Abstract: Blasting is one of the most common methods for exploitation of rock-fill dam materials, and its fragmentation not only affects the excavation and loading efficiency of material mining, but has a great impact on the compaction quality of dam construction. Therefore, adjusting blast design parameters to control the fragment distribution of mining materials is a key measure for real-time blasting control. Aimed at the deficiency of traditional models in predicting blasting fragmentation, a Levenberg-Marquardt (LM) algorithm based neural network model of two hidden layers is developed for the prediction. Through a case study of blast test fragmentation in a water conservancy project, the validity and practicability of this model and the method are verified.

Key words: hydraulic engineering, blasting technology, dam material excavation, blasting fragmentation prediction, neural network, LM algorithm