Although there is a sound and rigid theoretical background describing the evolution of chemical reactions of simple and well-defined substances, when real life materials, e.g., mineral ores, are involved, both equilibrium and kinetic approaches are not sufficient to model the chemical process and give accurate predictions, because critical parameters of the theoretical models are unknown and cannot be easily determined. In this work, we use ML and signal analysis methods to model the reduction of dried or calcined (Nchwaning and Comilog) manganese (Mn) ores using H2, in the context of HAlMan [1] project. The experimental data, provided by NTNU, for each initial mass and crucible temperature consisted of the measurements of the mass loss and the crucible temperature every 5sec for 2hr, for a total of 40 experiments.

The model developed involved the following steps: 1) Denoising and curve fitting of each experimental curve by a sum of exponential functions, to determine the maximum weight loss of the specific sample due to the H2 reduction, 2) Developing a generic theoretical model to describe the time-dependent reduction process, using the generic parameters $A$ (pre-exponent factor), $E_a$ (activation energy) and $n$ (order of the reaction) of the Arrhenius kinetic equation [2], 3) Developing an appropriate algorithm to compute the parameters {$A$, $E_a$, $n$} for each experiment, 4) Using ML methods to model the reduction process for the Mn ores under the specific experimental conditions (pre-calcination, crucible temperature) [3,4].