Control of an industrial hydrolysis batch process using a model-based control strategy is described. For this purpose, non-linear artificial neural network models of the process dynamics are identified with the aid of a semi-empirical model. Neural networks are trained using an augmented training data set. Neural network models predict the TiO2 precipitation rate for each batch. Using this prediction, some of the process parameters can be optimised through the application of the flexible recipes concept to control the quality of the intermediate product and the duration of each batch. A simulator has been designed to study the application of flexible recipe instructions. Different adaptation mechanisms which use on-line process measurements are discussed to improve model accuracy and to counteract changes in process dynamics.