Periodic Reporting for period 1 - STARWARS (STormwAteR and WastewAteR networkS heterogeneous data AI-driven management)
Reporting period: 2023-01-01 to 2024-12-31
The multidisciplinary STARWARS project (STormwAteR and WastewAteR networkS heterogeneous data AI-driven management) aims to address the challenges associated with managing heterogeneous geospatial data by exploiting and developing
artificial intelligence (AI) tools. The solutions developed within this framework are designed to be generic while using wastewater networks as an illustrative application for issues such as data completion, exploiting the complementarity of
multi-source information, and managing the diversity of data formats, based on real data collected. Wastewater networks encompass a wide variety of data that includes geographic information systems (GIS), pipe video inspections, analog or digital maps, as well as textual or .pdf intervention reports. The available information, however, is often fragmented, imprecise, dynamic, and sometimes contradictory, coming from multiple sources. The objective of the STARWARS project is to develop AI-driven solutions capable of representing, managing, merging, explaining, and querying complex data, while considering their diversity and unique characteristics. These solutions are based on innovative models and tools that use logical and graph-based representations of heterogeneous data. Specifically, we aim to represent different data types — such as geographic information systems, ITV inspection videos, and maps — as annotated graphs, while addressing the uncertainty arising from incomplete information, undetected elements (e.g. street names in the videos), or missing crucial data (e.g. manhole details in GIS).
The approaches proposed within the STARWARS framework are designed to integrate and combine uncertain, conflicting, and dynamic data to respond effectively to queries, even when the available information is incomplete or insufficient, while also considering the explainability of results, ensuring that the models incorporate mechanisms to justify their outputs. This project also aims at promoting knowledge exchange between partners of this project. It aims to produce new knowledge and to promote knowledge exchange between EU and non-EU partners, with a strong will and a plan to encourage knowledge sharing between researchers involved in this STARWARS project. Members of this consortium have very complementary skills, key to the realization of this project, in research areas of Water Sciences and Artificial Intelligence.
artificial intelligence (AI) tools. The solutions developed within this framework are designed to be generic while using wastewater networks as an illustrative application for issues such as data completion, exploiting the complementarity of
multi-source information, and managing the diversity of data formats, based on real data collected. Wastewater networks encompass a wide variety of data that includes geographic information systems (GIS), pipe video inspections, analog or digital maps, as well as textual or .pdf intervention reports. The available information, however, is often fragmented, imprecise, dynamic, and sometimes contradictory, coming from multiple sources. The objective of the STARWARS project is to develop AI-driven solutions capable of representing, managing, merging, explaining, and querying complex data, while considering their diversity and unique characteristics. These solutions are based on innovative models and tools that use logical and graph-based representations of heterogeneous data. Specifically, we aim to represent different data types — such as geographic information systems, ITV inspection videos, and maps — as annotated graphs, while addressing the uncertainty arising from incomplete information, undetected elements (e.g. street names in the videos), or missing crucial data (e.g. manhole details in GIS).
The approaches proposed within the STARWARS framework are designed to integrate and combine uncertain, conflicting, and dynamic data to respond effectively to queries, even when the available information is incomplete or insufficient, while also considering the explainability of results, ensuring that the models incorporate mechanisms to justify their outputs. This project also aims at promoting knowledge exchange between partners of this project. It aims to produce new knowledge and to promote knowledge exchange between EU and non-EU partners, with a strong will and a plan to encourage knowledge sharing between researchers involved in this STARWARS project. Members of this consortium have very complementary skills, key to the realization of this project, in research areas of Water Sciences and Artificial Intelligence.
The first activity of the project focuses on the data collection and completion phase, which takes place within Work Package 1. The majority of the data, whether already collected or to be gathered, pertains to wastewater and stormwater networks, the primary case study for our Starwars project. The datasets are available in various formats, including geographical information systems (GIS) (see Figure 1), images, maps (see Figure 2), videos (see Figure 3), texts, and more. Figure 4 provides an overview of the types of data collected or expected to be collected during the project. Alongside the data collected, we developed an ontology to model a hierarchy of concepts related to physical objects within the context of wastewater networks. A screenshot showing part of the ontology, named SewerNet, is displayed in Figure 5.
Within Work Package 2, we explored description logics and graph-based representations to model the data, capturing the interconnections within wastewater networks (see Figure 6). To handle uncertainty and inconsistencies, we applied possibilistic logic and a partial order relation to assertions. Figure 7 summarizes the different representational languages considered for this project.
In Work Package 3, we proposed tractable methods for conditioning and managing conflicting information. For conditioning, we introduced a syntactic reformulation of FH-conditioning that maintains consistency while incorporating new information. For inconsistency management, we developed more efficient characterizations of elect and possibilistic approaches in partially ordered lightweight ontologies. Additionally, we focused on the classification and extraction of wastewater network components from maps and inspection videos to identify manholes, pipelines, flow directions, etc.
In Work Package 4, we addressed query answering and explainability in the context of imperfect, incomplete, or inconsistent knowledge, or knowledge subject to exceptions. The key question explored here is how to define and compute explanations when dealing with uncertain axioms that may allow for exceptions, as opposed to classical description logic knowledge bases, where axioms are fully certain. Furthermore, rather than estimating missing data, we adapted query answering mechanisms by assigning confidence levels based on data availability.
Finally, we use meta-information to enhance query answering, conflict resolution, and explanation generation. These mechanisms include the use of weights assigned to criteria, a partial order applied to assertions within the knowledge base, and a total preorder on defeasible axioms reflecting their typicality or exceptionality. By leveraging these meta-information techniques, we improve the relevance of query responses, facilitate handling inconsistencies, and support the generation of more informative explanations.
The Starwars project website provides further details on the results obtained, along with a list of publications available in open access.
Within Work Package 2, we explored description logics and graph-based representations to model the data, capturing the interconnections within wastewater networks (see Figure 6). To handle uncertainty and inconsistencies, we applied possibilistic logic and a partial order relation to assertions. Figure 7 summarizes the different representational languages considered for this project.
In Work Package 3, we proposed tractable methods for conditioning and managing conflicting information. For conditioning, we introduced a syntactic reformulation of FH-conditioning that maintains consistency while incorporating new information. For inconsistency management, we developed more efficient characterizations of elect and possibilistic approaches in partially ordered lightweight ontologies. Additionally, we focused on the classification and extraction of wastewater network components from maps and inspection videos to identify manholes, pipelines, flow directions, etc.
In Work Package 4, we addressed query answering and explainability in the context of imperfect, incomplete, or inconsistent knowledge, or knowledge subject to exceptions. The key question explored here is how to define and compute explanations when dealing with uncertain axioms that may allow for exceptions, as opposed to classical description logic knowledge bases, where axioms are fully certain. Furthermore, rather than estimating missing data, we adapted query answering mechanisms by assigning confidence levels based on data availability.
Finally, we use meta-information to enhance query answering, conflict resolution, and explanation generation. These mechanisms include the use of weights assigned to criteria, a partial order applied to assertions within the knowledge base, and a total preorder on defeasible axioms reflecting their typicality or exceptionality. By leveraging these meta-information techniques, we improve the relevance of query responses, facilitate handling inconsistencies, and support the generation of more informative explanations.
The Starwars project website provides further details on the results obtained, along with a list of publications available in open access.
In terms of scientific results, several outcomes of this project go beyond the state of the art. The ontology developed in this project for wastewater networks is entirely new. The Sewer Network ontology (SewerNet) defines the structure of wastewater and stormwater networks, their components, and their attributes. The representation of diverse and heterogeneous data related to sewer networks in the form of graphs is a novel contribution that surpasses the current state of the art. Regarding query-answering and explainability, we have proposed methodologies that enable this within the framework of inconsistent or infeasible information. Furthermore, at the end of the project, datasets and ontologies, both multi-source and of various types, will be made available in open access, and will be valuable for the artificial intelligence and data science communities. Publications, along with the data, are being uploaded progressively as they are completed on the project's web page.
With regard to transferring research results to public institutions, particularly wastewater network managers, we have shared some initial findings with relevant stakeholders. While the potential for technology transfer remains to be assessed, we are exploring first-stage demonstrations to further assess the feasibility of our approach within the Starwars project.
At the halfway point of the Starwars project, one of the notable impacts so far is the exchange and sharing of knowledge. We are fortunate to have a multidisciplinary project bringing together researchers in water sciences and artificial intelligence. The various
secondments that have been scheduled have facilitated a transfer of knowledge between these two fields. These knowledge transfers are numerous, ranging from the representation of wastewater networks using geographic information systems (GIS) to ontologies, knowledge representation and computer vision.
With regard to transferring research results to public institutions, particularly wastewater network managers, we have shared some initial findings with relevant stakeholders. While the potential for technology transfer remains to be assessed, we are exploring first-stage demonstrations to further assess the feasibility of our approach within the Starwars project.
At the halfway point of the Starwars project, one of the notable impacts so far is the exchange and sharing of knowledge. We are fortunate to have a multidisciplinary project bringing together researchers in water sciences and artificial intelligence. The various
secondments that have been scheduled have facilitated a transfer of knowledge between these two fields. These knowledge transfers are numerous, ranging from the representation of wastewater networks using geographic information systems (GIS) to ontologies, knowledge representation and computer vision.