Most of us never think about what happens after water spirals down a drain. But for the people who run wastewater treatment plants, keeping the promise of clean water is a race against time \u2013 and data.<\/p>\n
Treatment plant operators face a major challenge: critical water-quality tests take days to complete, while operators often need to make decisions within hours.<\/p>\n
A growing field of artificial intelligence \u2013 explainable AI, or XAI \u2013 is helping bridge that gap, said Fuad Nasir, a water supply specialist with the Wisconsin Department of Natural Resources.<\/p>\n
During his graduate studies, Nasir (\u201925 PhD Civil Engineering) delved into research involving data collection from a local treatment plant. And he noticed something important happening in the broader world of technology: \u201cAI and machine learning were becoming used in many sectors,\u201d he said, \u201cbut they were not using AI in wastewater treatment very much in the United States.\u201d<\/p>\n
The problem wasn\u2019t lack of interest \u2013 it was lack of trust. Traditional machine learning has been compared to a black box: It takes in data and produces predictions.<\/p>\n
\u201cOperators were hesitant to use it because they couldn\u2019t see the process for how it arrived at its prediction,\u201d Nasir said.<\/p>\n
For wastewater operators making high-stakes decisions about chemical dosing and treatment timelines, that opacity was a deal-breaker.<\/p>\n
That led him to explainable AI. XAI doesn\u2019t just predict an outcome; it highlights the variables that shaped that prediction. \u201cIt reveals what\u2019s going on in the background,\u201d Nasir said. \u201cYou can literally visualize it when you apply XAI.\u201d<\/p>\n
A key example illustrates why this matters. Wastewater plants rely on a lab test called biochemical oxygen demand (BOD) to gauge how much organic material remains in treated water. High BOD can harm aquatic life, so operators need to adjust treatment chemicals to keep it low. But BOD testing takes five days, too long for operators to make decisions.<\/p>\n
XAI-guided models can predict BOD using historic data, showing which factors \u2013 such as temperature, flow rate or ammonia \u2013 are driving the levels up or down.<\/p>\n
XAI has grown rapidly. The medical field adopted explainable AI earlier, but wastewater research is now catching up and citing Nasir\u2019s work in the process.<\/p>\n
\u201cI\u2019ve seen a surge of people using it in wastewater in the last few years,\u201d he said.<\/p>\n
Real-world deployment will take time, he admitted, because utilities need instruments, training and funding, but the direction is clear.<\/p>\n
Nasir arrived at 51ÁÔÆæ in 2019 as a master\u2019s student from Bangladesh, attracted by 51ÁÔÆæ\u2019s strong research culture and environmental engineering faculty, including Professor Jin Li, his advisor.<\/p>\n