Integration of heating/cooling and evaporation to improve product quality in a batch crystallization process

Crystallization processes have been widely used for separation in many fields to provide a high purity product. In this work, dynamic optimization and neural network (NN) have been applied to improve the quality of the product: citric acid. In the dynamic optimization, optimization problems maximizing both crystal yield and crystal size have been formulated. In this work, a neural network forward model has been designed to provide estimations of crystallizer temperature, concentration of solution and jacket temperature as well as a neural network inverse model has been developed to predict jacket temperature set point. The Levenberg Marquadt algorithm has been used to train the networks and optimal neural network architectures have been determined by a mean squared error (MSE) minimization technique. In controller design, neural network direct inverse control (NNDIC) and neural network model predictive control (NNMPC) strategies have been applied to control the crystallizer temperature. The simulation results have shown that the obtained crystal size from optimization problem is 19% and 30% larger than cooling and evaporation methods, respectively moreover yield increase more than 50%. Both neural network forward and inverse models show good accuracy for the prediction of the system. The robustness of controller is investigated with respect to parameters mismatch. The results have shown that the NNMPC controller provides superior control performance in all case studies.结晶工艺已在众多领域广泛应用于分离过程，以制备高纯度产物。本研究将动态优化与神经网络（Neural Network, NN）应用于柠檬酸产品的品质提升工作。在动态优化环节中，本研究构建了同时最大化晶体产率与晶体粒径的优化问题。本研究设计了神经网络前向模型，用于预测结晶器温度、溶液浓度与夹套温度；同时开发了神经网络逆模型，用以预测夹套温度设定点。采用莱文贝格-马夸特（Levenberg Marquadt）算法训练神经网络，并通过均方误差（Mean Squared Error, MSE）最小化方法确定最优神经网络架构。在控制器设计阶段，本研究采用神经网络直接逆控制（Neural Network Direct Inverse Control, NNDIC）与神经网络模型预测控制（Neural Network Model Predictive Control, NNMPC）两种策略，对结晶器温度进行控制。仿真结果表明，通过本优化问题得到的晶体粒径分别比冷却法与蒸发法高出19%与30%，且产率提升超过50%。所构建的神经网络前向与逆模型均对系统具备良好的预测精度。本研究针对参数失配情况，对控制器的鲁棒性展开了分析。结果表明，在所有案例研究中，NNMPC控制器均展现出更优异的控制性能。