De Micco, Maurizio (2022) Deep learning methods for the image-based assessment of the physical stability of liquid formulations. [Tesi di dottorato]
Anteprima |
Testo
DeMicco_Maurizio_35.pdf Download (31MB) | Anteprima |
| Tipologia del documento: | Tesi di dottorato |
|---|---|
| Lingua: | English |
| Titolo: | Deep learning methods for the image-based assessment of the physical stability of liquid formulations |
| Autori: | Autore Email De Micco, Maurizio maurizio.demicco@unina.it |
| Data: | 13 Dicembre 2022 |
| Numero di pagine: | 100 |
| Istituzione: | Università degli Studi di Napoli Federico II |
| Dipartimento: | Ingegneria Chimica, dei Materiali e della Produzione Industrialea |
| Dottorato: | Ingegneria dei prodotti e dei processi industriali |
| Ciclo di dottorato: | 35 |
| Coordinatore del Corso di dottorato: | nome email D'Anna, Andrea anddanna@unina.it |
| Tutor: | nome email Verdoliva, Luisa [non definito] Villone, Massimiliano Maria [non definito] Zonfrilli, Fabio [non definito] |
| Data: | 13 Dicembre 2022 |
| Numero di pagine: | 100 |
| Parole chiave: | deep learning; interpretability; liquid formulation; fabric softener; instability detection |
| Settori scientifico-disciplinari del MIUR: | Area 09 - Ingegneria industriale e dell'informazione > ING-IND/26 - Teoria dello sviluppo dei processi chimici |
| Depositato il: | 23 Dic 2022 11:26 |
| Ultima modifica: | 09 Apr 2025 14:20 |
| URI: | http://www.fedoa.unina.it/id/eprint/14640 |
Abstract
Liquid formulations, such as liquid detergents and fabric softeners, arouse great interest in the industry of everyday products. Such products are characterized by the complexity of their composition, which derives from the presence of several additives to increase the customers’ experience. Since they are dispersions, liquid formulations are thermodinamically unstable. Thus, part of the job of manufacturers is to ensure a minimum period of time over which formulations will resist changes in their physicochemical properties, namely, shelf-life. Instabilities lead to changes in appearance and performance of the products and are not welcome. For this reason, a lot of effort has been invested over the years to provide methods for the stability assessment of liquid formulations. Among the many approaches proposed to this end, solutions based on the visual inspection of samples present numerous advantages: they are cheap, easy to perform, non-intrusive, repeatable. However, conventional image-processing-based detectors do not ensure a sufficient reliability. The aim of this thesis is to propose a novel deep-learning-based method for the stability assessment of liquid formulations. Towards this end, we leverage a dataset comprising a large number of macroscopic images of fabric softeners displaying various types of instabilities in different phases of their development. The dataset is strongly unbalanced, due to the dominance of stable samples. Therefore, we implemented a number of solutions to improve the classifier performance without increasing complexity. Eventually, a deep-learning-based instability detector has been implemented, achieving great results in early detection compared with the classical image analysis methods. Interpretability is provided in order to get insight on the phenomena of interest.
Downloads
Downloads per month over past year
Actions (login required)
 |
Modifica documento |
