Streams of Data
An ML classifier running on the Particle Boron detects water contaminants more rapidly, and less expensively, than existing methods.
From drinking to food production and domestic use, water serves to maintain our physical health and wellbeing. But when this most important of all resources is tainted, it can be transformed into a silent killer. According to the United Nations, water-related diseases are responsible for a staggering 80 percent of all illnesses and deaths in the developing world. These diseases include some very serious conditions like cholera, dysentery, typhoid, and polio. One particularly prevalent problem with drinking water in many locales is E. coli contamination which typically results from the presence of fecal matter. Up to two billion people worldwide rely on a water source with some level of fecal matter contamination for their drinking water.
Detecting water contamination is the first step in preventing water-related morbidities. This gives an indication that precautions need to be taken, like boiling water before drinking it, for example. Unfortunately, current methods for detecting fecal matter in water require expensive equipment, and persons with extensive training to operate that equipment — this can be a disqualifier in the developing world. Moreover, results can take up to 24 hours to be ready, but the need for drinking water is immediate. These factors make regular monitoring of water supplies in many regions either impossible or impractical.
Researchers at the University of Colorado Boulder have developed a new method of detecting fecal matter in water supplies by using fluorescence and machine learning. This method is able to provide real-time feedback, can run on inexpensive hardware, and does not require any special expertise to operate.
The team chose the Particle Boron microcontroller development board as the basis for their hardware platform. The Boron has a powerful Nordic Semiconductor nRF52840 SoC onboard, and also comes with an LTE modem for wirelessly transmitting data over cellular networks. Connected to this board is a UV LED light source, which is shined on a water sample. It is then possible to measure the amount of light that is absorbed and re-emitted at a higher wavelength, which can reveal potential contaminants.
This is a tricky business, however, because the data is sensitive to many factors, like water temperature, which makes it very difficult to interpret the results. To help in understanding the complex interplay between these factors, the team turned to a machine learning model. A binary classifier was built that can predict when there is a high risk of fecal contamination. Validation of the model showed it to be accurate in 83% of cases.
The researchers note that their methods could be tweaked such that many other sources of contamination could also be detected, either biological or chemical. In the future, they would like to continue developing the technology such that it can be installed on a home’s main water line coming from the well (wells are highly susceptible to contamination due to septic system malfunction and closeness to livestock). The continuous sampling and real-time results that this device provides would then be able to trigger immediate alerts before illnesses have a chance to break out. The method described by the team has been awarded a patent by the U.S. Patent Office, and they have entered a partnership to bring a product to market, so this innovation may be making a real difference in people’s lives very soon.