Datenbestand vom 29. November 2022
Tel: 0175 / 9263392
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
aktualisiert am 29. November 2022
978-3-8439-4900-2, Reihe Technische Chemie
Stefan Felix Heisel
Contribution to Quantification of Crystalline Product Quality using Image Analysis
161 Seiten, Dissertation Technische Universität Dortmund (2021), Softcover, A5
For understanding and control of crystallization processes, the characterization of the crystalline product is essential. By default, crystalline products are characterized regarding their particle size distribution which alone cannot be used to make a statement about shape, so imaging methods are necessary. The shape can describe different aspects of crystals, e.g., particle agglomeration, referring to the irreversible fusion of particles. Agglomeration can occur within the entire crystallization process chain and strongly may affect the product’s properties. For an improved understanding of crystallization, it is therefore of importance when and how strongly the system agglomerates.
For a quantification of agglomeration, imaging techniques are coupled with multivariate analysis in order to distinguish single crystals and agglomerates. In order to investigate at what point the model system adipic acid/water begins to agglomerate, an online measurement apparatus was developed. It is shown that loose aggregates overlap the measurement of the solid agglomerates and that after sample dispersion and offline measurement, a significant part of crystals consists of single crystals only.
To differentiate between single crystals and agglomerates, discriminant factor analysis and artificial neural networks are compared. It can be shown that ANNs use fewer variables than DFA to achieve the same or better classification accuracy. Through a targeted selection of crystals of different sizes, the accuracy is increased. From the data obtained, the distribution is separated into its subpopulations of single crystals and agglomerates. However, a disadvantage of the method is that the classification algorithm can only be applied to crystals with a very similar shape. This disadvantage is compensated by using training crystals of different shape. The algorithm generated can correctly differentiate different crystals into single crystals and agglomerates.