When AI helps water usage: Part 1

When AI helps water usage: Part 1

Water, a major challenge for cities

A civilization’s fate seems to be closely linked to its water usage. As a result, the roman aqueducts, Arabic gardens as well as the Paris’ sewers (created by the Haussmann baron) are the perfect example of how pivotal these eras were. Before being a source of pleasure, power or money, water is the source of all life. Despite all this, its management remains widely imperfect. Worse: each year, 2.2 millions of people die because of unsanitary water sources.

This goods, as rare as it is plentiful, constitutes one of the main challenges our cities have to face. Therefore, how can an AI help us manage our water better? To bring forth the first answers to this question, I met the CANN Forecast startup in Montreal as well as the water treatment plant of la Baumette in Angers. This latter stands out of the crowd by its ability to turn dirty water into real resources (electricity, fertilizer) for the Angevin region.

From predictive maintenance…

For nearly a year, the water treatment plant of la Baumette in Angers has been using AI technology to conduct predictive maintenances on its equipment. Under Veolia’s management, the plant uses connected units, called BOB, to anticipate mechanical failures. BOB is born from a collaboration between the startup Cartesiam and Eolane. This technology specifically uses an embedded AI (NanoEdge). Before being operational, the solution is put to a testing period (unsupervised). For nearly 7 days, Bob will listen to the vibrating signal of the equipment it will then have to supervise. After the testing phase, the connected unit is able to signal any defect which can potentially cause a breakdown or failure within the equipment.

As well as being particularly easy to use (BOB plugs itself onto the equipment), the solution is energy-friendly (one battery is enough for 3 years). The anticipating time of BOB is crucial as a breakdown could result in terrible environmental and economical circumstances for the plant. For example, a failure in an air-compressor could affect the plant’s entire operation chain and

thus result in the Maine (river where treated waters end up) being polluted.

As for CANN Forecast, it uses another part of the AI (supervised apprenticeship) to do predictive maintenance on the aqueducts. “In Montreal, as in some big cities, clean water conducts are reaching the end of their life. However, when a conduct breaks, cities lose both resources and money” (Naysan Saran, co-founder of CANN Forecast). The cost in repairing aqueducts was of 3 billion dollars in America in 2018. To tackle this kind of problems, CANN Forecast has developed a proactive solution (InfoBris) 5 times more precise than traditional methods. The startup, already working with ten cities on this matter (including Montreal), is therefore able to provide its partners with a list of aqueducts that could potentially break. To do so, its AI system analyses data linked to the breaks’ history but also to their areas and environmental conditions. CANN Forecast is part of a large ecosystem (including Mila, McGill Université and Environnement Canada) to create this solution.

… to smart management of water

Besides InfoBris, CANN Forecast has also developed InfoBaignade. This solution is able to predict, with a 95% accuracy, the concentration in E.Coli (unit of measure used to determine if the water is swimmable) in rivers. “Right now, measuring the concentration in E.Coli takes between 18h to 48h. In a dynamic environment such as a river, water contamination can vary from one day to the other. This method is therefore inefficient.” (Nicoles St-Gelais, flooding and other variables) the Montreal startup is then able to propose a smarter water management model which takes into account the dynamic aspect of rivers.

InfoBaignade’s platform

BOB allows for an optimized management of its equipments. The solution offers the possibility to check remote infrastructures and to replace periodic checkups by regular and targeted interventions.

An augmented intelligence rather than an artificial one

It is interesting to see that BOB doesn’t send data but insights via a LoRa network. Anyone using this unit has therefore access to an understandable and legible platform. This legibility is essential for the solution to work properly. Indeed, “BOB is an assistant to the plant’s agents” (Joel Rubino, Co-founder of Cartesiam).

Same challenges at CANN Forecast: “We try to understand how we can integrate this [InfoBris] within the decisional context of a city.” (Nicoles St-Gelais). For Cartesiam as for CANN Forecast, it is a matter of increasing the decision-making authority rather than replacing it.