Funded under the 2023 “Transmisiones” call by CDTI–AEI, ACCIONA is coordinating INARTRANS 4.0, “Digital transition towards an advanced industry in artificial intelligence solutions for the transport infrastructure sector,” a project aimed at promoting the use of artificial intelligence solutions in the transport infrastructure sector in order to accelerate digital transformation and increase the competitiveness of the construction industry.

This effort will be applied across the entire infrastructure value chain: construction, operation and maintenance, with a strong focus on information systems and data governance.

The project involves the companies INDRA, AZVI, VIRTUALMECH and JIG, and the technology centers CTCON, TEKNIKER, INTROMAC, UAH and UPM.

But how can AI improve decision-making throughout the entire life cycle of a rail infrastructure asset?

This article presents the solutions being developed by the consortium for each phase in the life of an infrastructure asset.

Construction

First, INARTRANS incorporates and develops new technology that helps to digitalize key aspects during tunnel construction and earthworks associated with railway sections.

Tunnel construction, whether using a TBM or conventional methods, still presents limitations that can be addressed through machine learning and artificial intelligence models. In INARTRANS, using a sufficient volume of real data, the main excavation and geological parameters are analyzed by connecting to the TBM’s real-time database. Out of more than 7,000 variables, 23 have been selected from TBM sensors, along with 4 parameters used to define geotechnical units.

By building regression models (geology–machine), it becomes possible to predict TBM advance in terms of parameters that are relevant for machine control, helping inform the operator’s decisions for optimal progress under safe conditions and within the machine’s operating ranges. The model is trained and calibrated with real tunnel-boring data and compares actual excavation results with the model’s output. This allows the operator to obtain optimal operating values that can be applied while driving the TBM, improving penetration in any type of ground.

Second, for tunnels excavated using conventional methods, a tool is being developed to perform automated geological mapping using only a smartphone. Through image capture and LiDAR scanning, the system generates a numerical output of geomechanical quality.

In addition, based on this result, the over-excavation generated in the tunnel cross-section can be predicted and a blasting plan is recommended according to the optimal values. This tool uses artificial intelligence algorithms for feature recognition in images and machine learning models trained with real data to generate these predictions.

Another disruptive element is the automated construction monitoring system, which combines image processing and sensorization for earthworks. Today, it is possible to install CCTV systems or IP cameras to view work fronts from a control room, but these systems generally include no intelligence and do not allow the use of captured image data — unlike what is already common in other areas such as traffic monitoring or access control in transport stations. INARTRANS will integrate this functionality into its platform to analyze data from earthmoving activities and propose optimized work plans and routes based on real-time and historical data.

Operation and maintenance

Railway operation and maintenance faces a wide range of challenges that are now being addressed with enabling technologies capable of covering longer inspection distances in less time and with high precision. INARTRANS contributes advanced sensors, automated data capture and interpretation, and the generation of digital twins to deliver an in-depth understanding of the condition of both infrastructure and rolling stock.

So, what techniques are applied to achieve this? Below is an overview of the models being implemented:

• AI/ML-based models that improve operations and maintenance by applying intelligent image processing to characterize different parameters of rolling stock and cargo as they pass specific control points along the infrastructure; by determining different rail parameters through the deployment of smart sleepers that provide critical information for operations and maintenance at key locations; and by improving rolling stock positioning, train effective length measurement and train integrity monitoring.

• Hybrid predictive models for catenary evolution: AI/ML algorithms trained on both real datasets (depending on their volume, quality and availability) and synthetic datasets. To generate these synthetic datasets, physical digital twins will be developed based on multibody simulations of vehicles, physical models of the infrastructure, weather conditions, operations and degradation models.

Digital twins, together with the validation of AI/ML algorithms trained using the synthetic data they generate, will help overcome one of the main barriers to introducing AI technologies in predictive maintenance of linear railway assets: the limited quality and volume of historical datasets currently available to rail infrastructure managers, due to the low inspection frequency.

End-to-end data management platform

How is data managed from the moment it is captured by IoT sensors until it is used in digital twins and decision-making dashboards?

To ensure that the results obtained through the application of artificial intelligence models can be directly used by future users, a platform has been designed to control construction, operation and maintenance processes through real-time digital twins and decision dashboards.

The core technological contribution lies in having an IT infrastructure that enables the secure management and exploitation of large volumes of data from multiple external sources, with heterogeneous and complex formats. To achieve this, a common ontology and architecture has been defined to guarantee data messaging, storage, transformation and representation in dashboards or digital applications.

This infrastructure is structured through a multilayer system (Local–Edge and central Cloud node) that is vertically integrated via Communication Middleware (CMV), enabling data processing through ETL engines and dashboards.

All AI-driven solutions under development in this project are deployed through this infrastructure — along with intelligence embedded in the infrastructure itself.