UC Irvine

Online continuous monitoring of water quality variables is a crucial aspect to ensure the correct performance in Water Resource Recovery Facilities. In the last few decades, with the advent of water reclamation and the use of new technologies to lower the energy consumption of processes, real-time measurements have gained a larger role in both process monitoring and control. In order to obtain on-line data, sensors and analyzers are essential, which technology in time have been improved to obtain reliable high-frequency measurements, and to optimize maintenance and costs of operation.

Despite the improvements and the development of new technologies for measuring different water quality, sensors are often expensive and usually present problems of fouling, poor calibration, and lack of reliability. A common perception is that sensors represent the bottleneck for implementing online process control of wastewater treatment, especially in consideration of the harsh environment they are deployed and the number of water professionals sufficiently confident to rely entirely on them.

In this dissertation we focused on ammonium sensors, typically based on Ion-Selective-Electrode technology, which are often found to incur into faulty behavior compared to other water quality sensors, but they have a large potential application for different monitoring and control systems. In this study we analyzed the field limitations of ISE-NH4+ sensors installed in different sections of the activated sludge tank at a full-scale treatment plant. Among the different field influencing factors, a special focus was given to the development of fouling on the sensor’s membrane, both in terms of composition of the film layer and effect on the functional behavior of the sensor when subject to fouling. The results show that the sensors presented a reduced accuracy when working in the lower ammonium range, due to the larger interference of cations like K+ and Na+, which exposes a limit to the possible deployment of these sensors for effluent monitoring. The fouling development was found to be increasing as a negative exponential against time, with a high content of Iron in the inert fraction. The sensor was affected by the fouling development, showing a negative drift in time, therefore leading to an underestimation of the ammonium reading.

In consideration of the demonstrated low reliability of such sensors, especially in effluent installations and concurrently the high request for the use of ammonium sensors for effluent measurement, this study aimed at improving the sensor’s accuracy and reliability for low range installations. A machine learning methodology, based on the use of Artificial Neural Networks and Principal Component Analysis, was employed to develop a fault detection chart to promptly detect the occurrence drifts, biases and in general sensor faults. In detail, the fault detection methodology shows promising results, requiring only few hours to detect an incorrect calibration procedure and one day to detect a fouling drift of the sensor.