Application of artificial neural networks in modeling limestone-SO2 reaction

Four varieties of limestone distinguished on the basis of pore-size distributions were exposed in dynamic 10-, 20- and 50-ppm SO2 atmospheres for up to 500 h at 25 degrees C and 100% rh. The resulting conversion of the limestone was measured as a function of the reaction product formed and the change in porosity. These conversions could be predicted correctly using either the shrinking unreacted core model or the distributed pore model. An artificial neural network (ANN) was also trained for the purpose. All three approaches predicted conversions that fitted well with the observed data; however, those predicted by ANN were the most accurate. Further, the weights determined for ANN on the basis of three limestone varieties also accurately predicted the conversions of the fourth variety for which no information was supplied in the training phase, showing that ANN can also be used successfully to estimate gas-solid noncatalytic reactions.