP Model achieved at minimum 25% and 27% higher level of accuracy on root-mean-square-percentage-error than the latter ones respectively. It implies a more accurate blast-induced vibration is developed from ANN and an aggressive maximum instantaneous charge weight can be used for each cycle of the blast.
This research also focuses on the application of Genetic Algorithm (GA), which is widely used over the world for analyzing and optimizing different engineering problems, to optimize the blasting parameters and improve the blast performance for tunnel blasting when excavating a tunnel with limitation on blast vibration constraints. Same blast monitoring records from the previous study on blast-induced vibration prediction using ANN was used with the genetic algorithm for optimization, the authors discovered that the Pull Length and Powder Factor for a tunnel blast could be optimized to its maximum value within the pre-defined vibration limit so that the explosives charge weight will also be maximized for a blast with different tunnel profiles. The results are reviewed and agreed with blasting professionals and found reasonable and acceptable for practical use for future blasting works.