Cleaner water through fluorescence spectroscopy and artificial intelligence

U.S. Army Reserves Staff Sergeant Chris Miller travelled to Honduras, Guatemala, and Panama while an engineering student. He was part of an Army engineering construction unit. The group did community building, including carpentry, electrical work and even putting up buildings.

“Maybe they needed something, and they’d call it a school,” Miller said. “But you often wondered. We knew it was a school, but it could be used for something else in the future.”

Miller earned his Ph.D. from the University of Iowa during his service. After 13 years in the reserves, he left service while pursuing tenure as an associate professor of Civil Engineering at the University of Akron in Ohio.

Miller focused his research on drinking water quality monitoring and management, along with treatment optimization. It was an interesting choice, since his formal training was principally experimental in a laboratory. But in his new role, he wanted to approach the drinking water quality equation in a novel way.

Miller’s work occupies two spaces. One is making greater use of fluorescence spectroscopy to measure more variables in the water before and after purification. The other is applying known chemistry simulations that could provide additional insight for water utilities.

His approach offers utilities insights into treating drinking water.

Water treatment plants do not control the water quality of their source water. Rain events, temperature shifts and unique groundwater features create a lot of variability.

“For water quality, the cards they’re dealt is what they get,” he said. “It’s not like making oil and gas at an oil refinery, where they can control the raw materials. You have raw materials that you don't have any control over.”

Temperature, turbidity, pH, alkalinity, hardness, and dissolved organic materials affect water quality. These variables come in different combinations and change over time. Treatment plants have to make treatment decisions on the back end and determine if they met multiple water quality targets that are chemistry driven.

The water sources are also changing due to climate.

“We get more intense rainstorms, so you get more turbidity (murkiness) coming in. You have different alkalinity now and the ability to manage these complexities is stressing the various plants’ technical capacities,” he said.

There are, however, chemicals and treatment processes operators can control, with multiple outputs and multiple objectives. Within limits, that is. For example, the objectives in the regulatory environment alone are increasing.

In treating water, Miller said some items could be in conflict. More treatment chemicals don’t always achieve all the objectives. Sometimes you can actually improve one condition and make the other worse.

“That's the engineer in me,” he said. “If you make this one 5 percent worse, which may be acceptable, we'll get 20 percent better in another one. I’m just helping (the treatment plants) quantify those impacts and decisions. It was pretty much where the engineering piece comes in.”

Miller is trying to help close the knowledge gap. How? By having more specific monitoring techniques. One is his use of HORIBA’s Aqualog. He said it gives him better information to make those decisions.

The Aqualog is a spectrofluorometer that simultaneously measures both absorbance spectra and fluorescence Excitation-Emission Matrices.

“The Aqualog provides insights that water treatment plants normally don't get about the dissolved organic components of what it must treat."

Scientists at the plants are monitoring what's coming into the plant and monitoring the core treatment process efficiency. Better insights and outcomes result from more specific information.

Miller has spent his career building databases of water composition from various sources. He uses the Aqualog to build profiles of these resources.

“I have fingerprints from Lake Erie, from Nova Scotia, from the Ohio River, and from inland reservoirs,” he said.

About 12 years ago, Miller wanted to begin collecting full scan fluorescence at these water treatment plants of the water coming into the plant and after their core treatment. Then, he thought, he’d figure out everything else later.

So he began building a database.

“I'm pretty confident both in the quantity and diversity of my data. I have one of the largest databases in the world for drinking water.”

Miller is making use of the conventional data, including all the research that's out there, the known chemistry, to make calculations that water utilities are not going to actually perform, because they lack the technical training, he said.

And he’s using that data for machine learning purposes to provide decision support.

Besides teaching at the University of Akron, Miller is Founder and CEO of Fontus Blue, Inc. The company is a technical resource for people seeking solutions for improving drinking water quality. The company provides specialized technical support via its Decision Blue® software platform for advanced water quality data monitoring, analysis, and treatment optimization.

Decision Blue® addresses core water quality compliance and treatment concerns. It’s designed to optimize treatment options and costs for water treatment plants. The software takes into account water quality forecasting, chemical dosing support, utility operations and regulatory compliance.

“Regardless of the complexity of the model, it grows and gets smarter over time, along with how you make decisions off of the models,” he said.  “The software also continues to support new regulatory challenges including the recently released (United States Environmental Protection Agency’s) proposed Lead and Copper Rule (LCR), which includes a suite of actions to reduce lead exposure in drinking water.

The machine learning begins with the source water quality, water treatment plant and chemicals added to treat the water. The software is learning what those interactions and combinations are, to predict what will come out at the end.

“It’s kind of like autonomous vehicles,” Miller said. “It has to account for speed and the things it observes. And those things are changing all the time. It’s predictive modeling.”

Miller continues to build on his model.

“I would put in enhanced monitoring, new sensors, and new analytical tools to gain better insights and better outcomes,” he said. “With those additional insights, you still have to build new machine learning models to incorporate that input into your prediction. If nobody is actually taking a full scans of fluorescent samples, you can't build the model for that plant because you don't have the experimental data to say, here's what's happening.”

Follow Us