The Watering IoTs (WITS) project focuses on the use of IoT for monitoring and optimizing Water Supply Systems, by combining short and long term decisions based on continuous and massive data collection, signal on network processing and AI analysis.
WITS key objectives are:
- Smart water IoT infrastructure design;
- Massive IoT (through LPWAN IoT technologies) design to accommodate an efficient collection of data from devices scattered in the WSS;
- Use graph signal processing to provide new and efficient AI methods;
- Smart contracts definition.
WITS is part of Spoke 8 – Intelligent and Autonomous Systems
Project PI: Francesca Cuomo
WP3
We developed an innovative approach based on graph theory to optimize the placement of gateway concentrators within smart water distribution networks. This method is based on classifying network nodes according to their centrality, using advanced graph theory tools to identify optimal concentrator positions. Additionally, we applied Graph Signal Processing (GSP) to model pressure values as signals on graphs. The primary objective is to establish a data fusion approach where both radio and application data are employed consciously to enhance application performance. Through this technique, we were able to identify strategic node locations, maximizing network coverage and reducing infrastructure redundancy. This approach allowed us to optimize the number of required gateways, significantly reducing the overall energy consumption of the system. To validate our methodology, we applied these techniques to realistic scenarios, using simulation tools such as EPANET to model hydraulic properties and ns-3 to simulate radio properties. These simulations enabled us to test the effectiveness of the proposed system on large-scale IoT networks.
We started a research activity on the development of a digital twin for water distribution networks. The first phase of the activity involves a comprehensive study of the existing state of the art. We proceed to the definition of platform requirements. This platform is envisioned to support multiple streams, including real data from the physical water distribution network, simulation data from the digital twin, and machine learning algorithms. The goal is to create a system that can leverage all these data sources effectively, providing valuable insights for decision-making and predictive maintenance.
WP4
We extended some results of the WP4 by leveraging the use of topological signal processing to model data in higher-order networks, focusing on node signals for pressure and edge signals for water flow. We introduced tools like graph signal processing and cell-complex techniques to represent water flows, factoring in user demand. A learning framework for reconstructing pressures and flows in water distribution networks is proposed, validated with real networks, with the aim to use fewer samples. The study extends to a dynamic model addressing water demands and leak detection. A leakage detection algorithm is developed, showing improved leak identification using cell complexes. The approach demonstrates topology-based learning's potential in effectively monitoring WDNs.
WP5
In the WP5 we have studied the methodologies underlying the disaggregation based on on-Intrusive Load Monitoring (NILM) in the electric field and by translating these studies also in the water management field. n WP5, we developed a Multi-AppliaNce TRAnsformer (MANTRA) model for NILM, enabling scalable estimation of individual water usage from total household consumption. MANTRA captures long-term dependencies and outperforms other deep learning models in multi-target disaggregation across various time resolutions.
We studied the reconstruction of nodal pressures in water distribution systems using Graph Neural Networks (GNNs). We used an already know GNN that was first trained in a controlled environment, and then perturbed with sensor noise and randomness to assess its robustness under real-world conditions. Inspired by the methodologies based on Graph Laplacian Denoising were added to improve the GNN's robustness. After extensive hyperparameter tuning, a modified approach was proposed for better reconstruction of unobserved junction pressures in water networks. This method will be applied to leak detection.
We developed an innovative approach based on graph theory to optimize the placement of gateway concentrators within smart water distribution networks. This method is based on classifying network nodes according to their centrality, using advanced graph theory tools to identify optimal concentrator positions. Additionally, we applied Graph Signal Processing (GSP) to model pressure values as signals on graphs. The primary objective is to establish a data fusion approach where both radio and application data are employed consciously to enhance application performance. Through this technique, we were able to identify strategic node locations, maximizing network coverage and reducing infrastructure redundancy. This approach allowed us to optimize the number of required gateways, significantly reducing the overall energy consumption of the system. To validate our methodology, we applied these techniques to realistic scenarios, using simulation tools such as EPANET to model hydraulic properties and ns-3 to simulate radio properties. These simulations enabled us to test the effectiveness of the proposed system on large-scale IoT networks.
We started a research activity on the development of a digital twin for water distribution networks. The first phase of the activity involves a comprehensive study of the existing state of the art. We proceed to the definition of platform requirements. This platform is envisioned to support multiple streams, including real data from the physical water distribution network, simulation data from the digital twin, and machine learning algorithms. The goal is to create a system that can leverage all these data sources effectively, providing valuable insights for decision-making and predictive maintenance.
WP4
We extended some results of the WP4 by leveraging the use of topological signal processing to model data in higher-order networks, focusing on node signals for pressure and edge signals for water flow. We introduced tools like graph signal processing and cell-complex techniques to represent water flows, factoring in user demand. A learning framework for reconstructing pressures and flows in water distribution networks is proposed, validated with real networks, with the aim to use fewer samples. The study extends to a dynamic model addressing water demands and leak detection. A leakage detection algorithm is developed, showing improved leak identification using cell complexes. The approach demonstrates topology-based learning's potential in effectively monitoring WDNs.
WP5
In the WP5 we have studied the methodologies underlying the disaggregation based on on-Intrusive Load Monitoring (NILM) in the electric field and by translating these studies also in the water management field. n WP5, we developed a Multi-AppliaNce TRAnsformer (MANTRA) model for NILM, enabling scalable estimation of individual water usage from total household consumption. MANTRA captures long-term dependencies and outperforms other deep learning models in multi-target disaggregation across various time resolutions.
We studied the reconstruction of nodal pressures in water distribution systems using Graph Neural Networks (GNNs). We used an already know GNN that was first trained in a controlled environment, and then perturbed with sensor noise and randomness to assess its robustness under real-world conditions. Inspired by the methodologies based on Graph Laplacian Denoising were added to improve the GNN's robustness. After extensive hyperparameter tuning, a modified approach was proposed for better reconstruction of unobserved junction pressures in water networks. This method will be applied to leak detection.
As a key outcome of the WITS project, we have initiated the design of a comprehensive system in which a real water distribution network placed near Rome, outfitted with LoRaWAN devices, will be linked to the WITS platform and incorporated into the RESTART ecosystem. This endeavor will be amplified within the framework of the RESTART digital twin smart grand challenge. The activities are supported by UNIPA and UNIDATA partners.
A notable outcome of our efforts within the project is the successful attainment of institutional membership in the LoRa Alliance by our CNIT partner, marking a significant step forward in our collaborative endeavors.
WP3
We developed a comprehensive IoT integration framework to enhance the monitoring and optimization of Water Distribution Systems. This framework includes components for hydraulic analysis identification, graph signal processing application, strategic placement of measurement points, and a simulation environment, with a particular focus on a LoRaWAN network for controlled system testing and refinement. Moreover, the additional tasks of the WP3 explores the amalgamation of smart orchestration, workload control, and edge computing to manage water network complexities. By adjusting data stream flow based on application features and utilizing edge computing, it aims to boost the overall performance and reliability of water distribution systems while facilitating timely decision-making.
WP3
Motivated by remarkable advancements in non-intrusive load monitoring (NILM), we have explored the potentiality of deep learning by proposing a neural network for near real-time multi-appliance water disaggregation. Positive results across various sampling intervals prompt discussion on sparse data challenges. Overall, such investigation aims to explore deep learning for water disaggregation as a powerful tool to monitor, manage and save water resources more effectively in the residential sector.
WP4
Based on recent literature on topological signal processing (TSP), we have developed a novel method to characterize interacting data related to water distribution networks. TPS aims to study signals associated not only with graph nodes (as classical graph signal processing) but with any order structures in order to model in a more appropriate way the complex data interactions. In this context, we propose a novel framework to simultaneously reconstruct flows and pressure values in a water distribution networks that is also able to detect water leakages.
Full conceptualization, implementation and analysis of graph-based modeling and signal processing to monitor a water distribution network in efficient and accurate ways. Identification of a flow reconstruction algorithm and of the best node ranking on the graph in order to save energy in case there is the need to reduce the number of measurements in the network still keeping a given level of accuracy. Testing of the proposed approach in numerical simulations performed in a realistic LoRaWAN WDS scenario. We demonstrate that the water flow is accurately reconstructed by strategic placement of a reduced number of sensors, leading to significant energy savings, i.e. around 73%.
SOCIETAL IMPACT:
The proposed approach for designing and implementing smart water managemtn and smart contracts for smart water distribution networks has significant societal impact. It enhances resource management by ensuring efficient and equitable water distribution and promotes sustainability by identifying and mitigating inefficiencies. Additionally, it empowers local communities to manage their water resources actively, leading to more responsive and tailored solutions. The tool also offers economic benefits by reducing operational costs and water bills, and encourages innovation and technological advancement in water management, driving further research and development.
Papers:
Domenico Garlisi, Gabriele Restuccia, Ilenia Tinnirello, Francesca Cuomo and Ioannis Chatzigiannakis, “Real-Time Leakage Zone Detection in Water Distribution Networks: A Machine Learning-based Stream Processing Algorithm”, International Symposium on Algorithmic Aspects of Cloud Computing (ALGOCLOUD 2023), Amsterdam 05/09/2023
Redemptor Laceda Taloma, Jr., Danilo Comminiello (Sapienza University of Rome, Italy) Patrizio Pisani (Unidata SpA, Italy) Francesca Cuomo (University of Rome Sapienza, Italy), "UNet-WD: Deep Learning for Multi-Appliance Water Disaggregation", IFIP 1st International Workshop on Smart Water Management (SmartWater) 2024, Salonicco, June 2024
Dimitrios Amaxilatis (Spark Works Ltd., Ireland), Tiziana Cattai (Sapienza University of Rome, Italy), Antonino Pagano (University of Palermo, Italy),Ioannis Chatzigiannakis (Sapienza University of Rome, Italy), Redemptor Laceda Taloma, Jr. (Sapienza University of Rome, Italy), Domenico Garlisi (University of Palermo & CNIT Italian National Consortium for Telecommunications, Italy), Themistoklis Sarantakos (Spark Works Ltd., Ireland), Varvara Vythoulka (University of Patras, Greece), Christos Zaroliagis (University of Patras, Greece), "A Tool to Facilitate the Design of Smart Contracts in Smart Water Distribution Networks", IFIP 1st International Workshop on Smart Water Management (SmartWater) 2024, Salonicco, June 2024
A notable outcome of our efforts within the project is the successful attainment of institutional membership in the LoRa Alliance by our CNIT partner, marking a significant step forward in our collaborative endeavors.
WP3
We developed a comprehensive IoT integration framework to enhance the monitoring and optimization of Water Distribution Systems. This framework includes components for hydraulic analysis identification, graph signal processing application, strategic placement of measurement points, and a simulation environment, with a particular focus on a LoRaWAN network for controlled system testing and refinement. Moreover, the additional tasks of the WP3 explores the amalgamation of smart orchestration, workload control, and edge computing to manage water network complexities. By adjusting data stream flow based on application features and utilizing edge computing, it aims to boost the overall performance and reliability of water distribution systems while facilitating timely decision-making.
WP3
Motivated by remarkable advancements in non-intrusive load monitoring (NILM), we have explored the potentiality of deep learning by proposing a neural network for near real-time multi-appliance water disaggregation. Positive results across various sampling intervals prompt discussion on sparse data challenges. Overall, such investigation aims to explore deep learning for water disaggregation as a powerful tool to monitor, manage and save water resources more effectively in the residential sector.
WP4
Based on recent literature on topological signal processing (TSP), we have developed a novel method to characterize interacting data related to water distribution networks. TPS aims to study signals associated not only with graph nodes (as classical graph signal processing) but with any order structures in order to model in a more appropriate way the complex data interactions. In this context, we propose a novel framework to simultaneously reconstruct flows and pressure values in a water distribution networks that is also able to detect water leakages.
Full conceptualization, implementation and analysis of graph-based modeling and signal processing to monitor a water distribution network in efficient and accurate ways. Identification of a flow reconstruction algorithm and of the best node ranking on the graph in order to save energy in case there is the need to reduce the number of measurements in the network still keeping a given level of accuracy. Testing of the proposed approach in numerical simulations performed in a realistic LoRaWAN WDS scenario. We demonstrate that the water flow is accurately reconstructed by strategic placement of a reduced number of sensors, leading to significant energy savings, i.e. around 73%.
SOCIETAL IMPACT:
The proposed approach for designing and implementing smart water managemtn and smart contracts for smart water distribution networks has significant societal impact. It enhances resource management by ensuring efficient and equitable water distribution and promotes sustainability by identifying and mitigating inefficiencies. Additionally, it empowers local communities to manage their water resources actively, leading to more responsive and tailored solutions. The tool also offers economic benefits by reducing operational costs and water bills, and encourages innovation and technological advancement in water management, driving further research and development.
Papers:
Domenico Garlisi, Gabriele Restuccia, Ilenia Tinnirello, Francesca Cuomo and Ioannis Chatzigiannakis, “Real-Time Leakage Zone Detection in Water Distribution Networks: A Machine Learning-based Stream Processing Algorithm”, International Symposium on Algorithmic Aspects of Cloud Computing (ALGOCLOUD 2023), Amsterdam 05/09/2023
Redemptor Laceda Taloma, Jr., Danilo Comminiello (Sapienza University of Rome, Italy) Patrizio Pisani (Unidata SpA, Italy) Francesca Cuomo (University of Rome Sapienza, Italy), "UNet-WD: Deep Learning for Multi-Appliance Water Disaggregation", IFIP 1st International Workshop on Smart Water Management (SmartWater) 2024, Salonicco, June 2024
Dimitrios Amaxilatis (Spark Works Ltd., Ireland), Tiziana Cattai (Sapienza University of Rome, Italy), Antonino Pagano (University of Palermo, Italy),Ioannis Chatzigiannakis (Sapienza University of Rome, Italy), Redemptor Laceda Taloma, Jr. (Sapienza University of Rome, Italy), Domenico Garlisi (University of Palermo & CNIT Italian National Consortium for Telecommunications, Italy), Themistoklis Sarantakos (Spark Works Ltd., Ireland), Varvara Vythoulka (University of Patras, Greece), Christos Zaroliagis (University of Patras, Greece), "A Tool to Facilitate the Design of Smart Contracts in Smart Water Distribution Networks", IFIP 1st International Workshop on Smart Water Management (SmartWater) 2024, Salonicco, June 2024
Our project's innovation is significantly enhanced by the active participation of our industrial partners. Unidata SpA brings a powerful infrastructure for collecting real LoRaWAN measurements for smart metering, providing accurate and reliable data essential for effective water management. Thanks to Unidata we get in touch with Acqua Pubblica Sabina SpA who has contributed valuable insights into their operational needs and challenges, ensuring that our solutions are tailored to real-world applications and are highly practical for end-users.
Our project's innovation is significantly enhanced by the active participation of our industrial partners. Unidata SpA brings a powerful infrastructure for collecting real LoRaWAN measurements for smart metering, providing accurate and reliable data essential for effective water management. Thanks to Unidata we get in touch with Acqua Pubblica Sabina SpA who has contributed valuable insights into their operational needs and challenges, ensuring that our solutions are tailored to real-world applications and are highly practical for end-users.
Our project's innovation is significantly enhanced by the active participation of our industrial partners. Unidata SpA brings a powerful infrastructure for collecting real LoRaWAN measurements for smart metering, providing accurate and reliable data essential for effective water management. Thanks to Unidata we get in touch with Acqua Pubblica Sabina SpA who has contributed valuable insights into their operational needs and challenges, ensuring that our solutions are tailored to real-world applications and are highly practical for end-users.
Program partners:
- Università di Catania
- Università di Roma Sapienza
- Consorzio Nazionale Interuniversitario per le Telecomunicazioni (CNIT)
WITS project took part in/organized important dissemination events:
- Redemptor Jr Taloma (PhD student) participated as speaker in the Live Webinar Servizi a Rete “Progettazione di soluzioni IoT per reti idriche intelligenti” on April 27th 2023, discussing the applications of machine learning in the literature about smart water management.
- Two public talks given in Paris, at CNAM, by Tiziana Cattai, “Graph model for Water Distribution Networks with IoT applications” and by Francesca Cuomo “Towards Edge Computing in LoRaWAN: new architectural models and future applications”.
- Talk in Rome at the AEIT 2023 conference by Tiziana Cattai, “A graph based method for efficiently monitoring of water supply systems”
- WITS researchers are organizing 1st International Workshop on Smart Water Management (SmartWater) in IFIP/IEEE Networking 2024, to be held in Thessaloniki, Greece, June 3-6, 2024
- R. Taloma participated to “Individuare le perdite idriche con l'IA?” for podcast “Tutto Connesso”
- R. Taloma delivered a 5-minute interview about the RESTART Grand Challenge “Digitalizzare l’ambiente per un mondo più sostenibile”
- WITS researchers organized the 1st International Workshop on Smart Water Management (SmartWater) in IFIP/IEEE Networking 2024, held in Thessaloniki, Greece, June 3-6, 2024 (https://networking.ifip.org/2024/index.php/workshops/smartwater).
- R. Taloma participated to “Individuare le perdite idriche con l'IA?” for podcast “Tutto Connesso”.
- Francesca Cuomo gave a Keynote entitled “Unlocking IoT Potential: Empowering LoraWAN for Secure, Distributed Smart Water Management” at the IFIP NETWORKING Conference in Salonicco (Greece) in June 2024
- Publications
Expected: at least 9 publications on 36 months
Accomplished: 2 (2 conference publications)
Readiness level: 66% - Joint Publications
Expected: >=30% joint publications on 36 months
Accomplished: 2 joint publications over 3
Readiness level: 66% - Talks/Communication events
Expected: 15 talks or event chairing/organizing within WITS activities on 36 months
Accomplished: 6 (among dissemination events and conference presentations)
Readiness level: 100% - Demo/PoC
Expected: 1 PoCs expected by the end of the project
Accomplished: 0
Readiness level: 0% (work according to plan) - Project Meetings
Expected: > 36 meetings
Accomplished: 20 meetings
Readiness level: 55% - Personnel Recruitments
Expected: 1 RTD-A
Accomplished: 1 RTD-A
Readiness level: 100%
Milestones:
- First year report including project dissemination activities and the delivery of D1 and D2 (due date: M12)
- Second year report including project dissemination activities and the delivery of D3 and D4 (due date: M24)
- Final report including project dissemination activities and the delivery of D5 and D7 (due date: M36)
- D1 - Deliverable on Smart Water Supply Systems (due date: M5)
- D2 - Deliverable on Massive IoT Access (due date: M12)
- D3- Deliverable on Continuum data collection and smart network orchestration (due date: M24)
- D4- Deliverable on Sensed processing using graph signal processing (due date: M17)
- D5- Deliverable on Development of utilities/smart contracts for users and operators (due date: M27)
- D6 - Performance assessment – Case Studies (due date: M36)
Researchers involved: We expect about 63 PM as overall. We already spent about 36 PM up to M21.
Collaboration proposals
Ioannis Chatzigiannakis and Domenico Garlisi are preparing a proposal titled “Driving the Digital Transformation of Water Supply Systems for Sustainability in the Next Generation and Beyond” (WaterBeyond) to be submitted to the HORIZON-MSCA-2024-DN-01-01 call.
The proposal aims to provide the required expertise and effort in research and training to a workforce of 15 Doctoral Candidates (DCs) in water management technologies and data analysis challenges at the frontier of the future ICT network sector.
The WITS project is open to cooperation with experts as well as nonprofit organizations in the following areas:
- water network degradation prevention;
- water consumption in precision agriculture;
- water network monitoring in emergency scenarios;
- water distribution system incorporating self-managing features;
- water demand forecasting.
WITS received a first collaboration proposal by University of Gabes-Tunisia, for the formation of a consortium for the application to the PRIMA project (Main topic: Sustainable Water Management). The project is seeking for other partners interested in this Consortium.
For any proposal of collaboration within the project please contact the project PI.