As a technology, machine learning’s potency lies in its ability to discern patterns and produce intelligence out of volumes of data that no single human being could analyse. For infrastructure businesses like water companies, where processes and patterns are everywhere, there are many opportunities to bring this level of analysis and intelligence to the services they offer. In a pioneering piece of work, we have worked with Northern Ireland Water on a research project, funded by the UK government’s innovation-focused Small Business Research Initiative, to explore how machine learning could optimise and improve the reliability of water filtration.
Rapid gravity filtration is a key process in the production of safe, clean drinking water worldwide, removing particles and pathogens through physical straining, sedimentation and electrostatic attraction. The performance of the filters changes through time and the filters must be cleaned regularly – it’s an energy-intensive process, traditionally carried out on a fixed schedule. The performance of these filters determines the cleanliness and safety of the drinking water. How could this analogue process produce digital insights?
Our idea was simple: develop a machine learning model that can analyse historic and live data to predict the performance of rapid gravity filters. The model uses data from observed changes in the quality parameters of the water passing through the treatment works, to make predictions about the future performance of the filters, helping operators to optimise their cleaning and operation.
We built the model in Python, initially as a desk study, to test the model’s ability to process and learn from real historic data and make accurate predictions. In its next development stage, the model will begin ingesting live data, deepening its understanding of the filters’ performance. By the end of winter 2021 it will be offering plant managers a detailed picture of the filter performance for the first time, assisting in the running of an actual treatment works, helping the operator to reduce the risk of a water quality failure.
For the water industry, machine learning can clearly help operators to reduce the cost of clean water per megalitre. At a broader level, this project demonstrates that it’s possible to introduce machine learning into the sector’s core processes and optimise quality and safety, boost productivity, optimise maintenance, and reduce energy consumption at the same time. Given Northern Ireland Water’s own carbon neutral commitments, tools like this clearly have value by finding new ways to reduce waste and energy use. And with over 90% of the world’s water companies using filtration systems similar to those in use at Northern Ireland Water, we believe that this approach could represent a valuable step-change for the industry.
There are many other related industrial and chemical processes within water treatment plants, currently generating data, but not yet providing intelligence. We’re excited to help the sector develop its digital future.