Residence Time Distribution-Based Smith Predictor: an Advanced Feedback Control for Dead Time–Dominated Continuous Powder Blending Process

Abstract

Abstract Purpose In continuous manufacturing (CM), the material traceability and process dynamics can be investigated by residence time distribution (RTD). Many of the unit operations used in the pharma industry were characterized by dead time–dominated RTD. Even though feasible and proper feedback control is one of the many advantages of CM, its application is challenging in these cases. This study aims to develop a feedback control, implementing the RTD in a Smith predictor control structure in a continuous powder blender line. Methods Continuous powder blending was investigated with near-infrared spectroscopy (NIR), and the blending was controlled through a volumetric feeder. A MATLAB GUI was developed to calculate and control the concentration of the API based on the chemometric evaluation of the spectra. The programmed GUI changed the feeding rate based on the proportional integral derivative (PID) and the Smith predictor, which implemented the RTD of the system. The control structures were compared even on a system with amplified dead time. Results In this work, the control structure of the Smith control was devised by utilizing the RTD of the system. The Smith control was compared to a classic PI control structure on the normal system and on an increased dead time system. The Smith predictor was able to reduce the response time for various disturbances by up to 50%, and the dead time had a lower effect on the control. Conclusions Implementing the RTD models in the control structure improved the process design and further expanded the wide range of applications of the RTD models. Both control structures were able to reduce the effect of disturbances on the system; however, the Smith predictor presented more reliable and faster control, with a wider space for control tuning.