Iterative learning parameter estimation and design of SMB processes

The simulated moving bed (SMB) is a continuous separation chromatography process. The design of the SMB operation condition is crucially important as it affects the purity of the products. For the physical model based design, the parameters of the system are a pre-requisite and are determined in an offline experiment. More often than not, changes in the operation of the process mean the offline determined parameters may no longer be a true reflection of the process and need to be re-determined. In this work, a novel design strategy which combines the physical-based and the empirical based design in an iterative learning scheme is proposed. In the physical based design, the parameters of the physical model are recalculated through an on-line scheme which makes use of the current on-line measured data from the UV signals obtained in the real process. Using the standing wave equations, the obtained parameters are then used to design the operation condition. With new sets of measurements, the model parameters are refined and the operation condition is also updated until the parameters converge. To overcome the model structure mismatch, the empirical based design is employed for fine improvement. The differences between the desired output and the actual output are used to modify the flow rates until the goal is attained. By a sequence of the two-phase design, the desired quality can be attained systematically. The proposed method is applied to a virtual eight-column SMB process to verify the effectiveness of the proposed method.