Yorkshire Water is responsible for a 55,000km sewer network serving around 4.5 million people. This includes thousands of combined sewer overflows (CSOs) that relieve the system during periods of wet weather, when it might otherwise be overwhelmed by rainfall. But what happens if these overflows become blocked?
Artificial Intelligence (AI) systems are increasingly being used to solve real-world problems, particularly where a large amount of data is available, but analysing and interpreting that data is difficult.
Recently Yorkshire Water, the University of Sheffield and Siemens Digital Industries got together to develop a system to improve the performance of the sewer network. It can identify a blockage in the network quickly, so that it can be dealt with before a problem escalates.
The SIWA (Siemens Water) Blockage Predictor is part of Yorkshire Water’s Pollution Incident Reduction Plan 2020-2025 which aims to reduce pollution incidents by 50%, by focusing on early intervention.
Heather Sheffield is the operational planning and technical manager (wastewater) at Yorkshire Water, and she explains that if a CSO becomes blocked, “The sewage will back up over the weir in dry weather.” This can be released into rivers – pollution which could have disastrous effects on health and the environment.
“We’re regulated by OFWAT,” says Sheffield. “It incentivises us as a company to improve service to customers and the environment with what are called performance commitments.” Every pollution incident that could have been avoided results in stiff financial penalties, even before the cost of rectification.
Around 2,100 CSOs are monitored by level sensors which indicate the state of the system. But conventional level monitoring in a water system is typically set to a threshold value – that is, it is activated when a particular water level is reached. As Siemens points out, “this is a blunt method for detecting alarms, as it cannot differentiate between level rises due to normal operations or an issue”. A more sophisticated system would also look at external factors, and this is the basis of the SIWA Blockage Predictor.
“The tool has the ability to alert us to a restriction of flow under normal operating circumstances,” says Heather Sheffield. “It can tell that there’s a blockage [which] will start discharging flow via a CSO.”
The system uses an Artificial Intelligence (AI) model built specifically for each asset or location. This reflects its specific response to rainfall, which is influenced by local geography and the layout of the network. How this works is explained by Joby Boxall, professor of water infrastructure engineering at The University of Sheffield. He says: “By building a personalised fingerprint for the wastewater assets that reflects how the local network responds to rainfall and overlaying that on patterns of daily behaviour, we have been able to establish what each asset’s ‘normal’ response is.”
Sensor data from each asset is combined with real-time rainfall data: “Using an analytics tool called ‘fuzzy logic’, the system then applies a further level of intelligence to judge if the predicted level is significantly different to the observed level based on how big the difference is, including expected response to any recent rainfall. My team compared the findings of the AI against the current system and what was actually observed over 21,000 days of operation – and the AI came out on top.”
In fact, the Blockage Predictor was able to find around nine out of ten potential issues – said to be three times more successful than the existing statistical methods for pollution prediction – and to give up to two weeks’ advance notice of blockages.
One of the key features of the system, says Heather Sheffield, is that “this didn’t require new operational technology to be put in place: we had that already”. The network’s CSO level sensors – some digital, some analogue – already communicate with the regional telemetry system via a fixed line or a GSM wireless link. This data is sent to the AI system, built on Siemens’ MindSphere platform (see box) and hosted remotely.
THE IMPORTANCE OF THINGS
This is described as an IIoT (Industrial Internet of Things) application – in other words, it uses data gathered from sensors embedded in real-world ‘things’ to come up with usable information. Typically, the sensors are quite simple (measuring temperature, air pressure or fluid levels, for instance). Examples are pictured at left (Technolog Cello CSO ultrasonic level sensor) and right (Detectronic LIDoTT). Their collective data can be used to build up a detailed and sophisticated picture of a system. The conclusions can then be sent via a web interface to a computer or any internet-connected device.
Continues Sheffield: “The blockage could be anything: it could be wet wipes, or FOG [fat, oil and greases] or concrete poured down a manhole. The predominant factor in an urban environment is a build-up of flushable wipes – which in fact don’t break down – and the AI can tell us when there’s a significant change in that environment.
“It will come through to my team and on a daily basis we raise jobs, investigate and clear the blockage. Yorkshire Water has offices all round Yorkshire, but the head office is in Bradford, where colleagues – in the operational planning and resilience department – sit in a control room in the service delivery centre.”
According to Siemens, the SIWA system can deal with challenges such as partial blockages, abnormal behaviour during summer rainfall, and shifts in baseline figures which might indicate an issue with the level sensor itself.
Avoiding false alarms is a key advantage: “Our existing process has false positives,” says Sheffield, “and the new system has significantly reduced them”. In fact, the reduction is said to be around 50%.
There may be other applications for AI systems at Yorkshire Water, she observes: “We are moving away from being operationally reactive to being proactive as much as possible, and therefore we are relying on predictive analytical tools to help us understand when we may have a service impact that is about to occur”. Another potential application may be predicting internal sewer flooding – for example cellar flooding, when sewers fail or are blocked. “And in the longer term, we want to look at that for above-ground assets as well – pumping stations and sewage treatment works, for instance.”
BOX: The MindSphere platform
SIWA (Siemens Water) Blockage Predictor is an application based on the MindSphere platform. While other AI and Machine Learning platforms can also connect with the IIoT – from Microsoft’s Azure system to SAP’s Leonardo and IBM’s Watson – Siemens describes its system as ‘the leading industrial IoT as a service solution’.The company says: “Using advanced analytics and AI, MindSphere powers IoT solutions from the edge to the cloud with data from connected products, plants and systems to optimise operations, create better quality products and deploy new business models”.
The MindSphere platform has many other applications; for instance, its City Graph system has been used in the Aspern area of Vienna, Austria to improve forecasting of the charging demand of electric cars and to help understand their impact on energy infrastructure.
Another example is the Smart Machine Assistant, a self-learning application for determining the optimal settings of an industrial machine in a complex environment. This helps determine the correct settings to allow an operator to produce consistently-conforming parts while external factors such temperature, humidity and quality of raw materials vary – particularly on machines which are not fully automated.