An artificial neural network model to predict the performance of induced roll high-intensity magnetic separator while upgrading low-grade iron ore fines is presented in the present investigation. Key process parameters of induced roll magnetic separator operation, i.e., applied current, speed of rotor, splitter position, particle size, and feed rate, are used as the network input. A feedforward ANN is designed with one hidden layer, with units varying from 9 to 40 and capable of estimating the separation performance (iron content, silica content, iron recovery, and silica rejection of the magnetic product) in an induced roll magnetic separator. Simulated results obtained using the neural network modeling correspond closely to the actual results with a correlation of coefficient (R²) > 0.95 in training and testing stages. The developed models are further used to understand the sensitivity of the studied variables. The effects of speed of the rotor and splitter position are found to be more sensitive to the separation. Furthermore, an optimized solution is designed to predict the best achievable grade-recovery curve for this pre-concentration.

本研究提出了一种人工神经网络模型，该模型可预测诱导辊高强度磁选机的性能，同时提升低品位铁矿粉的质量。感应辊式磁选机运行的关键工艺参数，即施加的电流，转子的速度，分离器的位置，粒度和进料速度，都用作网络输入。前馈人工神经网络设计有一层隐蔽层，单位从9到40不等，并且能够估算感应辊式磁选机的分离性能（铁含量，二氧化硅含量，铁回收率和二氧化硅对磁性产物的排斥）。使用神经网络建模获得的模拟结果与实际结果非常接近，且相关系数为（R ²）在培训和测试阶段> 0.95。所开发的模型进一步用于理解研究变量的敏感性。发现转子的速度和分流器位置的影响对分离更敏感。此外，设计了一种优化的解决方案，以针对该预浓缩预测最佳可达到的品位-恢复曲线。