The rapid rise of artificial intelligence has people in many professions wondering if they will be automated out of their jobs.
Many who work in occupations once assumed safe against offshoring and robots — computer programming, accounting, advertising, graphic design, engineering, magazine writing — have looked over their shoulders at what AI might bring.
AI is certainly coming to water and wastewater treatment, but many experts believe operators have nothing to fear from it and much to gain. One of those is Martha White, an associate professor of computing science and director of the Reinforcement Learning and Artificial Intelligence Laboratory at the University of Alberta.
White is also a co-founder and CEO of RL Core Technologies, a company pursuing advances in water treatment through AI-driven real-time optimization of processes like chemical dosing and filtration. The company uses reinforcement learning to enhance existing automation systems without costly overhauls.
RL is a branch of AI in which software makes decisions to improve performance while interacting with its environment. It constantly adapts low-level setpoints and therefore frees operators to focus on higher-level goals. White believes this elevates the stature of operators and makes them more valuable in their roles. She talked about RL and its relationship to clean-water professions in an interview with Treatment Plant Operator.
TPO: In general, what benefits do AI and machine learning offer to the drinking water and wastewater treatment industries?
White: Its primary role will be to improve the automation systems operators use today, thereby making their overall functionality better. It will help optimize processes, enabling treatment plants to use less chemicals, consume less energy and potentially alert operators to any issues. Today’s facilities have more sensors than ever collecting more data, and we’re not necessarily using that data as effectively as we could. Bringing data-driven methods to water and wastewater treatment will only be beneficial for operators. It’s going to give them more tools.
TPO: How would you define RL for operators not familiar with AI and related technologies?
White: Operators could think of RL as a data-driven approach to real-time optimization. It uses as much sensor and historical information as it can to understand how the system behaves. Then in real time, it monitors all the sensors and adapts the setpoints based on the outcomes of different choices. It’s a bit like what an operator ordinarily would do. An operator might revise a setpoint — guess at and test a few values to see if, for example, increasing a chemical dosage improves performance after a few hours. RL is about trying things to determine how to make the system do better and better over time.
TPO: How would you differentiate RL from other offerings in the AI sphere?
White: Let me first explain how it differs from machine learning. Most machine learning in water treatment focuses on data analysis. Typically, it collects and analyzes historical data and makes predictions about future trends so that operators can use that information to improve performance. And it can provide alerts when something is behaving differently than expected. On the other hand, RL is about improving operations through feedback and seeing in real time the outcomes of different setpoint choices. So instead of giving operators support tools, it improves system operation directly.
TPO: How would operators use RL in their daily lives?
White: It doesn’t change the operator’s role that much. Mostly, it helps them operationally in that it can enable the plant to save on costs. Operators have a lot on their plate. They have to maintain the whole facility and make sure it is running well. They don’t necessarily have time to manually change setpoints for all the systems to try to get those operational gains. RL gives them cost reductions by more frequently changing the setpoints on their behalf so they can continue to focus on overall plant performance.
TPO: What is an example of how RL was applied to improve a specific process?
White: EPCOR Utilities in Edmonton, Alberta, used RL to optimize chemical feed to a hydrogen sulfide scrubber. Previously, setpoints were set at values that were rarely changed because the system performed adequately, though not optimally; there were more important things for the operators to do. Now RL operates behind the scenes to optimize the chemical feed while keeping the hydrogen sulfide removal rate as high as possible. So they perform better in terms of compliance without significantly changing what the operators do day to day.
TPO: Is there an example where the operators were more directly involved with the RL function?
White: In the town of Drayton Valley, Alberta, we looked at their ultrafiltration system for drinking water. They do daily jar tests to decide what level of coagulant they should be dosing. Now they still do their jar test in the morning, and then they put into our system where they want the coagulant dosing set. For the rest of the day, RL adjusts around that value within a range of what the operators specified. So it optimizes chemical usage, but keeps the UV transmittance of the outlet water within the parameters the operators asked for. Again, it helps the system function more efficiently without much change in what the operators do.
TPO: What would you say to operators who might worry that technologies like AI and RL will make their jobs obsolete or less rewarding, or make them subservient?
White: There is very little chance it will replace people. Operators need to have a holistic view of the plant, and RL is really just an operator assistant, helping them achieve their goals and make sure the plant operates better. At most I would say it shifts their decision-making to a higher level. So instead of having to do jar tests every day to set coagulant dosing, our system can help them by setting coagulant dosing automatically. When you enable operators to do less low-level, mundane work and focus on higher-level decisions, that makes them more valuable.
TPO: What reactions have you seen from operators who have used this technology?
White: Operators we’ve worked with have been very confident and innovative. They know how their system works, and they have a reasonable comfort level with deploying real-time optimization because they themselves might try different values as a matter of course. We’ve seen more caution where operators have had a very active hand; they might be changing setpoints every hour. They want to know: Can your system actually do what I’m doing here?
TPO: How specifically would RL be deployed in a treatment plant?
White: We would talk to the plant’s process experts to understand the setpoints they want to adjust to improve performance. We would then work with those experts to identify the highest-value processes to optimize, and also to identify the guardrails. The next step is to deploy the full solution. Software is installed to a computer on site that connects to the existing SCADA system. A local process engineer then adjusts the SCADA system so that operators can toggle between their standard operation and the mode of operation that uses the RL solution. Often, we’ll also ask for historical data, if it’s available, so the RL can start by understanding as much about the system as possible before it actually starts doing direct control.
TPO: Is there a requirement for additional sensors or instrumentation?
White: Generally speaking, no. The goal is to take the sensors and data that exist and strive to improve operations from there. However, in some cases, adding sensors to help optimize a specific additional process is beneficial.
TPO: How would you characterize the importance of RL and related technologies to the water and wastewater treatment sector?
White: There is a growing focus on environmental sustainability, and we’re seeing higher restrictions on outlet water quality. I believe that’s going to require more complex systems, and that’s where we start needing smarter automation approaches. Even if for now AI isn’t strictly necessary, in the near future it will become more and more useful to have technologies that help these complex systems run more efficiently.
