A novel application of neural networks for instant iron-ore grade estimation

The inverse problem of magnetic petrophysics is to determine magnetic contents of rocks/ores provided with their susceptibility readings already known. This has not been studied yet due to its unknown applications. This paper proposes a novel application of solving this inverse problem for instant estimation of iron-ore grade in mining. This application is based on numerical simulation using neural networks assisted with 2D interpolation for determining the magnetite and hematite contents through known magnetic susceptibility data. This study shows that a four-layer multilayer perceptron (MLP) trained properly is able to accurately simulate the magnetic contents of iron-ores, which can lead to instant estimation of iron-ore grade in situ in iron-ore mining.