The main objective of the proposed project SMaRT-OnlineWDN is the development of an online security management toolkit for water distribution networks (WDN's) that is based on sensor measurements of water quality as well as water quantity. To this extend, five main research aims are defined as 1) Smart sensors and alarm generation, 2) Online Simulation Model considering hydraulic state and water quality, 3) Optimal Location of Sensors, 4) Online Source Identification of Contaminants and 5) Risk analysis, identification and evaluation of impacts (real impacts and perceived ones). The project aims are discussed in more details below:
1 SMART SENSOR AND ALARM GENERATION IN WATER DISTRIBUTION NETWORKS
During the last years powerful Smart Sensors for water quality monitoring in supply networks have been developed. On one hand there are so called multi-parameter probes for the measurement of several physical and chemical water parameters (e.g. temperature, pressure, conductivity, pH, redox potential, diffusion), on the other hand, broadband sensor systems for the online measurement of toxic ingredients have been developed (e.g. bbe moldaenke DaphTox, microloan iToxControl).
In order to detect changes with both types of sensor systems, the measured relevant physical, chemical and / or toxicity parameters from the sensor systems have to be evaluated. The difficulty in implementing model-based monitoring systems lies in the formulation of a suitable process model. This approach requires detailed expert knowledge about the physical and biochemical interactions involved in the sensor principle. These corresponding tools in general use a large number of parameters that strongly depend on the water network and the asset specific properties of the water. The adaptation of this large number of parameters in these monitoring systems is very time consuming and – even more importantly – a suboptimal parameterisation will eventually generate false alarms, which is not acceptable for the water suppliers.
An alternative to analytic model-based methods for monitoring water quality is offered by data-driven modelling techniques. On the basis of measured historical sensor data, machine-learning approaches can be applied to automatically generate a model for monitoring the water quality. The advantage of this approach is that no analytically formulated expertise is needed a priori, and thus the user is not hampered by "unsafe" assumptions about the physical / biochemical sensor principle. In the recent past, the Fraunhofer IOSB has successfully developed several concepts to monitor industrial and biochemical processes based on machine learning methods. These methods will be adapted and extended during the project SMaRT-OnlineWDN in order to develop a reliable alarm generation module for water distribution networks.
2 ONLINE SIMULATION MODEL FOR RELIABLE WATER QUANTITY AND QUANTITY PREDICTIONS
The core of the online security management toolkit consists of a grid of smart sensors in combination with a simulation model. The boundary conditions of the network model are regularly updated by measurement data guaranteeing the compliance of the model with the observations. The consistency of the measurements is checked, for example by use of an Artificial Neural Network. With this information the online security management toolkit is able to reflect the current hydraulic state of the entire system. In addition, monitoring water quality parameters supports the detection of biochemical contamination of the drinking water.
Innovations of the proposed approach are:
Generation of real-time, reliable (i) flow and pressure values, (ii) water quality parameter values of the whole water network
Consideration of water quality parameters in the online hydraulic simulation model and vice versa
Semi-automatic aggregation of complex networks
Semi-automatic update and online‑calibration of the model
Take into account nature of consumers and impacts evaluation through risk analysis methods
3 OPTIMAL LOCATION OF SENSORS
Over the last decade a significant number of research projects have focused on the problem of finding an optimal sensor network design for detection of deliberate contamination in water distribution systems. The sensor placement problem is usually formulated as a multi-objective optimisation problem. The result is the trade-off between different, competing design criteria (e.g. maximising the area observed and minimising the time to detection). A common drawback of the existing methods is that they are based on offline hydraulic outputs. In general, an extended period simulation model is used whose demand patterns, operational states of control devices as well as the input mass of contaminant are assumed to be known. However in reality, the spread of contaminant as well as appropriate countermeasures in case of an emergency highly depend on the actual flow conditions of the network that for a looped network are unknown. This issue enhances the need for online measurements to overcome data uncertainties and random variations. For application of reliable hydraulic simulation in such a case it is mandatory that the simulation model is automatically updated by online measurements correcting the common offline outputs. As a consequence not only water quality sensors are required but also sensors that measure hydraulic values like pressure and flow.
Therefore in this research project, the optimal sensor placement problem is reformulated in the context of an online hydraulic and water quality monitoring system. The objective is to find the best locations for sensors that guarantee both the proper estimation of the hydraulic state of the system as well as detection of changes in water quality.
As a second innovation the sensor network optimisation problem is not understood as green field planning where the best locations for a given number of sensors has to be found rather than a successive improvement by expanding an existing sensor network. This approach better reflects the real circumstances of water supply utilities as usually a sensor network (at least for hydraulic data) already exists.
The new approach for optimal sensor placement shall also include the development of a performance index for the quality of the conclusions that are possible with the existing number of sensors in a network and the improvement resulting from additional sensors.
4 ONLINE SOURCE IDENTIFICATION OF CONTAMINANTS
In this work package existing deterministic and probabilistic methods for source identification that have been developed for offline models shall be integrated in an online simulation framework. It is assumed that an online simulation model exists and produces hydraulic network data (e.g. every 3 minutes). The online source identification requires that the online model is sufficiently calibrated and validated and reflects the actual hydraulic state of the real water distribution network with sufficient accuracy. The desired degree of accuracy of the model is one issue that has to be investigated. Based on the online-model a backtracking algorithm that uses the data history of the measurements has to be implemented. As a result of water quality sensor alarms the possible localisations of the intrusion of contaminant can be calculated. The more sensors that are available, the better the identification of location and time of intrusion.
5 RISK ANALYSIS, IDENTIFICATION AND EVALUATION OF IMPACTS (REAL AND PERCEIVE ONES)
In this project risk analysis and impact evaluation will be taken into account for the three aspects of sustainability: environmental, social and economical, combined with technical innovation provided in the other work packages. For instance, the social aspect will be studied to evaluate the importance of the perception of water quality for the consumers, in the management of the distribution of this resource. Indeed, this perception can modify the consumers’ behaviour in relation to their use of water and can have an impact on the economic or environmental aspects of the distribution. Then, the precise knowledge of the perception of the water quality can help the calibration of the model of distribution by integrating data about the way information about the contaminants will be perceived by the consumers, what are the levels of quality they expect, the general behaviour they adopt in their use of water and what trade‑off they can accept in terms of water systems, etc.