Bacteria vs The Bot, FIGHT!

It's #TechnicalTuesday and Austen Buck is exploring how artificial neural networks can help us manage future microbial challenges in drinking water.

I’m still traumatised by the time my year 7 maths teacher confiscated my Tamagotchi. He didn’t feed it, or clean up its mess, and it sadly died at the ripe old age of 20 days (that’s 20 years in human years). At the time, I was fascinated by how these toys worked, both in terms of the operating system and the electronic hardware behind it, so much so I used to take them apart to the dismay of my parents. As a kid, it was easy to begin to wonder if the little beastie on the screen was somehow sentient. Fortunately, it wasn’t…. I hope!

Technology has made incredible progress since the time of virtual pets and artificial intelligence (AI) has become mainstream from its initial inception in the 1940s with work into artificial neurons by McCulloch and Pitts. Artificial neural networks (ANN), a core component of AI, provides the brain-inspired networks that feed machine learning (ML) algorithms for AI to solve tasks. 

This image was created in under 10 seconds by an AI image generator tool using only 3 keywords: robot hitting bacteria. 

ANN has been used successfully in multiple applications such as stock exchange prediction, handwriting recognition, fraud detection, automated car systems, and chemical plant control, but its use in the protection of public water supplies is relatively new, especially in the field of microbial control.

An exciting study by Miao et al, 2022, investigated waterborne disease prevalence in urban areas and demonstrated how multiple linear regression models have been used in conjunction with ANN to accurately predict enteric pathogen concentrations, such as norovirus and rotavirus, in surface waters using physicochemical parameters alone. Typically, collecting and analysing water samples for microbial parameters can take a minimum of 24h but with techniques such as these, instant and accurate predictions can be made in the absence of real-time microbial concentrations.

If we can harness the vast amounts of telemetry data collected on water and wastewater treatment works in conjunction with ANN the possibilities are significant:

• Abstractions could be temporarily shut down if surface water microbial concentrations were predicted to exceed disinfection processes.
• Disinfection requirements could be temporarily reduced in real-time during periods of improved raw water quality, thus saving on power and dosing chemicals allowing for financial benefit, and a reduction in chemical usage.
• Conversely, disinfection could be increased in times of poor raw water quality, enhancing public health protection.

There’s also nothing limiting the use of ANN to potable water supply and significant advantages could further be levered in catchment management, wastewater treatment, cooling systems and industrial processes both for microbial and physicochemical risk management. For instance, live “traffic light” style warning systems could be implemented at bathing waters to inform bathers of the microbial risks based on recent weather patterns and local river flows.

The water sector currently has multiple priorities that need addressing and whilst the solution to many of these problems are traditional ones, ANN should be considered as a potential solution now and a priority for future investment.

In commemoration of my Tamagotchi that lost its life, and the future billions of microbes that might be up against systems controlled by artificial neural networks. RIP.