(658e) Methodology and Pitfalls When Calibrating a PBM: The Case of Twin-Screw Wet Granulation
First, the model grid needs to be defined (i.e. number and location of size classes). Different measurement techniques (laser diffraction, QICPIC, sieve analysis) use different grids. As each measurement technique has its own peculiarities, it is difficult to compare measurements performed with different techniques. Second, an objective function needs to be defined. The current practice to calibrate PBM is similar to how this was done for time series and using deterministic models. Sum of Squared Errors (SSE) and Root Mean Squared Error (RMSE) are thus typical gold standards. Similarly, the information of a whole particle size distribution can be condensed into some characteristic numbers, such as mode, mean, span, and the Sauter diameter. Proper evaluation is needed to confirm whether this is indeed the way to go to obtain a good modelling practise for PBM. A lot of freedom is available when dealing with particle size distribution, but the question is how to deal with this kind of freedom and how it is best coupled to the modelling objective. Third, a technique to find the minimal value of the objective function has to be selected. In this study, predictions of a selected PBM formulation are generated using a global parameter space exploration by means of a large set of Monte Carlo simulations. Changing the aggregation and breakage kernel parameters yields different size distributions. Different objective functions are evaluated to determine the objective function whose optimum yields the optimal agreement between the simulated and measured particles bearing in mind the objective. This optimal agreement, within a predefined error range, is called the calibrated model for that process setting. This study is repeated for different process settings of the twin-screw granulator. By comparing the calibrated results for different process settings, the aggregation and breakage mechanisms can be identified, and the most dominant regimes can be found. Guidance on the different choices to be made during the calibration process will be provided.