Inferential estimation and control of chemical processes using partial least squares based model

Abstract

The use of inferential estimation model as a strategy to overcome the lack of efficient on-line measurement for product qualities is proposed. This strategy makes use of easy to measure secondary variables, such as temperature and pressure to infer the value of non-measurable primary variables such as chemical composition. As a case study, a fatty acid fractionation column from a local company was considered. The plant that was simulated using HYSYSTM simulator provided all the required process data throughout the study. To provide the necessary process insights, analyses of dynamic behaviour were carried out. Appropriate secondary measurements with significant relationships with the product composition were then identified for the construction of the inferential estimator within MATLAB® environment. A number of models were considered but nested neural network partial least squares (NNPLS) model was found most proficient. The model was tested online and reasonable performances were obtained. Further refinements were proposed to improve the accuracy and robustness of the estimator. In particular, the issue of data scaling was elaborately addressed. Following the success implementation of the estimator, inferential control of the product quality was examined. In both regulatory and servo controls, better performances were obtained compared to the indirect strategy of controlling product composition using selected tray temperature. This was further improved by employing cascade control. The results obtained throughout this work have illustrated the potential of inferential control strategy and the capability of the hybrid neural network-PLS model as the process estimator. This should therefore serve as an alternative solution to the lack of measurement in chemical process industry. The model developed from the simulation stage is specified to a particular case and it should be verified against the actual process before practical implementation.