A Regression Model Based on Artificial Neural Networks for a Stone Conical Crusher

Abstract Modeling for design and simulation has gained significance in both academic and industrial domains, primarily due to the availability of specialized software. In the domain of modeling mineral beneficiation crushers, the focus is on controlling variables related to equipment mechanics and material characteristics. In this study, the comminution process of a conical crusher in a stone aggregates production plant was specifically modeled using artificial neural networks. A Feedforward artificial neural network topology was employed to simulate the comminution process. Different network configurations were tested through 50 laboratory experiments, where size data were collected at the crusher’s entrance and exit. The aim was to select the configuration that best matched the actual data by optimizing mean square error, error dispersion, and linear fit. After thorough evaluation, the network configuration with hidden layers of 70 and 40 neurons emerged as the most effective. This configuration yielded a mean square error of 6.24, an error dispersion of 35.05, and a robust linear fit of 0.99. Subsequently, this model was applied to new size distribution data from the crusher’s entrance and exit, resulting in an R parameter of 0.999. The study highlights the effectiveness of the trained artificial neural network model in accurately predicting particle size distribution at the exit of a conical crusher. This has significant implications for optimizing the beneficiation process in stone aggregate production.