Authors: Peyman Aminpour, PhD; Thomas Reynolds, PhD; Sanjay Konagurthu, PhD
Thermo Fisher Scientific

Purpose

Scale-up of blending unit operations from the laboratory scale to commercial scale can be challenging. Use of predictive tools for simulation of blending processes can be valuable in drug product manufacturability. This work provided a mechanistic model based on Discrete Element Method (DEM) for a pharmaceutical blending process. In-silico design of experiments (DOE) were conducted to evaluate process scale-up for V-blenders from 1.2 L to 2.5 L, 10 L and eventually to a commercial size 1250 L blender. The model showed capabilities in predicting the operating ranges a priori to running the actual process by identifying the critical process parameters and performance early in the blending process.

Methods

DEM models were developed based on a Visco-elasto-plastic frictional adhesive contact model. First, an FT4 powder rheometer (FreemanTechnologies) was used to characterize the mechanical properties of the powders in the formulation. The model API was in the form of spray-dried amorphous solid dispersion (ASD) and characterized as a cohesive poor-flowing material with bulk density 0.35 g/m. The mixture of excipients was a non-cohesive material with bulk density 0.5 g/mL. To calibrate the DEM model parameters, a predictive response model based on an artificial neural network (ANN) was trained on the results of more than 250 simulated FT4 powder rheometer studies (Freeman Technologies). The response model was then used to determine the input parameters, which produced the closest match to the experimental measurements (Figure. 1). In the DEM model, critical process parameters including blender design, rotational speed, filling level, and blending time were investigated. The response factor was set as the blend uniformity. The operating space was generated using a DOE in a laboratory-scale V-blender simulation. The developed DEM model was then used to simulate the commercial-scale blending process to evaluate the change of operating space during scale-up.

Results

An in-silico DOE was used to evaluate the blending speeds, blender sizes, and volumetric fill levels. For a fixed set of materials, the parametric space consisted of three blending speeds (15 rpm, 20 rpm, 30 rpm), three blender sizes (1.2 L, 2.5 L, 10 L), and three volumetric fill levels of 30%, 50% and 75%. Figure 2 illustrates the snapshots of the final blend in a 1.2 L V-Blender after 1-min blending with rpm=15 at three different fill levels. A computational micro-scale index, the Coefficient of Blending Performance (CBP), was used to quantify the mixing performance. As the mixing process takes places the CBP provides information on the contact number of similar and dissimilar particles. Therefore, variation in CBP index demonstrates the formation or breakage of undesired agglomerates. Figure 3 shows the CBP across different fill levels in a 1.2 L V-blender rotating at 15 rpm.

Conclusion

A DEM model has been developed as an efficient scale-up strategy for the blending unit operation of a formulation containing a spray-dried dispersion of a model API. For the development of the model, input parameters were defined using a calibration approach. The validated DEM model was used to simulate the blending process to evaluate the change of operating space during scale-up. Critical process parameters were identified by statistical studies and showed dependence on blender size, fill level and the rotation rate. DEM can be applied as a predictive approach to modeling and scale-up of pharmaceutical blending unit operations.

Figure 1. Figure 1. DEM simulation of the FT4 powder rheometer (left). Prediction accuracy of the ANN model (middle). Structure of the single-layer ANN model with 15 hidden nodes and TanH activation functions (right).

Figure 2. Snapshots of the final blends in a 1.2 L V-Blender after 1-min blending with RPM=15 at three different fill levels (top). Concentration of API, with red representing the highest concentration (bottom).

Figure 3. CBP vs. number of revolutions across different fill levels in a 1.2 L V-blender with RPM=15.