Taking the ‘fingerprint’ of complex water mixtures

The water sector urgently needs reliable, real-time, and cost-effective detection of a wide range of substances at low concentrations, so that timely mitigation measures can be taken when necessary. In the TKI project Development of a Water Quality Sensor, Electrochemical Impedance Spectroscopy (EIS) was investigated as a potential monitoring tool for saturated hydrocarbons and inorganic substances.

New application

EIS technology itself is not new—it is, for example, already used in studies to examine material degradation. However, its application as a monitoring tool for measuring certain types of pollution in wastewater is novel, explains Vermeij, Chief Technology Officer at Hypersoniq. “Our startup is a spin-off from the Dutch sensor company TWTG, which was also a partner in the project. TWTG initiated this new direction for the EIS technology. When it became clear that the sensor still required significant research, KWR and TU Delft got involved. Hypersoniq was then established, and soon afterward this TKI project began. This allowed us to deepen the technical development of the sensor and explore its business case and potential industry applications.”

Heavy metals

The principle of EIS is based on measuring electrical resistance in water as an indicator of its composition. Vermeij explains: “The sensor has electrodes made of conductive material. By applying an electrical current through a water sample at various frequencies and measuring the resistance, we gain insights into the substances present. We’ve successfully determined what properties the electrodes need to detect inorganic pollutants like heavy metals. Organic compounds, however, show little or no interaction with the electrode materials we’ve tested so far. They don’t show up in the signals. This means the electrode surfaces need further treatment to develop that selectivity. The technology isn’t there yet.”

Machine learning

Before the EIS sensor can detect water contaminants, it must be trained using machine learning models. KWR researcher Patrick Bauerlein explains how: “It works similarly to how a computer is trained to recognize images of cats or dogs. Once trained with enough examples, the computer learns to recognize them independently. In our case, the computer was trained to interpret data from known water mixture samples. We had already gained experience with machine learning through microplastics monitoring. For heavy metals, we had to visualize the data a bit differently, but when the model finally works—and the computer recognizes substances in the mixture on its own—you just go: wow! That’s when true monitoring becomes possible. We published an explanation of these machine learning models in the prestigious journal Nature Communications, along with findings on how the electrodes function for water quality monitoring. A great result from the project.”

Drinking water sector

Although the project focused primarily on wastewater quality measurements, the drinking water sector was also involved. “At KWR, we simulated a drinking water distribution system at laboratory scale,” says Bauerlein. “We tested the EIS sensor in this setup for 24 hours and analyzed the data. While the processing takes some time and is not fully real-time, it’s still much faster than collecting samples and sending them to a remote lab. The sensor’s ability to detect heavy metals like arsenic could be of interest to drinking water companies with joint ventures in countries where this is a concern. That opens up international opportunities. KWR will present the EIS sensor findings at upcoming meetings of the Chemical Safety theme group within Waterwijs—the collaborative research program of the drinking water companies—to gauge interest in this sensor technology.”

Unexpected turn

Beyond the advances in machine learning and monitoring of inorganic substances, the project took an unexpected turn for Hypersoniq, says Vermeij. “The core technology—the sensor and measurement principle—is not yet developed enough to bring what we’ve learned into practice. One major drawback is that the electrodes quickly become contaminated when placed in real wastewater. Pollutants stick to them. We realized that solving this within our startup’s financial timeframe wasn’t feasible. However, since we had gained experience in data analysis and model development through the project, we saw an opportunity in using data from commercial sensors that clients were already collecting but not using optimally. We focused on turbidity sensors used in Dissolved Air Flotation—commonly used in the food industry. In this process, tiny air bubbles attach to contaminants in water, causing them to float and be removed more easily. With the turbidity sensor data, we can develop models for clients to optimize chemical dosing, improve wastewater purification, reduce waste streams like sludge, and positively impact the environment. For one client, our models already cut chemical use by more than half. This marks the beginning of our commercial rollout.”

New focus

Essentially, Hypersoniq changed its business focus during the project. “We transitioned from a hardware startup to a software startup,” says Vermeij. “We now collect, visualize, and analyze data from water treatment plants and develop models to optimize them. We remain very interested in the further development of EIS technology—TU Delft is continuing that work. We’re especially grateful to KWR for helping us launch our data science team. Without their support, this transition wouldn’t have been possible. They also opened their network to us, which helped us gain insights into the world of sensors, regulation, and key industrial players. That’s the strength of TKI—you become part of an ecosystem that gives you access to people, resources, and knowledge you’d never gather on your own.”

Project partners

The project ‘Development of a water quality sensor using Electrochemical Impedance Spectroscopy for measuring (micro)pollutants’ was carried out with the following partners: Hypersoniq, KWR, TWTG, Vopak Ventures, and with support from TU Delft.