Potential of Deep Learning Methods for Deep Level Particle Characterization in Crystallization

Abstract

Crystalline particle properties, which are defined throughout the crystallization process chain, are strongly tied to the quality of the final product bringing along the need of detailed particle characterization. The most important characteristics are the size, shape and purity, which are influenced by agglomeration. Therefore, a pure size determination is often insufficient and a deep level evaluation regarding agglomerates and primary crystals bound in agglomerates is desirable as basis to increase the quality of crystalline products. We present a promising deep learning approach for particle characterization in crystallization. In an end-to-end fashion, the interactions and processing steps are minimized. Based on instance segmentation, all crystals containing single crystals, agglomerates and primary crystals in agglomerates are detected and classified with pixel-level accuracy. The deep learning approach shows superior performance to previous image analysis methods and reaches a new level of detail. In experimental studies, L-alanine is crystallized from aqueous solution. A detailed description of size and number of all particles including primary crystals is provided and characteristic measures for the level of agglomeration are given. This can lead to a better process understanding and has the potential to serve as cornerstone for kinetic studies.

Agglomeration Kernel Determination by Combining In-Process Image Analysis and Modeling

“…Transfer learning is a method where pretrained weights of different often more general applications are used as the starting point for training. Here, the weights of a similar task are used, where single crystals and agglomerates are differentiated but by a different image acquisition method . The training process itself is performed for 30 000 iterations with a learning rate of 4 × 10 –4 .…”

“…Here, the weights of a similar task are used, where single crystals and agglomerates are differentiated but by a different image acquisition method. 67 The training process itself is performed for 30 000 iterations with a learning rate of 4 × 10 −4 .…”

“…However, a limited twodimensional (2D) frame is available with only a few crystals on an image, along with hidden crystals or crystals on the edge. A further characterization regarding PSD of agglomerates or even sizes of primary crystals, which form an agglomerate, as it was performed in Lins et al, 67 is not possible but can provide important insights into the agglomeration kinetics. Consequently, depending on the application and the level of information desired, the image acquisition and processing must be evaluated.…”

“…The continuous, noninterfering recording of images is performed by an inline image probe. 32 Due to the inferior image quality of inline setups, images are processed by a deep learning approach, and a CNN 66 inspired by a previous work 67 is adapted to the inline captured crystal images. In a first instance, the potential of using deep learning approaches for image processing of in situ-taken images is discussed, before it is used to evaluate the images during the experiments and is applied for parameter estimation.…”

Automated Image Analysis for Retention Determination in Centrifugal Partition Chromatography

“…Therefore, quantitative online monitoring of the retention was hardly performable, and the total number of parameter sets being investigated according to their retention was severely limited [6]. With rapidly evolving image processing technology and the resulting potentials, the goals of the work reported here are, therefore, to reduce the duration of the evaluation on the one hand and to increase the accuracy on the other hand [7]. For this reason, a Matlab-based routine was developed and implemented, automating the analysis of raw data supplied by a high-speed camera [8].…”