Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer

The rapid advancement of AI algorithms presents new opportunities for sensing technologies based on image recognition, such as real-time crystallization monitoring. This work investigates the use of computer vision to detect and size crystals in a lab scale draft tube baffle crystallizer (DTBC). A s...

Full description

Saved in:
Bibliographic Details
Main Authors: Laura Marsollek, Julius Lamprecht, Norbert Kockmann
Format: Article
Language:English
Published: MDPI AG 2024-11-01
Series:Crystals
Subjects:
Online Access:https://www.mdpi.com/2073-4352/14/12/1045
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850241421207928832
author Laura Marsollek
Julius Lamprecht
Norbert Kockmann
author_facet Laura Marsollek
Julius Lamprecht
Norbert Kockmann
author_sort Laura Marsollek
collection DOAJ
description The rapid advancement of AI algorithms presents new opportunities for sensing technologies based on image recognition, such as real-time crystallization monitoring. This work investigates the use of computer vision to detect and size crystals in a lab scale draft tube baffle crystallizer (DTBC). A specially developed analytical bypass was implemented on the DTBC to enable a low-influence analysis without invasive intrusion into the process. By utilizing AI models such as YouOnlyLookOnce version 8 (YOLOv8), YOLOv8 Segmentation (YOLO8seg), and the convolutional network for biomedical image segmentation U-Net, this study assesses their effectiveness in determining crystal size distributions from photometric images. While U-Net was deemed unsuitable due to computational demands and accuracy issues, YOLOv8 and YOLO8seg performed better in terms of efficiency and precision. YOLO8seg, specifically, achieved the highest accuracy, with a mean average precision (mAP) of 82.3%, and excelling in detecting larger crystals, but underperforming with crystals smaller than 90 µm. Despite this limitation, YOLO8seg was able to compete with the manual methods and was superior to the state-of-the-art algorithm mask region convolutional neural network (Mask R-CNN) in terms of accuracy. The study suggests that specific training and adaptation of the imaging conditions could further improve the crystal detection performance.
format Article
id doaj-art-42512691b02f49cbb6425fc9ceaaa11f
institution OA Journals
issn 2073-4352
language English
publishDate 2024-11-01
publisher MDPI AG
record_format Article
series Crystals
spelling doaj-art-42512691b02f49cbb6425fc9ceaaa11f2025-08-20T02:00:37ZengMDPI AGCrystals2073-43522024-11-011412104510.3390/cryst14121045Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle CrystallizerLaura Marsollek0Julius Lamprecht1Norbert Kockmann2Department of Chemical and Biochemical Engineering, Laboratory of Equipment Design, TU Dortmund University, 44227 Dortmund, GermanyDepartment of Chemical and Biochemical Engineering, Laboratory of Equipment Design, TU Dortmund University, 44227 Dortmund, GermanyDepartment of Chemical and Biochemical Engineering, Laboratory of Equipment Design, TU Dortmund University, 44227 Dortmund, GermanyThe rapid advancement of AI algorithms presents new opportunities for sensing technologies based on image recognition, such as real-time crystallization monitoring. This work investigates the use of computer vision to detect and size crystals in a lab scale draft tube baffle crystallizer (DTBC). A specially developed analytical bypass was implemented on the DTBC to enable a low-influence analysis without invasive intrusion into the process. By utilizing AI models such as YouOnlyLookOnce version 8 (YOLOv8), YOLOv8 Segmentation (YOLO8seg), and the convolutional network for biomedical image segmentation U-Net, this study assesses their effectiveness in determining crystal size distributions from photometric images. While U-Net was deemed unsuitable due to computational demands and accuracy issues, YOLOv8 and YOLO8seg performed better in terms of efficiency and precision. YOLO8seg, specifically, achieved the highest accuracy, with a mean average precision (mAP) of 82.3%, and excelling in detecting larger crystals, but underperforming with crystals smaller than 90 µm. Despite this limitation, YOLO8seg was able to compete with the manual methods and was superior to the state-of-the-art algorithm mask region convolutional neural network (Mask R-CNN) in terms of accuracy. The study suggests that specific training and adaptation of the imaging conditions could further improve the crystal detection performance.https://www.mdpi.com/2073-4352/14/12/1045crystallizationcomputer visionYOLOanalyticslab scale equipmentDTBC
spellingShingle Laura Marsollek
Julius Lamprecht
Norbert Kockmann
Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer
Crystals
crystallization
computer vision
YOLO
analytics
lab scale equipment
DTBC
title Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer
title_full Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer
title_fullStr Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer
title_full_unstemmed Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer
title_short Investigation of AI Algorithms for Photometric Online Analysis in a Draft Tube Baffle Crystallizer
title_sort investigation of ai algorithms for photometric online analysis in a draft tube baffle crystallizer
topic crystallization
computer vision
YOLO
analytics
lab scale equipment
DTBC
url https://www.mdpi.com/2073-4352/14/12/1045
work_keys_str_mv AT lauramarsollek investigationofaialgorithmsforphotometriconlineanalysisinadrafttubebafflecrystallizer
AT juliuslamprecht investigationofaialgorithmsforphotometriconlineanalysisinadrafttubebafflecrystallizer
AT norbertkockmann investigationofaialgorithmsforphotometriconlineanalysisinadrafttubebafflecrystallizer