Evaluation of AI-method for measuring and characterizing particles on-line in drinking water treatment

Sammanfattning: The municipal association Norrvatten produces drinking water at Görväln WTP for approximately 700,000 residents in 14 member municipalities in the northern Stockholm region. To ensure the drinking water meets quality criteria, the water must be carefully monitored by the WTP. At Norrvatten there are several rapid filters that remove different types of particles, mainly residues from the previous flocculation step. The running time of the filters can be limited by a filter breakthrough, which means that the filter must be regenerated through backwashing. To detect a filter breakthrough, turbidity measurement is used. A new advanced AI method, a particle meter, from the manufacturer Uponor is installed at several locations at Görväln WTP. The particle meter is being evaluated as a possible complement to the standard turbidity measurement. The particle meter is a type of advanced image interpretation software that measures and categorizes particles that may indicate disturbances in the drinking water production.I n this project, particle meters placed on-line in three different rapid filtrates were compared with existing on-line turbidity measurements. The aim was to investigate whether the particle meter could detect a filter breakthrough earlier than a turbidimeter and whether the particle meter added any additional valuable information for drinking water production. Data from the period 1 May 2022–30 April 2023 was evaluated in Acurve and Excel. Periods where filter breakthrough occurred in filters denoted A, B and C were evaluated to see which method indicates a filter breakthrough the fastest. During the studied period, eight filter breakthroughs occurred in filter C (quartz sand), three in filter A (Filtralite NC 0,8-1,6 mm) and none in filter B (Filtralite 70% NC 0,8-1,6 mm and 30% HC 0,5-1 mm. Rapid filter B has a lower flow rate than filter A and C, which contributes to no breakthroughs being found. The particle meter could not detect a filter breakthrough faster than existing turbidimeters. However, total particles correlated with the trend of turbidity between March–June, which could be explained by the higher abundance of B-particles. During the remaining months, turbidity and total particles followed completely different trends. Therefore, the particle meter could potentially be used to detect algal blooms online early during spring, compared to the weekly laboratory analysis of algae. This is also supported by algae in raw water correlating with the trend of B-particles in the rapid filtrate. Periods where the turbidity was below 0,10 FTU and total particles exceeded 100,000 pcs/ml in filters A, B and C were selected to investigate whether the particle meter provided any additional valuable information. Six events for total particles exceeding 100,000 pcs/ml were found in filter C, two in filter A and none in filter B during the examined period. The lower flow rate as well as the material combination in Filter B could contribute to no events being found. The material in Filter A could also contribute to a lower number of events compared to Filter C. The small particle category, 3