MERCURIO, an Innovative Monitoring System Employing Multisource Data for the Safety of Railway Infrastructure

In 2020, the global space economy has increased of 4.4% compared to a revised 2019 total of $428 billion with a total value of $447 billion, a number that has increased of approximately 55% compared to ten years ago, with a predominance of the commercial segment (“The Space Report 2021 Q2”). In this framework, the performance of the earth observation market and its services is particularly significant. In 2021, the global turnover of Earth Observation (EO) data and value-added services amounts to €2.8 billion (“EUSPA-2022 EO and GNSS Market Report”). In 2023, the global revenues from EO data and value-added services amounted to €3.4 billion and almost half of this total is generated by the top three segments: Agriculture, Climate, Environment and Biodiversity and Urban Development and Cultural Heritage (with projections that suggest significant growth in the Insurance and Finance segment). In a decade, the overall revenues of the global EO data and value-added services market are projected to reach nearly €6 billion. The global EO data market is expected to see a CAGR of 5% by 2033, resulting in total revenues of almost €1 billion while the EO value-added services market, considerably larger, has generated a total of around €2.8 billion in 2023. Starting from 2023, the EO value-added services market is expected to experience a CAGR of almost 6%, achieving total revenues of almost €5 billion by 2033 (“EUSPA-2024 EO and GNSS Market Report”). It's in the framework of space value-added services that the rail sector plays a fundamental role between the ones of always more growing interest towards EO services employment. As the “EUSPA-2024 EO and GNSSMarket Report” states, in the rail sector, the major stakeholders include government agencies and infrastructuremanagement or private railway operators and manufacturers. In this application context, GNSS and EO-basedsolutions could help ensure increased safety of railway infrastructure while allowing reduced costs for on-siteinspections. Nowadays, various regions across Europe are modernising their rail infrastructure also in the opticsof sustainability. In this wake, the European Green Deal aims to invest almost €90 billion in rail-relatedinfrastructure upgrades. Due to the increasing need of having always more ready measures and monitoring ofsome events on the line, EO and the use of drones as well as of in situ sensors could help improving thereliability of rail infrastructure management and predictive as well as proactive measures to ensure safety forusers. Within the Multi-Annual Work Program of Europe's Rail Joint Undertaking (EU-Rail) is underlined theexplicit indication for the creation of intelligent railway infrastructure asset management systems. In the FlagshipArea 3, explicit reference is made to the ability to apply remote observation technologies as one of thedevelopment directions. It’s in this European strategic scenario that the idea and technological objectives of the MERCURIO (acronymderived from the Italian “SisteMa di monitoraggio intElligente e multipiattafoRma per la siCURezzadell'Infrastruttura ferrOviaria”) project fully fits. MERCURIO is a research project whose aim is fitting into thetransformation process towards a concept of smart railways which, thanks to the intensive use of ArtificialIntelligence applied to the processing of heterogeneous data, can create a digital twin of the asset (over the lastten years, the growing interest in the application of Artificial Intelligence in the railway sector has resulted inaround 50 projects at European and international level, with a 70% share earmarked for the development of newtechnological solutions dedicated to maintenance activities and infrastructure inspection). MERCURIO project willdevelop an innovative multi-platform monitoring system to support the inspection and maintenance process ofregional railway infrastructures in Campania (Italy) through the integration into a single decision support tool of allthe results of an advanced processing chain that will lead to a modular and scalable data fusion strategy throughthe intensive use of Artificial Intelligence techniques. The project takes on board the needs to provide regionalrailway infrastructures with a flexible tool that responds to the specific criticalities of the territories on which theyinsist, optimising maintenance procedures, thus preventing, in a timely manner, critical situations that may affectpassenger safety and the costs dictated by line stoppages and on-site inspection interventions, especially forsome use-cases of interest such as: • Ground movement analysis; • Soil moisture trends in the vicinity of the railway network; • Monitoring of civil works; • Monitoring of tunnels and galleries. To guarantee the response to heterogeneous problems, MERCURIO will implement a modular approach tointegrate heterogeneous data into a single platform: satellite data, distributed sensors and sensors on drones, willbe collected in an open architecture, to allow the evolution of the entire monitoring system based on technologicaladvances and the needs of particular use cases in the optics of the use of innovative asset monitoringtechniques. The acquired data will be suitably analysed through machine learning algorithms with the dualobjective of exploring possible correlations potentially useful for the definition of risk reduction strategies andidentifying the presence of areas of greater danger. In particular several techniques will be used: • unsupervised machine learning techniques, such as clustering algorithms that will allow the subdivision of datainto groups to highlight similarities and differences for the different data categories. Clustering algorithms willmake possible to perform a multivariate analysis by grouping data of different nature (such as mechanicalcharacterisation of soils, hydrological data, moisture data, vegetation data, etc.) into a number of similarityclasses. The different classes are associated with different degrees of criticality and visualised through a colourindex (green-yellow-red). Soft-type algorithms allow the attribution of the same data to several clusters, assigningthe data a probability of belonging to a specific class. This characteristic makes these algorithms useful for betterdefining the boundaries of zones of different criticality. The different types of algorithms will be implemented in thePython language. Since the clustering analysis is sensitive to the number and type of data, a stability analysis ofthe solutions will be performed, whereby the algorithms will be applied to different datasets, taking into accountdifferent scenarios; • supervised machine learning techniques, such as artificial neural networks (ANN) that will allow the definition offuture hazard scenarios based on the training of the available data; the algorithms will have to learn to identifyareas with several degrees of risks and assign a degree of warning to monitor a railway line. During the trainingphase, the algorithms will be provided with known cases, each of which will be accompanied by factors and variables characterising the event and will generate hundreds of mathematical models that will differ in therandom combination of those variables. The ‘model validation’ will consist of an iterative process in which theperformance of the models in correctly classifying cases that are known but never entered into the algorithmduring the training phase, will be measured. The integration between Artificial Intelligence and data derived from the several sources will then make it possibleto implement a Geospatial Artificial Intelligence (GeoAI) model that can be defined as the collection, analysis andexploitation of images and information to describe, evaluate and virtually represent the physical characteristicsand geographical activities detected on the territory. The monitoring system, depending on the different scenariosanalysed, will produce a series of data that are extremely heterogeneous in terms of type, frequency andsignificance. Through the data fusion operation, the intention is to combine several elements with the aim ofgenerating integrated and objective information to understand the evolution of phenomena and allow the bestquantitative characterisation of the conditions in the analysed areas. The objective is to constitute a data set ofinformation on which the project WebGis/DSS will be based. The monitoring system that will be set up for theMERCURIO project brings an intrinsic complexity linked to the heterogeneity of the data sources, the produceddata format and the information contained therein. Specifically, with regard to the information extracted frommulti-source data sources, these can be directly correlated with each other if they refer to the same entity, suchas temperature, height, etc. or they can be correlated with each other by indices generated through appropriateprocessing of multisource data. Furthermore, for a correct fusion and interpretation of the data generated by thesystem, there is a need for a careful study of the areas investigated to allow the elaboration of a context model,which contemplates the geological, environmental and infrastructural characteristics. The process of datahomogenisation, fusion and correlation will be followed by an information management phase through the use ofa WebGis/DSS platform. In conclusion, by employing the methods and tools explained earlier in this abstract, the MERCURIO platformcould reach some functional and research goals in order to make the platform be used to monitor some alreadyidentified use cases that nowadays represent the main criticalities in terms of safety for the regional railwayinfrastructure and lead the way in the employment of such system in a wider range of applications, possibly alsoin other sectors of interest.