关键词: 入料块度, 破碎能耗, 块度识别, 线性回归

Abstract: In order to realize the quantitative prediction of energy consumption based on block char‐ acteristics and provide reliable data support for energy consumption analysis of ore crushing，based on deep learning and image recognition technology，a block intelligent recognition and energy consumption modeling method for ore crushing process is proposed. On this basis，a block recognition system is de‐ ployed in a molybdenum mine crushing station to study the relationship between coarse crushing lumpiness and energy consumption of the mine.The blasting pile image is collected by the drone，and the ore block recognition data set containing multiple data augmentation processing is constructed. The Segment Any‐ thing Model (SAM) of Meta AI is used as the basic network for training and optimization，and the automatic segmentation and particle size parameter extraction of ore particles are realized.Through the image input， particle size analysis and result visualization functions，the key indicators such as total particle number， maximum particle size，F80 and F90 can be output in real time，and the current data of the main motor of the crusher can be collected synchronously，and matched with the identified block parameters，and the ef‐ fective data can be screened out.Through scatter plot and fitting analysis，it is found that the maximum lumpiness，F80 and F90 of the ore are significantly positively correlated with the crushing energy consump‐ tion，and the correlation between the maximum lumpiness and the power consumption is particularly obvi‐ ous.Based on this，a linear regression model is established as W=5.1×10-3 K+2.564 1，which realizes the quantitative prediction of energy consumption based on block characteristics，and provides a new technical path for promoting intelligent control and energy-saving optimization of crushing operation.

Key words: feed lumpiness, crushing energy consumption, lumpiness identification, linear regression