Iterative Learning Control for Preparative Chromatography

Sammanfattning: There is a desire for a fully automated downstream process in pharmaceutical protein production. One part of the downstreaming process is preparative chromatography. A good separation between the product and other proteins, as well as good productivity, are desired. Due to the batch nature of chromatographic separation, an iterative learning controller (ILC) could be a suitable choice for achieving these goals. ILC based on time-varying perturbation models have been successfully applied to control batch reactors in the past. The purpose of this master thesis was to test the application of time-varying perturbation model-based ILC for automation of preparative chromatography. The application of an ILC was performed using simulations of an ion-exchange chromatographic purification process. Since protein purification by chromatography is commonly performed with gradient elution, the slope of the gradient was chosen as the input parameter for the controller. The slope was controlled via the gradient time, i.e. the time it takes for the elution buffer to go from its initial to its final concentration. First, the resolution between two peaks was used as the output parameter of the controller. The controller was able to successfully reach the desired resolution, however using only the gradient time resulted in a non-linear process trajectory, which resulted in difficulties in process control. Secondly, an objective function was constructed using the resolution and the productivity of the process. The objective function had a local extrema, which was considered the process optimum and thus the derivative of the objective function was used as output parameter. This configuration showed promise, although the estimation of the derivative during live runs was a limiting factor. ILC shows promise for use in preparative chromatography. Multiple-input-multiple-output configurations should be considered for future applications, as such configurations could circumvent the problems caused by the non-linear process trajectory. Alternatively, a different objective function could possibly be applied. A natural next step is to apply the ILC to a real process, thus coming closer to full automation of preparative chromatography.