Peter Chalk Centre

University of Exeter

Stocker Road

Exeter

EX4 4QD

Tel: +44 (0)1392 263637

E-mail: CCWI2019@exeter.ac.uk 

17th International Computing & Control for the Water Industry Conference

1st - 4th September 2019
University of Exeter, UK
  • Facebook Social Icon
  • Twitter Social Icon
  • Instagram

4C Data driven analytics

Please note the time shown on this page is automatically adjusted by the server according to the time zone set in your computer.

Collaborative

Kegong Diao

Chair:

to

A comparison of feature descriptors for automatic CCTV sewer analysis

Joshua Myrans

Presenter:

Authors:

Joshua Myrans, Richard Everson and Zoran Kapelan

Water companies must regularly survey their sewers to prioritise maintenance to ensure the effective operation of the wastewater network. For any pipe too small for human entry, analysis is performed using CCTV cameras, recording footage of a pipe’s interior. This work presents a comparison of image feature descriptors used in conjunction with an existing automated sewer fault detection method. This method aims to automatically label faults from raw CCTV survey footage. The calculation of feature descriptors forms a crucial part of this process, identifying key qualities of a CCTV frame to be classified by a random forest classifier. This work explores the application of HOG, GIST, LBP and ORB feature descriptors. Furthermore, this work has been performed in conjunction with UK water company South West Water. Enabling access to a new dataset of labelled images, the automated sewer fault detection methodology is shown to be robust and flexible, achieving accuracies of up to 85% when applied to independent CCTV frames.

to

Data Analytics for Automated Detection of Blockages in Sewers

Talia Rosin

Presenter:

Authors:

Talia Rosin, Michele Romano, Zoran Kapelan and Ed Keedwell

Sewer blockages are a major issue in cities around the world. In the UK alone there are approximately 300,000 blockages every year, resulting in costs of £100 million and causing issues such as sewer flooding and environmental pollution. A novel Event Detection System (EDS) is presented in this paper that enables the detection of blockages and other unusual events at, or in the proximity of, a Combined Sewer Overflow (CSO) in near real-time. The methodology consists of two different blockage detection modules, (i) evolutionary artificial neural network discrepancy analysis and (ii) statistical trend based analysis. Both modules utilise statistical process control techniques to identify unusual behaviour in the CSO caused by blockage events. A simple inference engine is used to combine the results from the two modules to reliably detect the presence of a blockage. The EDS was applied to a case study site and demonstrated to detect a blockage event quickly, in under two hours, with no false alarms. It is envisioned that the system will be valuable to wastewater utilities, enabling proactive management of blockage events and therefore reducing the impact on the customer and the environment.

to

Lessons learned from comparing smart meter water consumption data with measured wastewater flow in the drainage system

Nadia Lund

Presenter:

Authors:

Nadia Lund, Morten Borup, Jonas Kjeld Kirstein, Ole Mark, Henrik Madsen and Peter Steen Mikkelsen

Many urban areas face water-related issues, such as flooding, sewer overflows, and water scarcity. Other cities wish to optimize their existing systems in respect to electricity consumption, customer relations, etc. Many professionals see ‘digitalization’ as a mean to solve some of these issues. Various ideas are being proposed, and research within digitalization of urban water systems has increased notably over the past few years. Data is a key aspect of digitalization, and many utility companies in the industrialized part of the world have digitized their asset data and used this to build detailed dynamic models of their systems to better understand their behaviour. Many utilities also increase their focus on observing this behaviour by collecting flow, level, pressure, and velocity measurements. But what is the actual quality of the data collected, and how easy are they to utilize? We investigate these questions in a practical “learning-by-doing” study where we compare spatially distributed water consumption data from the city of Helsingoer, Denmark, with flow observations in the sewer system using a detailed hydrodynamic model. Here, we find a mismatch between the measured water consumption and the wastewater flow. What is the reason for this difference? Are we missing a large consumer? Is the smart meter or wastewater flow data erroneous? Is someone using water from the harbour and discharging it to the sewer system? Is the water distribution network leaking directly and into the sewer system? Is another part of the city connected to this part of the sewer system, without this being registered in the utility company’s asset database? The possible reasons are many, and investigating these requires the activation of either the employees within the utility or several of their external contractors, which makes it even more complicated to find the answers. Our detective work is still on-going.

to

A data assimilation scheme tailor-made for real-time modelling of urban drainage tunnels

Rocco Palmitessa

Presenter:

Authors:

Rocco Palmitessa, Peter Steen Mikkelsen, Adrian Wing Keung Law and Morten Borup

Around the world, urban drainage systems are struggling to fulfil their function due to pressures from climate change, increased urbanisation and stricter environmental regulations. As a response, many cities adopted drainage tunnels, which provide additional storage volume while conveying water to final disposal. To take full advantage of the storage capacity of tunnels, these need to be optimally controlled. Model-based control has the potential of informing risk-based decisions, but is only useful if the initial conditions of the forecast are kept close to the real state of the system. This can be achieved using data assimilation, which corrects the model states dynamically based on a comparison with observations from sensors. Ensemble-based assimilation methods have the added advantage of producing uncertainty estimates, but require a large computational overhead. We developed a data assimilation scheme tailor-made for real-time modelling of drainage tunnels. By restricting the assimilation to the tunnel model, instead of the full catchment, the scheme retains the benefits of distributed models and ensemble-based methods, while reducing the computational overhead. We applied the assimilation scheme to the MIKE URBAN model of the Damhus tunnel, Copenhagen, Denmark, for testing and validation. For a single intense rain event, the scheme achieves a significant improvement in the estimate of water levels 3.4 km upstream of the observation location along the tunnel. Future research will investigate how well the assimilated water level improves model forecasts at different horizons.

to

Applying Deep Learning-based Data Analytics with Hydraulic Model Calibration for Anomaly Detection and Localization

Zheng Yi Wu

Presenter:

Authors:

Zheng Yi Wu

An integrated solution framework has been developed for anomaly detection and localization. Figure 1 illustrates the solution procedure. A comprehensive solution method is proposed for anomaly detection and localization by formulating and integrating the data analytics methods with the Pressure Dependent Leakage Detection (PDLD) that has been developed by the PI. Data analysis for anomaly detection proceeds in multiple steps including (1) data-preprocess to eliminate and correct error data records, (2) decomposition of time series data to ensure stationarity, (3) outlier detection by statistical process control methods and the Deep Learning with Extended Kalman Filter (DL-EKF), (4) anomaly event classification per sensor, so-called sensor event, and finally (5) system anomaly classification by correlation analysis of multiple sensor events.

to

Identifying the Least Amount of Data for Reliable Life Cycle Assessment of Water Distribution Networks

Mohsen Hajibabaei

Presenter:

Authors:

Mohsen Hajibabaei, Sina Hesarkazzazi, Florian Gsch�sser and Robert Sitzenfrei

During the last years, life cycle assessment (LCA) has been widely used to evaluate the different parts of urban water systems (UWSs). Water distribution networks (WDNs), as one of the main components of UWSs, play an important role in providing public services. Using LCA for WDNs, some studies have focused on the comparison of different materials used for pipes and piping process. Although these types of comparisons in WDNs could be useful to understand the environmental impact of different materials, it is not clear how much data is actually needed for reliable LCA. In other words, the effects of availability of various types of data for LCA of WDNs has not been systematically investigated. To close this gap, in this study, the amount of information required to attain certain confidence in the environmental impact is systematically evaluated based on analyses of uncertainties and failure propagation. The results demonstrate the ability of the proposed approach for defining the least amount of relevant data needed for LCA of WDNs.

to