A digital water strategy based on the digital water service concept to support asset management in a real system

Water distribution networks (WDNs) must provide a reliable, continuous, and quality service to ensure human consumption and support the execution of economic activities. There are multiple factors that have increasingly impaired water supply, such as pipeline deterioration, population growth (Sinha 2014), and water scarcity (Giustolisi et al. 2016). Deteriorated pipes are associated with higher background leakages and head losses (Lambert 1994), which increases the probability of pressure-deficient scenarios. For this reason, asset management in terms of leakage reduction is fundamental to guaranteeing a constant water supply with a perspective that seeks a rational and efficient use of resources.

The classification of leakages, also known as real water losses, includes two categories: bursts and background leakages (Lambert 1994). Bursts are associated with large outflows and supply disruption, while background leakages are small outflows from joints, fittings, and small cracks along pipelines, that are too low to be detected by active leakage control actions (Germanopoulos 1985). Consequently, background leakages have a high impact on the long-term mass balance of WDNs, and thus they are usually referred to as volumetric leakages (Berardi et al. 2018). Moreover, pressure levels that are higher than required for a correct service increase volumetric leakages that can cause accelerated pipe deterioration, resulting in a higher rate of bursts and/or failures of larger dimensions.

In this sense, leakage reduction diminishes the waste of water resources and increases the service life of pipelines, which boosts the hydraulic capacity of WDNs. Furthermore, it decreases the carbon footprint and costs related to water treatment and energy consumption for pumping (Colombo & Karney 2005). The reduction of volumetric water losses in the short-term (i.e., operational) and medium-term (i.e., tactical) time horizons, may be achieved through pressure control strategies such as the installation of pressure control valves (PCVs) (e.g., Saldarriaga & Salcedo 2015; Berardi et al. 2018) and district metered areas (DMAs) definition (e.g., Ferrari & Savic 2015). In addition, medium-term and long-term (i.e., strategic time horizon) solutions include pipelines rehabilitation (Alvisi & Franchini 2009), enlargement of existing tanks or renewal of pumping stations (Giustolisi et al. 2016), which require higher investments in comparison with short-term actions.

In this context of decision-making complexity, water managers need to be assisted by reliable, robust, and flexible tools for asset management decisions; nowadays, these can be based on digital technologies, exploiting the potential of Digital Transformation. In fact, the growing power of computers, as well as data acquisition and storage, has accelerated the introduction of digital tools in the field of WDNs. Such decision-support tools require, among others, the integration of (i) the concept of Digital Twin; (ii) advanced hydraulic modelling with a pressure-driven approach, accounting for variable topologies due to different operational conditions of the WDN; (iii) a Geographic Information System (GIS) environment with a consistent data format for the interoperability with analysis tools; and (iv) multi-objective optimization algorithms aimed at returning sets of feasible solutions with a variable budget.

This integration is here presented within a digital water strategy based on the paradigm of digital water services (DWSs), that represent the last mile for digital transition in WDNs since they promote the interaction with water operators, who may modify optimal alternatives based on expert knowledge. DWSs are developed as plug-ins in the GIS environment, incorporating advanced hydraulic analysis and optimization capabilities.

Hence, this research presents a comprehensive digital water strategy framed within a structured and replicable workflow that implements DWSs as decision-support tools for each step of the asset management workflow. This scheme consists of analysis of the topological domain of the WDN, and hydraulic model calibration based on mass balance as a starting point, followed by design of hydraulic DMAs integrated with pressure control and pipeline rehabilitation. Hydraulic networks of the cities of Modugno and Bari (Italy) are used to demonstrate the methodology application.

The presented digital water strategy is supported by advanced hydraulic modelling based on a pressure-driven approach embedded in the WDNetXL-WDNetGIS system (Berardi et al. 2018). It includes advanced hydraulic analysis capabilities, that fit crucial pre-requisites to support asset management and outperform the limitations of traditional modelling software, consistently with the digital transformation paradigm. The majority of traditional software (generally based on the traditional EPANET paradigm) only uses a demand-driven approach, which does not allow a consistent analysis of pressure-deficit scenarios and pressure-dependent leakages at a single-pipe level (Giustolisi & Walski 2012). In contrast, the advanced modelling technique used herein implements a pressure-driven analysis paradigm, allowing a more accurate evaluation of WDN functioning in several operational conditions, thus evaluating different features such as (a) the real volume of water provided to each consumer and (b) the volumetric water losses at the pipe level.

Regarding water demand representation, recently traditional models tried to embed a pressure-driven function (Wagner et al. 1988; Giustolisi & Walski 2012), although still maintaining water demands as aggregated at nodes. This condition does not consider the different connection configurations that each user may have and is ill-adapted to the scenario that digital transformation is outlining, for example, single user telemetry that water companies are using for billing consumption. Conversely, the advanced modelling approach here used accounts for each user connection with three possible configurations: direct connection, direct connection accounting for floors number of supplied building, and connection with private storage tank, thus having water storage capability and a dedicated pumping system. Hence, the used model incorporates the real elevation and the georeferenced location of each user meter, as well as the database of water consumption of each consumer, which is usually available at water providers. Such a database is kept apart from the model so that changes in the status and number of connections do not require changes in the model definition.

In addition, traditional approaches model leakages as emitters, that correspond to nodes with a pressure-dependent outflow. However, background leakages are distributed among the network, and they are an indicator of pipe deterioration, thus they can be more realistically represented as outflows for each pipe (Giustolisi et al. 2008a). In this sense, the method used here represents leakages as uniformly distributed outflows at the pipe level considering the dependence on pressure and on deterioration through the structure of the Germanopoulos' model (Germanopoulos 1985) (β parameter), or through the FAVAD approach (May 1994; Van Zyl & Cassa 2014). This modelling technique improves the accuracy of the hydraulic analysis and provides information about the deterioration level of every pipe of the system, which is essential to support decisions related to asset management, such as pipe rehabilitation (Laucelli et al. 2023).

In the case of variable WDN topology (e.g., due to works and/or interruptions), traditional models consider a fixed topology while assigning a very small flow rate to pipes to close instead of removing them from the hydraulic analysis; this introduces accuracy problems and does not contemplate the possible disconnection of some sections of the WDN. The advanced hydraulic modelling here used accounts for WDN topology really connected to water sources before and during hydraulic simulations since it automatically detects network segments based on the status of devices, such as pressure reducing valves or flow meters (Giustolisi et al. 2008b). In this sense, the used tools exploit the complex network theory (CNT) metrics tailored for WDNs to perform the topological analysis of networks and drive DMA design in an optimal way (Laucelli et al. 2017). Moreover, the used advanced hydraulic modelling is fully interoperable with GIS given its consistent data format.

This work shows the application of a thorough digital water strategy that integrates DMA planning, pressure control, and system rehabilitation in a GIS environment to find robust and flexible alternatives to support asset management decisions for system operators. This application is performed through DWS using the Digital Twin of the WDNs (Ciliberti et al. 2021; Laucelli et al. 2023).

In the first step, the consistency of the topological and demand data is integrated, verified, and corrected when necessary. Next, an analysis of the topological domain is performed to identify the most relevant elements (pipes and nodes) of the network graph before any hydraulic simulation, to improve the understanding of WDN behaviour. Subsequently, two parallel processes are executed: (a) a calibration based on mass balance using hydraulic measurements and topological data and (b) a topological segmentation into virtual districts to identify the candidate positions for flow meters and closed sectioning valves. The latter provides alternatives for the next hydraulic DMA design integrated with pressure control. Finally, optimal rehabilitation plans are identified based on cost–benefit analysis.

The reduction of volumetric water losses is achieved in two complementary ways: the DMAs design integrated with pressure control and the pipeline rehabilitation. First, closing valves to delimit DMAs reconfigures the water paths to achieve little reduction of average pressure throughout the network while avoiding pressure deficit at consumers. Additionally, the implementation of PCVs in optimal locations puts under full control the pressure head on the WDN as well as the water losses. Second, pipe rehabilitation/replacement in a pressure controlled WDN can further reduce leakages, while preserving the necessary pressure levels for a correct service at users, avoiding unintended pressure surges due to new pipes in a network with uncontrolled pressure. Further details on each step will be provided in the next sections.

This paper presents the above-mentioned integrated approach on the WDN of Modugno (36,000 inhabitants) and Bari (330,000 inhabitants), two cities in the Apulia region, Italy. The hydraulic model of Modugno WDN consists of 1,814 pipes and 1,585 nodes, with three reservoirs. One of these feeds both Modugno and an area of Bari, the largest city in the area. The hydraulic model of Bari WDN consists of 9,057 pipes and 7,762 nodes, with eight reservoirs. The network of Bari was modelled and analysed considering six hydraulically independent sub-networks: Torre a Mare, Japigia, Bari Downtown, Carbonara-Ceglie, Loseto, and Santo Spirito-Palese. The sub-networks correspond to neighbouring municipalities to the city of Bari.

Results on Bari WDN will be reported to demonstrate the application on a large system, while the Modugno WDN will be used to illustrate the details of the procedure along with relevant considerations for each phase, focusing on how DWSs support the execution of every phase (Laucelli et al. 2023).

The topological and hydraulic data of the Modugno WDN was provided by the water company Acquedotto Pugliese in EPANET and ESRI^® shapefile format. The information consisted of:

• a geodatabase that included the attributes of pipes, tanks, and devices;

• the georeferenced water meters and consumption database for each user;

• information about the replaced and recently built pipelines;

• data recorded at flow and pressure meters in 2019.

As a first approach, the data were analysed to detect possible inconsistencies and errors by comparing hydraulic measurements and topological data (e.g., inconsistent flow/pressure data). Initially, one missing PCV was identified in the Modugno WDN. Subsequently, the advanced topological functions based on CNT were used to perform two verifications. First, this tool simplified the topology of the pipes with isolation valves including them as a feature of the adjacent pipes instead of separated pipes with one valve each. Second, double nodes were identified and joined into one to restore hydraulic connectivity. The georeferenced water meters and the water consumption data were included in the hydraulic model. The software allows specifying the type of consumer (e.g., commercial, industrial, household), the status of the service (i.e., active or inactive) and the attributes of the connection (e.g., number of floors and height of the buildings, location of private tanks, directly connected taps) whenever available. Hence, the value of the data was increased by integrating and verifying topological, hydraulic and water demand information into the WDN digital twin.

The analysis tools allowed identifying the existing monitoring districts based on available flow meters and closed gate valves. These are here defined as pseudo-districts since there are no flow meters at the inlet/outlet of the districts, while pressure is attempted to be controlled by closing several valves and each one of them is fed by a single different reservoir.

WDN topology mostly determines its hydraulic regime. The analysis of the topological domain of WDNs involves the evaluation of centrality metrics derived from the CNT, that are reformulated for WDN infrastructures. The metric applied for the WDN topological analysis was the Tailored Edge Betweenness, which measures the importance of each pipe accounting for the intrinsic relevance without considering hydraulics (Giustolisi et al. 2019). The integration of CNT metrics allowed us to conceptualize WDNs as graphs and identify the prevailing elements of the network considering the shortest paths of the system (Giustolisi et al. 2020). In such a context, domain analysis is essential for the methodological proposal since it allows to define the main paths of the system, whose hydraulic resistance is likely to have a strong influence on the pressure status of the network, and thus deserves careful calibration. Moreover, such analysis supports the next virtual segmentation of the WDN since it identifies the most relevant pipes as candidate locations of flow meters during DMA design, hydraulically consistent with the connectivity of the system.

The hydraulic model was calibrated with an innovative approach based on the concept of mass balance (Berardi & Giustolisi 2021), which leverages the capability of separating the consumers' water demands (stochastic demand component) and the volumetric water losses (deterministic demand component) from the monitored total input volume m thanks to the pressure-driven analysis implemented in the advanced hydraulic model. This procedure allows us to identify, for each monitored area, primarily: the β parameter of the leakage model for each pipe, and a unique consumers' demand pattern. In the case of Modugno WDN, the uncertainties about the pseudo-districts shown above makes it technically reasonable to consider the entire network as a unique area with monitored inflow close to the three reservoirs.

Therefore, a unique demand pattern is assumed to be observable and calibrated for the entire system. Furthermore, this strategy also allows us to simultaneously calibrate the hydraulic resistances of the most relevant pipes, which are identified based on the topological domain analysis, while the hydraulic resistances for the remaining pipes were assumed from previous models and verified considering the ranges in technical literature. The initial values of the β parameter were assessed for each pipe based on the propensity to fail as a function of diameter and number of connections (Berardi et al. 2008) since they have proved to influence the development of leakages.

The calibration was performed using the flow measurements at the inlet points from the three reservoirs. Five representative days (i.e., working/holiday, winter/summer conditions, and New Year's Day) were chosen to describe the different characteristic hydraulic status of the network. Based on this information and the information on annual demand for each consumer, the DWS named DigitalWaterMass_Calibration was used to identify a unique demand pattern together with the β parameters at a single-pipe level (Berardi & Giustolisi 2021), consistently with pressure the pressure regime based on available data.

The DMAs design is carried out in a two-step process: (1) topological segmentation, that identifies the candidate locations for two types of devices: closed sectioning valves and flow meters and (2) the hydraulic DMAs design, in which a device is assigned at each candidate location. It is worth noting that the location of possible PCVs is left to the user based on engineering judgement and practical constraints; it is recommended that PCV are at the inlet points of the network since they enable more flexibility in WDN operation, also considering the time lag between the design of the DMA and its implementation.

Hydraulic DMAs design is based on a multi-objective optimization that finds the optimal positions for flow meters and closed sectioning valves, considering the conceptual cuts identified by the topological segmentation as candidate locations. The objectives are to maximize the reduction of volumetric losses, while simultaneously minimizing the number of flow meters between DMAs and unsupplied user's demand. If the WDN incorporates PCVs, their set-points are included as decision variables.

The above is accomplished by DigitalWaterDMA_Design and DigitalWaterDMA_Analyzer, that are services used to optimize the design of hydraulic DMAs and visualize the corresponding solutions, respectively. The DWS DigitalWaterDMA_Analyzer allows to select the configuration of actual DMAs in a GIS environment, which is useful to support the final design, allowing the user ad hoc modifications of some positions of gate valves/meters. It is worth noting that removing flow meters means to join contiguous DMA, while preserving the information of the best locations of future meters to create nested DMAs into the existing ones.

Moreover, this DWS allows checking the expected values in flow observation points to evaluate the metrological effectiveness of these devices considering their accuracy and precision to reduce the uncertainty of mass balances in each DMA. For example, placing flow meters where flow inversions or small water velocities are expected should be avoided.

The DWS DigitalWaterRehabilitation supports the choice of rehabilitation plans. This service allows for an assessment of pipe replacement interventions in terms of the expected reduction of volumetric losses, the required investment and other efficiency indicators. This DWS sets the colour of each pipe based on the value of its linear leakage indicator (M1a) as in Figure 7, which enables the user to manually select the pipes to be replaced. Alternatively, external rehabilitation plans may be loaded, if any. Figure 7 displays the interface of this DWS and the expected results for a certain budget for the Modugno WDN, in which the replaced pipes are highlighted in yellow.

Therefore, this DWS allows evaluation of the reduction of leakages for different investment levels, which is aligned with the scope of digital transformation since it enhances and supports decision-making through digital tools that make part of a comprehensive workflow.

This section presents the results obtained after applying the strategy to the Modugno WDN and shows how it can be applied to the larger and more complex WDN of Bari.

The results of the DMAs design in the Modugno WDN are shown in Table 1. The C_WDN coefficient is computed for Modugno WDN before the intervention, while all the other values are evaluated after the DMAs and pressure control. The C_WDN coefficient is the ratio between the M1a indicator and the average pressure in the network. In the case of Modugno WDN, the C_WDN parameter is much greater than 1, which implies that the water losses are mainly due to the deterioration of the infrastructure rather than high-pressure levels (Berardi & Giustolisi 2021). Therefore, in this case, a greater reduction of volumetric water losses is expected with rehabilitation in contrast with DMAs design and pressure control. In this sense, although the DMA design generated a significant reduction of water losses (i.e., around 18%), the M1a indicator is still high.

The results of the DMA design of Bari WDN for each sub-network are shown in Table 2. The total amount of closed sectioning valves is 124, which corresponds to 101 valves inside each sub-network plus 23 additional valves that separate the sub-networks. The only sub-network that was not divided into DMAs was Loseto since the M1a of the calibrated model is 25.3 m³/day/km, which indicated that no interventions were necessary to further reduce leakages. The sub-network with the highest reduction of volumetric water losses is Torre a Mare, that had the lowest C_WDN in its original state. Similarly, the sub-network with the lowest reduction is Japigia, that had the greatest original C_WDN, which implies a significant deterioration level.

The reduction of the volumetric water losses for the whole network does not exactly correspond to the sum of the values for each sub-network because the feeding of a zone between the sub-networks of Torre a Mare and Japigia was modified. The M1a indicator for the whole network of Bari is 77.2 m³/day/km, with a total reduction of volumetric water losses equal to 13,097 m³/day, that corresponds to a decrease of 19.95%.

The rehabilitation plans for Modugno WDN considered a replacement cost of up to 25% of the cost of replacing the system, given the significant magnitude of the C_WDN coefficient. The plans included the pipes with the maximum cost–benefit ratio. In this way, the reduction of the volumetric water losses that can be achieved with an optimized replacement plan of 25% is 4,021 m³/day, with a cost of € 7.05 million for 183 pipes. Note that the reduction of leakages reached with DMAs design and pressure control (see Table 1) may be obtained with an investment of only 8.32% for 20 pipes. Hence, rehabilitation is a necessary task in the Modugno WDN to generate a substantial reduction of volumetric losses, together with DMAs design integrated with pressure control.

The rehabilitation plans for the Bari WDN considered a pipeline replacement cost of up to 20%. Table 3 displays the reduction of volumetric water losses that can be achieved with an optimized solution of around 20% for each sub-network. The greatest predominance of leakage decrease with rehabilitation plans over the DMAs design integrated with pressure control strategy is obtained in the sub-networks with the highest C_WDN coefficient (i.e., Japigia and the City of Bari).

This reduction should be added to that already achieved with the hydraulic DMA design and pressure control. Like the Modugno WDN, rehabilitation is also an essential task in the Bari WDN to obtain a significant leakage decrease and a strategic action for asset management.

This work shows a possible application of the digital transition potential to WDNs asset management. The advantages of using DWSs as a consistent decision-maker support tool for operators have been demonstrated, integrating the use of artificial intelligence, Digital Twins, GIS, and advanced hydraulic analysis at the service of technical skills of water managers, engineers and researchers.

The example WDNs that were presented in this paper illustrate the high coverage potential that digital transformation has and how it can be applied to diverse case studies without excluding factors of size or complexity. At each step of the asset management workflow, the DWSs do not provide a unique solution but expose a set of technically optimal alternatives that the user can easily evaluate, manipulate, and verify to reach the final design configuration. The application on two real networks, as part of a larger planning activity for asset management, proves that the digital water strategy enables rationality, repeatability and flexibility to solve complex and specific technical problems.

This paper was partly financed by European Union – PON Ricerca e Innovazione 2014-2020 E FSC, call D.D. 13 July 2017 n. 1735, Project ‘Energidrica – Energy efficiency in hydraulic networks management’ (code ARS01_00625). Pictures and results from WDNetXL were provided thanks to IDEA-RT s.r.l. (www.idea-rt.com).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.