Machine learning of the mineralogical and chemical constituents’ influences on carbonation efficiency

Abstract Cheap and efficient computational tool is needed to understand the influences of rock’s mineralogical and chemical constituents on the mineral carbonation efficiency. This work employs machine learning technique to learn some subtle contributions of minerals and technical parameters to carbonation yields in earthly materials. After training different configurations of Artificial Neural Network models, statistical appraisal shows that ANN[17-4-1] possesses the best performance criteria. Using this best-performing network, it was found that carbonation efficiency becomes lower as the particle size increases with maximum effects of size change giving up to 88, 180 and 289% reductions in carbonation efficiency corresponding to 10, 20 and 30% particle size increases, respectively. The simulation further highlights that the effects of particle size on carbonation efficiency can be overcome or, at least, reduced, by prolonging the duration of the carbonation. The study shows that there is optimum level of increase in the calcium amount to yield increased carbonation efficiency, following which the marginal increase in % of calcium continue to result in decrease of efficiency. There is gradual rise in efficiency as the % of silica and iron increase in the minerals, with up to 1541 % rise in efficiency at 30% increase in iron content. This is connected to the relatively higher level of magnesium in the sample considered. It was found that the high temperature (> 100oC) combined with high pressure (> 10 bars) favours increased efficiency with rise in pH value. Similarly, low temperature (< 100oC) with low pressure (< 10 bars) increased efficiency with rise in pH value. On the contrary, low pressure with high temperature yields reverse trends in efficiency for increase in pH. Pressure, carbonation time and temperature, all have positive effects on carbonation efficiency. But, curiously, at very high original carbonation efficiency, e.g., above 50%, reverse trend was predicted by the ANN model for temperature rise. This may mean that at very high carbonation level, it is of no benefit to raise the temperature further, as this may lead to decarbonation or desorption of CO2 in the medium. This study is among the first set of publications, in the open literature, to utilize the concept of machine learning to predict and forecast the carbonation efficiency of rock materials based on mineralogical contents and experimental conditions.