Moment Estimation and Product CSD Prediction for Batch Crystallization Process via Adaptive Receding-Horizon Kalman Filter

To overcome the influence from time-varying process uncertainties of seeded batch cooling crystallization in engineering practice, an adaptive receding-horizon nonlinear Kalman filter (ARNKF) is proposed in this paper for the moment estimation and prediction of the product crystal size distribution (CSD). The proposed ARNKF consists of two parts: one is an adaptive algorithm constructed to estimate the initial values of the process and measurement noise covariances before taking a moving time window for state estimation, and the other is an adaptive fading factor strategy conducted inside the moving time window to timely adjust the noise covariances for computation, so that the estimation accuracy on moments related to the total number and mass of crystal population could be effectively improved along the cooling crystallization process. Based on the estimated moments in real time, a prediction algorithm of product CSD is established by using the updated kinetic model parameters such that the prediction accuracy could be substantially improved along the crystallization process. Simulation study and experimental results on the seeded batch cooling crystallization process of β form l-glutamic acid well demonstrate the effectiveness and advantage of the proposed method.