Image analysis in coagulation process control

Abstract

Coagulation-flocculation is a conventional process in drinking water, municipal and industrial wastewater treatment. This stage of wastewater treatment has become popular in Norway due to its superior ability to remove particles and phosphates. However, a conceptual model of the process still does not exist in the sense of an established and universally accepted model, due to the complex nature of the system. The treatment efficiency and economics of the process are mainly based on chemical (coagulants and flocculants) consumption, thereby emphasising the need for optimal coagulant dosing; in addition to reaching the required treatment efficiencies, optimal coagulant dosing reduces the operational costs related to chemicals and waste management as a result of reduced amounts of sludge. Optimisation of the process highly depends on the dosage control concept. A vast majority of treatment plants operate the process with a flow-proportional dosage concept, at best combined with pH, despite the fact that the optimal coagulant dosage highly depends on particles and phosphates. Innovative concepts that include these parameters, directly or indirectly, have reported savings of up to 30 % of coagulants. Despite the potential savings, the limited use of such systems may be explained partly due to the lack of a conceptual model and partly due to high costs of appropriate multi-parameter dosage control strategies. Additionally, not all the desired parameters can be measured affordably online, nor can they be used in feed-back control strategies due to the long time lag between the inlet and outlet measurements. Thus, sensors that can be placed inside the flocculation chamber to evaluate the process in situ would significantly increase the efficiency of the existing feed-forward systems. This would help to improve the dosage prediction models, as well as decrease the number of expensive and complicated sensors in existing multi-parameter based systems. The author’s PhD work concentrated on solving these issues. Such sensor concepts also contribute to the prediction of outlet qualities, which is another challenge in the water industry. Thus, the objective of this PhD work was to address the above challenges and needs by developing a low-cost sensor prototype based on image analysis of the flocs – particles aggregating during the coagulation-flocculation process. Different image analysis techniques were evaluated including conventional particle recognition methods and texture analysis methods, which are broadly used in other fields than water treatment. Texture image analysis methods were found to be a successful solution to challenges associated with wastewater flocs. The concept of characterising flocs images by texture image analysis techniques was first tested in the laboratory scale batch process (jar tests) with model wastewater. The non-intrusive image acquisition system was established to capture images of flocs during the slow mixing stage of the coagulation process. It was proven that the images of flocs have distinct texture features correlating to the coagulation conditions (type and amount of chemicals, time after the start of a slow mixing phase, etc.) and inlet wastewater parameters. The changes in flocs images, coagulant dosages and treatment efficiencies were studied. The correlations between textural features of the flocs images, coagulant dosages and treatment efficiencies were found using multivariate statistics.

After successful laboratory studies, the full-scale experiments were conducted in Skiphelle municipal wastewater treatment plant (Drøbak, Norway). The texture image analysis concept was proven to be applicable for real municipal wastewater flocs. The changes in inlet wastewater parameters and coagulation conditions were traceable with the images of flocs. The system was proven to predict the outlet turbidity values, which can potentially be used for troubleshooting as an early indication of coagulation failure. The low-cost floc sensor prototype, consisting of a single-board computer and camera module, was developed and tested at the same municipal wastewater treatment plant. Customised software was written to control the camera and adjust settings. The investigations documented that the images of flocs captured by the low-cost camera module could be used for optimal coagulant dosage predictions. Overall, the results of this PhD work confirmed the potential for the floc sensor to be a stand-alone online digital image analysis device that could increase the accuracy and affordability of the existing multi-parameter based coagulant dosage control systems.

Koagulering-flokkulering, også kjent som kjemisk felling, er en konvensjonell prosess i behandling av drikkevann og kommunalt og industrielt avløpsvann. Prosessen er blitt svært populær for avløpsbehandling i Norge på grunn av høy rensegrad for partikler og fosfor. Imidlertid finnes det fortsatt ingen etablert og universelt akseptert konseptuell modell for prosessen, noe som må tilskrives prosessens iboende kompleksitet. Renseeffektiviteten og økonomien i prosessen er i hovedsak kontrollert av kjemikalieforbruket (koagulanter og flokkulanter). Dette understreker behovet for optimal dosering av kjemikalier, som i tillegg til å sørge for at rensekravene overholdes, vil redusere driftskostnadene knyttet til kjemikalier og slambehandling pga. mindre slamproduksjon. Optimering av prosessen er nært knyttet til styring og regulering av kjemikaliedoseringen. De aller fleste avløpsrenseanlegg benytter i dag mengde-proporsjonal kjemikaliedosering, i beste fall kombinert med pH overstyring, til tross for at den optimale doseringen er sterkt avhengig av vannets innhold av partikler og fosfor. Innovative tilnærminger som inkluderer disse parameterne direkte eller indirekte i styringen av doseringen, har kunnet vise til en reduksjon kjemikalieforbruket på opptil 30 %.