8 | CONCLUSION
The first, second and third dissociation constants of eight amines of importance in carbon capture operations, namely, N-(2-aminoethyl)-1,3-propanediamine (n-2AOE13PDA), 2-Methyl-pentamethylene diamine (2-MPMDA); N, n-dimethyldipropylenetriamine (DMAPAPA) and 3,3’-Diamino-n-methyldipropylamine (DAOMDPA), Bis[2-(n,n-dimethylamino)ethyl]ether (2DMAOEE), 2-[2-(Dimethylamino)ethoxy]ethanol (DMAOEOE), 2-(Dibutylamino)ethanol (DBEA) and N-propylethanolamine (PEA) were measured in a temperature range varying from 298.15 K to 313.15 K with 5 K increments. As expected, the dissociation constant values for all studied amines decreased with increasing temperature. The study shows that all amines studied have a higher pKa than MEA. 2-MPMDA, with two primary amino groups, had the highest pKa1 and the second pKa 2. It should be an amine of great interest in carbon capture operations. DAOMDPA, with two primary and a tertiary amino group, also had a high pKavalue. A tertiary amine, Methyldiethanolamine pKa values were the lowest. Using computational chemistry calculations, the first protonated positions for the studied amines were predicted and the results agreed well with the literature.
In terms of modelling, the first pKa values of the studied amines at 298.15K were estimated using the original and modified PDS but also with the Qian-Sun-Sun-Gao (QSSG) method. This method provided the best estimation of pKa values when compared to the experimental data.
Finally, an artificial neural network (ANN) was developed to predict the values of the dissociation constants for the temperature range studied in this work. The input data included the molecular weight, critical temperature, and pressure to identify the compounds as well as temperature as pKa values are temperature-dependent. In addition, the density, dynamic viscosity, refractive index and surface tension were also used as inputs. The predicted values were in very good agreement with the experimental values. An optimum ANN architecture of 8-5-7-1 was selected, its predicted outputs were in a good agreement with targets, with a regression coefficient of 0.99424 and a mean squared error for training, validation and testing of 2.20E-05, 0.0094 and 0.0078, respectively.
The full ANN model was further simplified and optimized by only including the surface tension and the refractive index as inputs.