Impact of agglomeration on crystalline product quality within the crystallization process chain

Abstract

The quality of crystalline products, defined by e.g. purity or crystal size distribution (CSD), is primarily dominated by crystallization conditions but influenced by further downstream processes like solid-liquid separation and drying also. Through uncontrolled agglomeration within the crystallization process chain the purity or CSD can be negatively affected. Therefore, in context of process optimization, missing knowledge of the impacts on the final product can lead to product batches out of specification. To increase the understanding of agglomeration and to provide insight into the relevance of holistic process optimization the agglomeration behavior of L-alanine crystals is exemplarily quantified over the crystalline process chain. For the quantification the agglomeration degree (Ag) and the agglomeration degree distribution (AgD) are determined. The results show that the product quality achieved after crystallization is significantly affected by agglomeration during drying. Especially if washing after solid-liquid separation is omitted, a broadening of the CSD is observed. Moreover, the evaluation by the AgD indicates that the final product can be -despite similar characteristics of the CSD -highly different. Consequently, it can be concluded that the characterization of the product quality by the CSD alone is insufficient and the quantification of agglomeration is essential for process optimization.

Growth Rate Measurements of Organic Crystals in a Cone‐Shaped Fluidized‐Bed Cell

“…The wash liquid consisted of a mixture of ethanol and ultrapure water with a volume ratio of 4:1 in the first wash cycle and pure ethanol (Merck KGaA, absolute EMPLURA®) in the second cycle. This washing procedure was adopted from Terdenge et al . …”

Discrimination between Single Crystals and Agglomerates during the Crystallization Process

“…The QICPIC, equipped with a CMOS camera (resolution: 1024 × 1024 pixel, pixel density: 10 µm/pixel, frame rate used: 12.5 Hz) and the optical measurement module M6 (magnification 2:1, Sympatec), is capable of detecting particles > 5 µm. Due to the limited resolution though, the analyzation limit of 80 µm was established and successfully utilized in previous works , . Because using M6 does not allow to investigate the early process stages where the majority of particles is smaller 80 µm, the optical measurement module M4 (magnification 5:1, Sympatec) was added to the QICPIC, effectively decreasing the analyzation limit to 32 µm (Fig.…”

“…In previous works , , , the classifier generated was employed to classify final crystalline product batches. For classification of data where the crystal sizes change significantly during the whole crystallization run, it needs to be ensured that the classifier's accuracy is similar for all j size fractions of the PSD.…”

“…Up to this point, several authors , , used discriminant factorial analysis for particle classification while Heisel et al found that artificial neural networks (ANNs) are a more promising way for particle classification in direct comparison to discriminant factorial analysis. Common to all these authors, the accuracy of the generated separation function (classifier) has not been investigated with respect to size dependency of the classified particles, even though some authors mentioned to specifically use dimensionless image descriptors only , .…”

Crystal Morphology Monitoring based on In-situ Image Analysis of L-glutamic Acid Crystallization

“…The misidentification of these agglomerates into single crystals directly affects the distribution of crystal size and shape. L. M. Terdenge, A. Ferreira, S. Heisel, L. M. Terdenge, D. R. Ochsenbein used Discrimination Factorial Analysis (DFA) and Artificial Neural Networks (ANN) to divide agglomerated particles and single crystals into training sets and testing sets and conducted a variety of dataset crossover experiments, defined PI and Ag test indicators, the experiment proved that the recognition accuracy can reach 93% [17][18][19][20][21]. Y. Huo monitored particle agglomerations during crystallization by using microscopic double-view image analysis [22].…”