Modeling, control and measurement in continuous potassium chloride crystallizers
The objective of this study was to employ process control techniques to improve important quality and quantity variables for KCl crystallizers. In order to meet this objective, extensive experimental and theoretical investigations in measurement, modeling and control of KCl crystallizers were carried out. The experimental investigations were mainly focussed on enhancing crystal mean size in a continuous laboratory scale crystallizer. Two sensors for crystal size and concentration measurement in the control loop of the system were individually tried. Of these two sensors, one was an on-line double-sensor turbidimeter. This device, which is particularly suited for saturated slurry systems with suspended foreign insoluble particles, was developed in our laboratory. The other sensor was Par-Tec® 100 (Laser Sensor Technology, Redmond, WA, USA). The theory behind the sensor was extensively investigated. A model capable of predicting the response of the sensor in measuring cordlength distribution (CLD) of a suspension for spherical and ellipsoidal particles was developed. The model was employed to infer the actual particle size distribution (PSD) using the measured CLD. The model was validated using experimental data. These two sensors were employed in the control loop of a continuous KCI crystallizer for measurement of fines suspension. A feedback control loop was implemented to control fines suspension density. The performance of the control loop was evaluated by comparing open and closed loop responses. The theoretical study was aimed at applying a modern process control technique to control a KCl-NaCl crystallization system. To meet this objective, a model of a crystallizer that could replace the physical plant in the control loop was developed. The model was dynamic and could predict important variables of a continuous evaporative cooling KCl-NaCl crystallizer. A nonlinear model predictive controller was assigned to control the theoretical process. A black-box model was used for system identification and generation of a nonlinear model of the plant. The process was found to be MIMO and nonlinear, having significant interactions between its process variables, exhibiting non-minimum phase behavior with restricting constraints on its inputs and outputs. A specific type of model predictive control (MPC), namely extended quadratic dynamic matrix control (EQDMC) was developed and used to control such a difficult process. Closed loop responses of the control systems using EQDMC were compared to those of PID controllers. In complex control systems with several inputs and outputs, the usefulness of the EQDMC was more transparent. The performance of EQDMC in presence of noise was also evaluated.