The Smart Sewer AI Revolution


Join the Smart Sewer AI Revolution


2021 is likely to be a significant turning point in the development and deployment of the ‘smart sewer’. A revolution you might say.


Long talked about, solutions that can make accurate predictions in sewer pipe and pumping station flow during wet as well as dry weather conditions, as well as predicting sewer blockages, are finally making their debut. Accurate predictions now enable early intervention to remove blockages and prevent spillages that would otherwise have significant negative cost and environmental impacts.


It will be no surprise to learn that this huge step forward is enabled by new technologies such as AI (artificial intelligence), predictive analytics and cloud computing which are now beginning to provide case studies with unprecedented results.


Drivers for change

Urbanisation, climate change, and population growth challenge ageing wastewater and drainage infrastructure around the world. At a local level, sewer performance is also affected by varying levels of customer awareness so regrettably, wet wipes, fat, oil and grease continue to make their way into the sewers which can lead to the formation of “fatbergs” and sewer blockages. In the UK, removing sewer blockages costs more than £100 million each year, and if a blockage is not removed in time then it may escape into the environmental, causing a pollution or flooding event.


At current rates, it would take centuries to replace existing sewers. This means that for service to be maintained,  it’s crucial to maximise existing network capacity in a cost-effective way that keeps wastewater services affordable for customers.


Additionally, the current pandemic and an increasing realisation that we must do more to combat the climate emergency, has caused many in the water industry to re-examine traditional approaches.


Given these challenges, wastewater service providers must look beyond the status quo and embrace the AI alternative.




What are ‘smart sewers’?


Smart sewers are foul or combined sewer systems that use real-time data and machine learning techniques to improve the performance of utility/municipal wastewater networks. A smart sewer uses field level and flow sensors and telemetry data as inputs for a combined analytics and AI systems to predict flows and improve operations and decision making.


A smart sewer AI system has some major advantages over other approaches for smart sewers:


  • Real-time blockage alerting: AI “learns” the asset behaviour in a range of conditions using data logs then continually predicts and updates the dynamic thresholds for the expected asset performance and compares these to available real time data. When thresholds are breached an alert is automatically generated and issued to operational decision makers.
  • Ease and speed of implementation: Little internal resource is required to get a proof of concept or trial going. Typically, the system can be up and running in two weeks with usable results available immediately.
  • Highest degree of accuracy: AI is proving to be significantly more accurate than other methods.
  • Scalability: Highly scalable, being able to move from a catchment trial to full network deployment with ease
  • No need for modelling: No hydraulic model(s), or modelers, are required to build or maintain a model. This reduces complexity, time to market and cost.
  • Topology agnostic: The terrain of each asset is unimportant to the machine learning algorithms.


Why smart sewers now?


Several factors are coming together to collectively act as an industry catalyst, namely:


  • Increasing regulatory pressure : Increasing public expectations leading to increasingly stringent regulations raise the bar on acceptable levels of sewer escapes. Leading AI solutions have been proven at catchment level and are ready to scale.
  • Increasing level sensor deployment: The deployment of ever-increasing numbers of level sensors in response to tightening regulatory regimes especially in UK, Europe and North America and Australasia. For example, United Utilities in the UK are adding 14,000 level sensors in their network over the next 5 years.
  • Wastewater rising up the water company agenda: A rebalancing of water industry capital investment towards wastewater


What does a smart sewer solution look like?


The key to any smart sewer solution is the ability to accurately predict level thresholds for each single sewer asset in real-time. These dynamic thresholds are set based on key inputs such as sewer levels and hyperlocal rainfall data and then updated real-time, several times per hour, as new inputs are received. Any breach of these forward-looking thresholds by the actual sewer level data for a meaningful period of time will trigger an alarm. In turn, maintenance crews can then proactively investigate the alarm.


A leading solution will have a number of key components:


  • Bespoke machine learning: Customised machine learning for predictive (dynamic) sewer level thresholds and alarms. The machine learning algorithm suite must have been tried and tested in wet weather conditions as well as dry weather.
  • An approach for cleaning sewer level data: Sewer levels tend to increase and decrease rapidly when rainfall runoff flows into combined sewer systems. For example, in the minutes following a rainfall event, the sewer level can quickly rise to 8-10 times dry weather levels.


These sorts of spikes in rainfall data are not an ideal condition for standard data cleansing techniques. Standard (off the shelf) techniques tend to “clean out” this rainfall related ‘spiky’ data, so, AI companies must develop specific analytics to accurately clean sewer level data without removing the rainfall related data. These algorithms target false level data from things like benching, step irons etc. providing the cleanest possible data sets to input into the machine learning models.


  • Real-time rainfall forecast inputs: Real-time feeds from a hyperlocal weather forecasting system to help set dynamic thresholds.
  • Real-time sewer level inputs: The ability to easily accept real-time feeds (where available) from sewer level monitors. Any system must be monitor brand-agnostic given an average wastewater company typically deploys monitors from a range of providers.


What smart sewer examples are out there?


In the UK, Wessex Water have had excellent results with a 3-month trial during 2020. In this example the deployed solution used an advanced machine learning suite and hyperlocal rainfall forecasting to accurately predict sewer blockages with 95% accuracy, identified likely CAT3 spillages before they happened and enabled condition-based maintenance of the sewers. It also rationalised alarms in the control room, filtering our unnecessary alarms, allowing operations teams to focus on those related to real network problems i.e., blockages and anomalies.


In the US, cities like Cincinnati, South Bend and Kansas City have used AI to manage attenuation. In Kansas City a real-time decision support system has been deployed to actively control the flow of water and prevent combined sewage entering the Missouri River. This system increases the storage capacity in the sewer systems by using the in-line gates during heavy rains – similar to the smart traffic lights working during peak hours.


The benefits of smart sewers


There are several, powerful and proven benefits to using AI approach on smart sewers. These include

  • early identification of blockage & anomaly detection;
  • enabling condition-based maintenance;
  • rationalisation of control room alarms;
  • non-compliant spillage reduction;
  • and the smart attenuation of stormwater.


Given the potential of these benefits to give customer satisfaction, environmental compliance and efficiencies for wastewater companies, the AI smart sewer space will likely experience high growth over the next few years with a growing number of deployments, case studies and AI approaches.