Pilot project tests AI to improve water cleanliness

In a town west of Edmonton, experts are trying to figure out if a machine can be taught to “think” like the skilled technicians who help keep Alberta's drinking water clean.

In what is believed to be a firstof-its-kind pilot project, engineers and researchers have set up a mini water treatment machine inside Drayton Valley's larger facility, about 200 kilometres from the capital.

There, armed with artificial intelligence and sensors for everything from water transparency and temperature to Ph levels and conductivity, that small machine has been gathering baseline data since October on the multi-stage process used to clean water.

The goal is, by late spring or early summer, to ask the smart system to find efficiencies in the process and potentially create something that could be used by small communities struggling to have clean water.

The project is being spearheaded by Edmonton's ISL Adapt. Jason Kopan, the company's lead water infrastructure engineer, said professionals running treatment facilities are constantly tweaking the steps involved in cleaning water and the process has become more complicated as water has become more contaminated.

“I used to joke (with effective operators) that I would love to be able to download your brain because there's just certain things that operators get ... that aren't spelled out in any textbook. It just comes with tried and tested experimentation and knowledge,” he said.

The new ISL “brain” is aiming to monitor all the steps in water treatment and track how making a change to one component impacts the others.

Douglas Hallett, research information lead, said this type of “deep reinforcement learning” allows the algorithm to have a memory of sorts, by storing what it has learned to make future recommendations.

“(The system) can both understand the current state, and then more importantly, predict,” he said. “... and look for associations between parameters that human minds wouldn't necessarily understand.”

Kopan said tracking how one step impacts another is good for the environment because operators can be more efficient with chemicals. The same goes for tracking the amount of energy used for pumping, how often a filter needs to be cleaned and what impact those cleaning chemicals have on the filter itself.

Currently, water being tested through the ISL system is discarded and doesn't end up in Drayton Valley's drinking supply.

Even if ISL'S system gets up and running, it won't replace humans. Kopan said the machine would make suggestions for changes to water treatment but a person would make the final call.

If the pilot project, which runs until the end of 2022, is successful, the water system could be used to help out smaller communities.

“The current pilot is portable on a truck. So it can be delivered to a Northern community that has a water crisis, for example,” Kopan said.

It could also influence other sectors.

Martha White, University of Alberta professor and Alberta Machine Intelligence Institute fellow who is leading the research team, said the project is a chance for researchers to see if their theoretical reinforcement learning algorithms work in real life.

If the water tests succeed, the same idea could be used to make industrial plants efficient, she said, and potentially create an AI control industry in Alberta.

“I think that we have a lot of AI expertise, we clearly have a lot of industrial expertise, and we should bring those two expertise together,” she said.