Sewage is delivered to wastewater treatment plants (WWTPs) where benign microbial organisms within ‘activated sludge’ vessels contribute to the removal of harmful pathogens from the sewage. The activity and pathogen-removing ability of these helpful organisms is affected by many factors including temperature, numbers of fine particles, pH, ammonia, and the time available to remove the pathogens. Regulatory authorities require at least 90% (one log removal value, LRV) of the pathogens to be removed, but as WWTP operating conditions vary, the LRVs change. This problem led to recognition of the need to develop models capable of predicting relationships between plant operating parameters (such as temperature) and pathogen removal. This research reviewed published reports and datasets, then set up and ran an experimental activated sludge pilot plant to generate data about a range of operating conditions and pathogen removals. These datasets were used to develop models which had only a ‘poor’ predictive value for clostridia but were ‘good’ for giardia and ‘very good to excellent’ for the removal of other pathogens. These models need to be extended with more operating conditions but have the potential to be used to attribute LRVs and for future integration into online real-time monitoring.

Although water utilities recognise the value of online instruments that provide real-time monitoring capability, there are problems with visualising and interpreting datasets, and with distinguishing between data resulting from real-world changes in treatment plant operating conditions, for example changed turbidity or flow, and instrument failure. There are also challenges around instrument installation and operation. This project developed tools to support data visualisation and interpretation by building a prototype visualisation platform for analysing complex online UV spectral data in conjunction with weather and lab data (see Factsheet 2 ‘Development of an online platform’). To improve differentiation between instrument failure and real-world data a Bayesian Belief Network model was developed to analyse patterns and variations within datasets. Real operational, high turbidity data was used to demonstrate that this model could accurately identify different causes for the readings which included filter ripening, backwash and other causes (see Factsheet 3 ‘Improving decision making in water plant operability through Bayesian Belief networks’). Strategies for instrument installation and operation were illustrated through case studies.