Categorization of Sprays by Image Analysis with Convolutional Neuronal Networks. Issue 2 (18th November 2022)

• Record Type: Journal Article
• Authors: Pieloth, Damian; Rodeck, Matthias; Schaldach, Gerhard; Thommes, Markus
• Other Names: Sarangi Prakash Kumar guestEditor.
• Is Part Of: Chemical engineering & technology. Volume 46:Issue 2(2023)
• Page Start: 264
• Page End: 269
• Publication Date: 2022-11-18
• Subjects: Convolutional neural networks -- Droplet size -- Image analysis -- Machine learning -- Spray categorization; Chemical engineering -- Periodicals; Chemical processes -- Periodicals; 660.05
• DOI: 10.1002/ceat.202200356
• Languages: English
• ISSNs: 0930-7516

Abstract

Abstract: Spray characterization has been an issue for process and product characterization for decades. Because of this, a convolutional neuronal network was developed to determine the droplet size from spray images. The images were taken using a digital camera, a light source, and a dark room. These were subsequently employed to design and train a convolutional neuronal network using open‐source software packages and a desktop computer. The accuracy of the network droplet size determinations was checked with additional, independent images. The median drop size was assessed with a high accuracy of more than 99.8 % as the mean spray performance indicator. Additionally, the droplet size distribution measurements from the neural network method deviated from those from the reference method (laser diffraction) by less than 1.5 %. Convolutional neuronal networks can be applied to determine the spray performance using spray cone images. This approach could be useful for multiple applications. Abstract : An optical quality control system, capable of characterizing sprays with a predefined specification, is developed. Categorizing sprays by convolutional neuronal networks is described. The median drop size is determined with a high accuracy. The droplet size distribution estimations made by this method deviated from the measurements from the reference method (laser diffraction) by less than 1.5 %.