
Passive acoustic emissions were detected by a new rheometric device to monitor the manufacture and rheological changes of complex fluids, live and in situ; a simplified output was then transferred to machine learning algorithms. Power-law and Herschel-Bulkley model fluids were studied on the laboratory and pilot scales. Offline rheometry was used to validate the obtained rheological properties.
Abstract
The measurement capabilities of a newly developed in-situ rheometric device based on a single passive acoustic emission sensor and machine learning algorithms were investigated. Two surfactant structured fluids demonstrating complex non-Newtonian rheology (Power-law and Herschel-Bulkley models) were examined. Furthermore, a static evaluation on the laboratory scale in comparison to dynamic processing on the pilot scale was conducted. The results indicate that the machine learning algorithms of this technology can identify, in > 90 % of scenarios, the correct type of rheology or the manufacturing process step across both scales. This identification is based on solving a classification problem using quadratic support vector machine learning algorithms, which have proven to deliver the most robust predictions across a choice of 24 different algorithms tested. Additionally, a new format of in situ rheology display was introduced, referred to as RRF™ factor.