(778d) Effect of Bulk Properties of the Tracer on the Measurement of Residence Time Distributions in Continuous Powder Based Unit Operations

In continuous pharmaceutical manufacturing environment, the development of process mixing models is paramount to characterize the impact of upstream disturbances on the product, establish traceability of raw materials, and assess the process’s performance and robustness. Upstream disturbance dissipation is key to determine whether material will remain is specification given a process deviation. The number of potentially affected products by a disturbance can be quickly quantified based on the disturbance’s dissipation (i.e., back-mixing with in-specification material). Tracing raw material composition allows for product lot-to-lot delineation based on time and lots of raw material used. Process performance and robustness evaluation can allow us to determine potential failure modes and aid in the selection of appropriate process control action to correct deviations.

A commonly used tool in the characterization of mixing systems as well as mixing model development is Residence Time Distribution (RTD). Introduced in the 1950s by Danckwerts, RTD serves as a mean to describe non-ideal mixing in unit operations by characterizing the in-and-out time probability distribution function of a particle entering the unit at steady state [1]. To perform an RTD evaluation, a set of base and tracer materials need to be selected. The base material can be a single or multicomponent ingredient set, which resembles the actual process stream. The tracer is a material that can be quantified at the outlet using visual or chemical methods. Two major methods can be used for introducing a tracer into the system: pulse (or impulse) and step change. The selection of the methodology is very important for both the evaluation and characterization of the mixing behavior. Pulse experiments are effective at keeping the tracer material requirements low, while step changes often require large amounts of material to effectively produce a change in concentration at the system’s outlet. Nonetheless, pulse experiments are more difficult to correctly perform due to the nature of the assumption. Failure to introduce pulse materials evenly and instantaneously may lead to misrepresentations of the system’s mixing behavior [2].

RTD methods have been extensively used in liquid systems, and served the purpose of determining the mixing behavior inside of the unit. Over the last decade, RTD methods have been used to characterize and understand mixing in continuous powder (i.e., granular material) units, such as blenders and feed frames [1, 3, 4]. However, as we study in closer detail how the application of RTD methods compares between liquid and solids systems, we see a clear are of research that requires further evaluation: the selection of a tracer material. Similarly, to the selection of materials for performing unit characterization experiments, tracers are often selected based on a formulation, ad hoc, or based on the availability of materials with easy means of detection (e.g., easy to recognize spectral bands on an Near Infrared Spectra). Although this latter property is a key requirement for tracer materials, it is important to remember that in order to accurately measure the RTD, tracers introduced either through a pulse or step, must not alter the dynamic behavior of the bulk material within the unit operation [5].

Selecting the appropriate tracer requires understanding of the system material properties. Properties of a good tracer include: (1) an ability to be detectable from the bulk, (2) similar physical properties to the process stream, (3) no or minimal adherence to the equipment’s walls or surfaces, (4) can be well mixed in the system, and (5) is representative of the materials involved in the system [1]. Tracers whose material properties are vastly different from the bulk can alter granular dynamics within the unit, resulting in inaccurate measure of the RTD. Similarly, a step change, which causes substantial change in the behavior of the bulk powder within the unit operation, may result in an inaccurate characterization of the RTD. Tracer materials in several published works for the characterization of mixers in pharmaceutical manufacturing include pure active pharmaceutical ingredients (API), which tend to have poor flowing characteristics and adhere to the walls and surfaces of vessels [4, 6-9]. Nevertheless, the selection of API is often the most relevant given the fact that it is the only material whose composition is measured in the product.

This work constitutes the development of a scientific framework to select appropriate tracers (i.e., meet all five criterions) for measuring RTDs. The framework is based on the concept of quantifying disparity between bulk and tracer physical material properties using multivariate analysis tools, such as Principal Component Analysis (PCA) [10]. Using the orthogonal decomposition of PCA, several material property values were collapsed into a reduced set of principal components that express material vast set of properties. Materials were placed on a multi-dimensional principal component space, where their weighted Euclidian distance from each other was measured. The goal of creating this space was to determine the level of similarity (or dissimilarity) between materials. Greater distances between two materials on the multi-dimensional space indicated a greater disparity between material property values and henceforth, the materials themselves. Our work proposes an acceptable multi-dimensional principal component region in which materials that lie within it are considered as acceptable tracers.

Both pulse and step change RTD experiments were performed to assess the effectiveness of our tracer selection based on the multivariate material property strategy. A single bulk material was run through a continuous blender (Gericke GCM-250) at a fixed mass flowrate. The impeller speed, blade arrangement and angle of the exit weir were kept constant. Nine unique materials, at variable distances from the bulk material on the PCA scores plot, and consequently exhibiting variable disparity between its own properties and those of the bulk, were selected as tracers. RTD curves were obtained from each of these nine materials. Each experiment was repeated in triplicate. Statistical methods were used to compare the RTD curves to the ‘true’ RTD of the process. The true RTD was obtained by using dyed bulk material as the tracer. Materials, which resulted in curves that were statistically similar to the true RTD, were labelled as acceptable tracers, and vice versa. An acceptable Euclidian radius was proposed based on these findings. Step changes of increasing magnitude were performed for a binary mixture, and their RTD curves were obtained. The properties of the material post step were characterized and mapped on the PCA space. Step changes resulting in curves which were statistically similar to the true RTD were identified, and their Euclidean distance noted. An acceptable Euclidean radius and thus, an acceptable magnitude of the step change were proposed.