(154c) Evaluation of Real Time, Inline Characterization of Bubble Size and Shape Distributions in Highly Concentrated Dispersions

Today many different techniques for sizing transient fluid particle behavior in production vessels are available. Often samples are withdrawn over time, which are later diluted or stabilized, prior to their measurements. These sampling techniques neither guarantee that the drop or bubble sizes are frozen, nor that they are preserved during the sampling and have been criticized by many authors. Especially for technical applications with fast reactions and strong coalescence such kind of measurements are not suitable. Even for sampling times less than one second, the drastic change in the flow condition during sampling results in significant measurement errors. The control of the fluid particle size in such systems is an important goal for method development.

methods

Image-based set-ups are capable of a simultaneous analysis of the size and the shape of the bubbles and they are the number one technology in the industrial as well the academic world. Static and dynamic image analysis have been evaluated by the ISO group to standardize the multiple procedures. The results are given in the ISO13322-1 and ISO13322-2. Image analysis is a technique that has gained popularity in different applications. Two different set-ups have been studied to evaluate the capability of inline imaging connected with an automated image analysis software: a mixing vessel as well as a bubble column. Different image analysis algorithms and their results based on the named experiments are evaluated and categorized based on their advantages as well as disadvantages. These algorithms range from classical computer vision approaches to machine learning and artificial intelligence applications.

results

The influence of the stirrer geometry, the stirrer speed, the air flow rate as well as the air pressure on the bubble size distribution have been studied extensively. The different experiments do give a broad overview on possible process conditions in terms of Reynolds and Weber number. The different algorithms used give a greater understanding of the development and challenges in the application of image analysis in those circumstances. The following figure shows a typical image analysis result using a pattern matching algorithm. Despite the high hit rate and acceptable degree of precision, the image already shows the limitation of this algorithm if the shape of the object of interest varies from circular objects. For thes more challenging analysis task, as the degree of freedom is increasing, machine learning approaches have been applied with astonishing results. The presentation will show these results and the necessary preparations to achieve real time feedback from such a highly concentrated dispersion.

Fig. 1 - Example images of the detection results. The figures show the original image with the detection results (upper image) and the filtered and sharpend image (lower images). The results were obtained for the GDX-impeller at a stirrer speed of 1300 rpm.