Neural network-based optimal control of a batch crystallizer

In batch crystallization, the control of size and shape distributions of crystal product is known to be a difficult and challenging task. Although various model-based control strategies have been widely implemented, the effectiveness of such the control strategies depends heavily on the exact knowledge of crystallization of which the dynamic behavior is complicated and highly nonlinear. In this study, a neural network-based optimal control was proposed to regulate the batch crystallization of potassium sulfate chosen as a case study. A neural network model of the batch crystallizer was first developed to capture the nonlinear dynamics of crystallization in terms of the solution concentration within the batch crystallizer and the moment variables that relate to a crystal product quality over a prediction horizon. Then, the developed neural network model was incorporated in an optimal control framework to find an optimal operating temperature profile for improving the quality of the crystal product. The simulation results showed that the neural network can predict the final product properties and the optimal control integrated with the developed neural network gives a better control performance compared to a conventional linear cooling control technique.