The water sector is currently installing sensors in sewer networks across the globe. In the UK alone, it is expected that water companies will install more than 300,000 sensors by 2030.
Why install sensors in wastewater systems?
It is partly driven by regulation, but the current lean towards deployment is also enabled by a dramatic drop in the price of sensors that opens up more efficient and controlled operations of sewer and wastewater systems.
A monitored sewer system provides the means to improve many key objectives, including:
- forecasting and preventing CSO overflows
- identifying blockages, flaws, inflow and infiltration
- identifying the source of sewer contamination
- predictive maintenance and condition monitoring
- improved performance metrics, etc.
South Bend, Indiana, in the United States of America is a prime example of the benefits of installing sensors in sewers. Following regulator penalties in 2011 South Bend deployed a smart system in their sewers, which provided data to more efficiently control the network and consequent CSO spills, as well as decreasing the amount of new infrastructure required to manage flashy or intense weather events.
By 2020 the deployment of the sensors in the sewer limited the overflow per inch of rain from 42.8 million gallons in 2008 to 6.9 million. South Bend is now able to design their systems to their actual requirements rather than a ‘finger-in-the-air’ plan. The water company foresees that the results will bring CSOs down to zero, and provide control over 600 miles of sub-surface pipes.
Transitioning to a monitored system
The transition to equip the sewer system with sensors can be an enormous and extremely costly undertaking for water companies in England and Wales, and across the wider international sector. Public health is paramount, but there are other issues which come in to play. Regional regulations, geographical constraints, population movement, extreme weather events and ageing infrastructure all contribute. Water companies know they need good data to manage the systems, but that improved knowledge could also lead to increased penalties while performance is brought under control. Sensor deployment becomes a tricky decision.
There is currently a degree of trial and error to the decisions about how many sensors to purchase and where to put them. The cost of sensors, physical constraints such as the ease of access, local geography and IoT connectivity, add to expert opinion, but the “gut-feel” approach does not allow for measurable deployment and detection targets, so quantitative requirements are not accounted for. While Southern Water in the UK has taken the ‘blanket’ approach, casting a wider data net across their networks, Thames Water has looked at smaller areas to gather ‘niche’ understanding of network performance. Both companies are now changing tactics to fill in incomplete knowledge.
Both the Blanket and Niche approaches have their benefits and detractors, but without combining a portfolio of information there is no certainty that sensors have been deployed in the best places for the best intelligence gathering. With many parts of networks unmapped, there is the added risk that when the experts leave the business there are no reference points for why the deployment occurred in the way it did.
Using machine learning and data science for optimal sensor placement in water networks
The question we hear widely in the sector is, how can we reduce the number of experiments or sensors without sacrificing confidence in our results?
Optimal experimental design or *optimal sensor placement*, are two perspectives of the same underlying problem.
Research on the topic includes both mature methods such as "traditional" optimal design (see Goos & Jones, 2011) and sophisticated response-surface methods (see Garnett, 2023). While these approaches are highly generalised and can be applied to virtually any problem, their relevance to practical engineering problems, such as the placement of sensors in a sewer network, is not obvious. Additionally, much of the literature on the topic is impenetrable to non-statisticians so can be out of reach for engineering domain experts.
However, these methods have been tested and validated in other safety critical industries such as nuclear and aerospace. Acceptance into the water sector is slow but, where the approaches have been adopted, results are obvious. The City of Milpitas has saved $50m over the lifetime of their energy and water networks upgrade by introducing IoT sensing systems, smart water management and smart grid certainty.
This is where data science experts within water companies, design engineers, contractors and manufacturers can help. Their understanding provides the confidence to accept enabling tech as part of the matrix of solutions the sector will need to overcome the current issues. Understanding the data science and coding to reduce the number of sites while remaining confident that the data will provide excellent intelligence, is key.
Placing Sensors with Confidence
Monitoring a sewer system adequately is a complex and challenging task. Placing sensors is the first step towards a robust decision support system, but a sensor deployment strategy can make or break this system. Sensor placement typically relies on a variety of data which is chosen and assessed by an expert. While this can certainly result in good placement strategies, the efficacy of a given strategy is neither quantifiable nor traceable.
The sector needs an alternative, rigorous approach to sensor placement in water networks, where the decisions that determine the placement strategy are entirely transparent and traceable. Explainability is extremely valuable in safety-critical systems. Where it is fully automated to reduce the number of required sensors, the workforce can be redeployed on further complex projects where expert knowledge is required. This improves the current approach on multiple fronts:
- by reducing the cost of monitoring a sewer network,
- by simplifying the auditing of plans and proposals,
- and by reducing the workload for the sensor deployment team.
With good placement comes good data, and that leads to better outcomes for the sector. Fewer spills, mitigating extreme weather events, better asset management, improved performance and risk management, and less disruption in the transport network, are to name a few, but the consequent improvements give the public more confidence in their water systems, help to change damaging behaviours and give good outcomes for public health and leisure. Who doesn’t want that?
Dr Mikkel Lykkegaard is a domain expert in water systems and data science. Find him on Google Scholar and LinkedIn.
Sarah Brooks is an experienced business development and innovation lead for the water sector, with a background in both industry and the academic world.
digiLab is an Ofwat Discovery Challenge Finalist, currently developing senSiteUQ, a software tool which offers an easy-to-use, cloud-based software application for the optimal placement of sensors in sewer networks. It integrates high-level information, including sewer network models (InfoWorks, MIKE+, SWMM), weather data, surface conditions and IoT connectivity.
senSiteUQ places sensors based on a rigorous, scientific approach that ensures that each sensor contributes the most and best possible information about the network.
Sensor placement designs can be targeted to forecast CSO overflow, identify blockages, perform predictive maintenance, or a combination of all of the latter.
It is being developed in close collaboration with Ofwat, sector experts and industry partners in England and Wales and internationally, to ensure alignment with industry needs.
senSite Features
- Simulation-driven or simulation-free optimal sensor placement.
- Integration of geospatial data, such as IoT connectivity and surface conditions.
- Multi-objective optimisation for blockage detection, CSO overflow forecasting and contaminant source identification.
- Compatibility with industry standard modelling tools, such as InfoWorks, MIKE+, and SWMM.
- Simple and intuitive interface, made for practitioners.
