The evaluation of grinding process using artificial neural network

Abstract

Ball milling has been the subject of intensive research for the past few decades. It is indeed the most encountered mineral processing operation of size reduction. Known as the most energy inefficient process, focus has mainly been on ways of reducing the energy consumption incurred by the operation. There are programs for the computer design of mineral processing circuits, and these programs contain computer simulation models for ball mill design. These models need the input of characteristic breakage parameters for the mineral of interest and these are often determined in a small size laboratory ball mill and scaled up by the program to the conditions of a full-scale ball mill. Models and simulators have been used for plant technical analysis since 1970. Some of these models and simulators were developed for mineral processing operations, whereas some were dedicated to mineral processing operations. The prominent work for the mineral processing applications includes JKSimMet, MODSIM © and its derivatives. A neural network is able to learn complex relationships between related variables and therefore has been widely used as a tool for process modeling. It consists of many simple parallel processing units (called "neurons"), which can resemble the architecture of the human brain, and thus is capable of learning arbitrary nonlinear mappings between noisy sets of input and output factors. The grindability properties of the calcite sample belonging to the Mu?la region were investigated at batch grinding conditions based on a kinetic model. The obtained kinetic model parameters were used to estimate the product size distribution by artificial neural networks (ANN). Then, the experimental and neural network prediction results were compared. © 2015 Elsevier B.V.