Identifying toxic substances in water using machine learning — ScienceDaily
Waste from oil sands extraction, stored in tailings ponds, can pose a risk to the natural habitat and nearby communities when it seeps into groundwater and surface ecosystems. Until now, the challenge for the oil sands industry has been that proper analysis of toxic wastes has been difficult to achieve without complex and time-consuming testing. And there is a backlog. For example, in Alberta alone, there are about 1.4 billion cubic meters of fluid tailings, says Nicolás Peleato, assistant professor of civil engineering at the University of British Columbia’s Okanagan campus (UBCO).
His team of researchers from UBCO’s School of Engineering has discovered a new, faster and more reliable way to analyze these samples. It’s the first step, says Dr. Peleato, but the results look promising.
“Current methods require the use of expensive equipment and it can take days or weeks to get results,” he adds. “There is a need for an inexpensive method to monitor these waters more frequently to protect public and aquatic ecosystems.”
With master’s student María Claudia Rincón Remolina, the researchers used fluorescence spectroscopy to quickly detect key toxins in water. They also analyzed the results through a modeling program that accurately predicts the composition of the water.
The composition can be used as a reference for further testing of other samples, Rincón explains. Researchers use a convolutional neural network that processes data in a grid-like topology, such as an image. It’s similar, she says, to the type of modeling used to classify hard-to-identify fingerprints, facial recognition and even self-driving cars.
“The modeling takes into account the variability of the water quality background and can separate hard-to-detect signals, and therefore it can achieve very accurate results,” says Rincón.
The research looked at a mixture of organic compounds that are toxic, including naphthenic acids – which can be found in many petroleum sources. Using high-dimensional fluorescence, researchers can identify most types of organic matter.
“The modeling method searches for key materials and maps the composition of the sample,” says Peleato. “The results of the initial sample analysis are then processed through powerful image processing models to accurately determine comprehensive results.”
Although the results so far are encouraging, Rincón and Dr. Peleato caution that the technique needs to be further evaluated on a larger scale – just how possible it might be to incorporate screening for additional toxins.
Peleato explains that this potential screening tool is the first step, but it has some limitations since not all toxins or naphthenic acids can be detected – only those that are fluorescent. And the technology will need to be scaled up for future more extensive testing.
While it won’t replace current analytical methods that are more precise, Peleato says this approach will allow the oil sands industry to accurately filter and process its waste. This is a necessary step to continue meeting the standards and guidelines of the Canadian Council of Ministers of the Environment.
The search appears in the Hazardous Materials Journaland is funded by the Discovery Grants program of the Natural Sciences and Engineering Research Council of Canada.