Advanced process control of fatty acid distillation column using artificial neural networks

This thesis discusses the application of neural networks in a fatty acids distillation process control. The objective was to fulfill product purity specifications under varying feed compositions and large process disturbances. The research activities were divided into four main parts. First, a fatty acids distillation process was simulated using HYSYS.Plant1M software. The simulation model was able to closely match the actual process plant data. The model was therefore considered as a reasonable representation of the actual process. Second, a neural network estimator was formulated for a product composition. Here, secondary variables as well as their past values were carefully selected to formulate an accurate and parsimonious model.Third, the inferential model was implemented as an integral part of an inferential control scheme in the distillation column. Simulation results obtained affirmed the potentials of the proposed inferential strategy for composition control. Significant improvements were obtained compared to the widely used strategy of controlling a column temperature. Finally, a neural network based model predictive control scheme was investigated. The implementation on composition control of the distillation process also produced promising results. In conclusion, the works undertaken here have exposed the potentials of neural network models in solving industrial control problems, particularly for composition control in a distillation column