FlocEye: An Artificial Intelligence Based Coagulation and Flocculation Online Smart Control

Introduction.
Physical-chemical separation, using coagulation and flocculation, is a fundamental step for contaminant separation by clarification, flotation and filtration. Although it is a crucial process for many industrial applications, there are currently almost no automatic control options for the dosing of the reagents. In most of the plants, process control is manual, based on visual inspection of the flocs in the flocculation tank, and/or by frequent manual jar tests. Underdosing of the reagents results in poor effluent quality, usually leading to plant shutdown, extensive O&M activities and posing a risk for incompliance with the allowed effluent quality. Overdosing of the chemicals results in high operation costs, excess sludge formation and excess salt content in the effluent. Existing automatic systems for monitoring of the process are limited mainly to surface water treatment, and usually cannot be applied to wastewater treatment. The FlocEye Smart Control, developed by Elad Technologies ltd, is a novel online automatic monitoring and control system for coagulation-flocculation processes. Using a camera installed inside the flocculation chamber, pictures of the wastewater during flocculation are continuously collected and analyzed by an artificial intelligence (AI) algorithm based on an extensive labelled database. The system serves as a 'virtual technician's eye', providing continuous quantitative measurements of the process quality and accurate, automatic dosing of the coagulants and flocculants. Picture A shows one of the prototypes of the FlocEye, installed in an industrial wastewater treatment plant. Picture B shows the first commercial unit of the FlocEye, before installation.
Field Test Description.
This case study relates to a wastewater treatment plant of a PCB factory in Israel, in which the FlocEye was installed in a sulfide precipitation process. The plant utilizes an organic based coagulant containing aluminum (ECOMIX-RE5), an anionic polymer and an organic-sulfide based reagent (Enthol Antiplex). Throughout the first 3 months after installing the FlocEye, the reliability of the system was tested, both for the mechanical aspects (the wiper, the quality of the photos, etc.) and for the accuracy of the machine learning algorithm. Changes were continuously made to the wiper and casing, as well as to the AI algorithms used, to improve performance and robustness. In the following period, during September-November 2021, the FlocEye demonstrated excellent performance, identifying process problems long before effluent quality deteriorates. Consequently, in November-December 2021 tests for controlling the coagulant dosing with the FlocEye were initiated. Objectives. - Demonstrate the FlocEye's ability for early detection of changes in the flocculation process. - Compare the coagulant dosing rate (liter coagulant per m3 of wastewater) when dosing is controlled by the FlocEye to the dosing rate during normal operation. - Compare effluent's turbidity with and without the FlocEye control. Methods. FlocEye was installed in the flocculation chamber and was connected to the plant's PLC. When wastewater is flowing through the continuous treatment, FlocEye receives an indication and continuously sends the PLC the process score (Good/Bad). When controlling coagulant dosing, the FlocEye sends the PLC orders to increase or decrease the dosing rate. Coagulant dosing was continuously adjusted using a simple proportional method, based on changing the dosing time throughout a dosing cycle (XX seconds of dosing, out of YY seconds of cycle). Thus, when working without the FlocEye, the operators maintained a constant dosing rate most of the time, changing it manually according to jar tests or if turbidity increased. When controlled by the FlocEye, dosing rate was increased or decreased continuously by the PLC, according to the process score generated by the FlocEye. Coagulant actual dosing rate was continuously measured by recording the actual working time of the dosing pumps (seconds), divided by the wastewater flow totalizer, as measured with an electromagnetic flowmeter. Twice a day, the actual dosing rate of the pump (ml/sec) was manually verified in the injection point. During the tests, the turbidity of the effluent was continuously monitored by an online turbidity meter. Field Test Results. The field test demonstrated excellent reliability of the FlocEye, with an early detection of almost 100% of the bad process events. False positive events (i.e. FlocEye identifies a bad process, when process is not bad) were frequent during the first few days, but when learned by the AI module, their frequency decreased to less than 10% (i.e 10% of the 'bad process' scores generated by the FlocEye, related to a process quality that was defined in a jar test as a good process). Note that in all of the false positive events, there was indeed a visual change of the process, but separation in a jar test was good.
When FlocEye controlled coagulant dosing, a 36% reduction in coagulant dosing rate was recorded, yet achieving an even higher quality of effluent relative to the reference period. Figure 1 depicts an event, in which the process deteriorates due to a lack of polymer caused by a clogged dosing pump. It demonstrates the changes in process score, during process deterioration and during process improvement. It is shown that the FlocEye immediately identifies the changes in process quality, allowing for a fast reaction. Figure 2 demonstrates the early detection achieved by the FlocEye. FlocEye identifies process deterioration long before the turbidity in the settler increases. Figure 3 depicts the results of testing coagulant dosing control by the FlocEye. It demonstrates a 36% reduction in coagulant dosing rate during 11/29/21-12/2/21, compared to the reference period 11/22/21-11/28/21. Moreover – the turbidity of the effluent during the test period was lower and much more stable than during the reference period. Summary and relevance. The Floc-Eye system presents a novel approach in coagulation & flocculation control using machine learning algorithms. The system continuously monitors the effluent quality, alerts the operator or stops the plant when needed and allows for chemicals' dosing optimization such that chemical costs and sludge quantities are significantly reduced. The field test demonstrated excellent reliability of the FlocEye, resulting in a better effluent quality and a 36% reduction in coagulant consumption. The FlocEye is currently being tested in other flocculation applications, such as Gravity Belt Thickeners, DAF systems and surface water treatment.