Periodic Reporting for period 2 - RAILS (Roadmaps for A.I. integration in the raiL Sector)
Periodo di rendicontazione: 2021-10-01 al 2023-06-30
The Shift2Rail project RAILS addressed the topic “Artificial Intelligence for the railway sector”.
The project aimed to understand the potential of AI in railways, contribute to future research roadmaps, and address challenges associated with safety, dependability, security, and autonomy.
The main objectives included: identifying AI potential, aligning with ongoing railway innovation, recognizing necessary shifts, developing proofs-of-concept, creating benchmarks and simulations, outlining transition pathways, engaging relevant stakeholders, training young researchers.
To achieve these objectives, RAILS followed a Technology Road-Mapping Methodology, integrating research activities into comprehensive roadmaps for safety and automation (WP2), maintenance and inspection (WP3), and traffic planning and management (WP4).
The project assessed scientific, industrial, and regulatory landscapes, proposed pilot case studies, and outlined guidelines for applying AI approaches from other sectors.
Figure 1 provides a high-level view of the RAILS scope and research.
In conclusion, the roadmapping activities revealed that the railway industry is embracing AI but faces challenges such as the lack of standards, insufficient datasets, the need for innovative solutions like digital twins, testing methodologies rooted in mixed reality, and guidelines for the gradual introduction of AI in autonomous driving. The developed roadmaps provide insights and recommendations to guide future research, proposing innovative ideas, fostering advancements in railway technology, and facilitating consensus-building to bridge the gap between AI opportunities and their full exploitation in the railway sector.
The project aimed to understand the potential of AI in railways, contribute to future research roadmaps, and address challenges associated with safety, dependability, security, and autonomy.
The main objectives included: identifying AI potential, aligning with ongoing railway innovation, recognizing necessary shifts, developing proofs-of-concept, creating benchmarks and simulations, outlining transition pathways, engaging relevant stakeholders, training young researchers.
To achieve these objectives, RAILS followed a Technology Road-Mapping Methodology, integrating research activities into comprehensive roadmaps for safety and automation (WP2), maintenance and inspection (WP3), and traffic planning and management (WP4).
The project assessed scientific, industrial, and regulatory landscapes, proposed pilot case studies, and outlined guidelines for applying AI approaches from other sectors.
Figure 1 provides a high-level view of the RAILS scope and research.
In conclusion, the roadmapping activities revealed that the railway industry is embracing AI but faces challenges such as the lack of standards, insufficient datasets, the need for innovative solutions like digital twins, testing methodologies rooted in mixed reality, and guidelines for the gradual introduction of AI in autonomous driving. The developed roadmaps provide insights and recommendations to guide future research, proposing innovative ideas, fostering advancements in railway technology, and facilitating consensus-building to bridge the gap between AI opportunities and their full exploitation in the railway sector.
The first steps for all roadmaps were done in WP1, by setting the scientific, industrial, and regulatory landscapes through a review of the scientific literature, the examination of the ongoing work in European and overseas projects, the analysis of the EU regulation and directives at the date, and a survey among the railway stakeholders on the challenges and the state-of-practice of the usage of AI in railways.
Following the outcomes of the first phase of the project, two pilot case studies were suggested for each technical WP to investigate innovative and strategic applications of AI and a Proof-of-Concept (PoC) has been developed for each case study. The PoCs were conceived as benchmarks to experiment with AI solutions and identify promising directions but also limits and issues to be faced.
The six pilot case studies (and related PoCs) proposed in RAILS are the following:
• WP2 CS1: “Vision-based Obstacle Detection on Rail Tracks”.
• WP2 CS2: “Cooperative Driving for Virtual Coupling of Autonomous Trains”.
• WP3 CS1: “Smart Maintenance at Level Crossings”.
• WP3 CS2: “AI-based Rolling Stock Rostering”.
• WP4 CS1: “Primary Delay Prediction”.
• WP4 CS2: “Incident Attribution Analysis”.
For each of the above case studies, AI-powered approaches have been proposed and results have been produced through the experimental PoCs. They are reported in the project's deliverables.
The overall activities have produced valuable knowledge and pinpointed specific research directions toward intelligent control, autonomous trains, enhanced safety, and optimized operations
Figure 2 provides a mapping between the road-mapping steps, the project outcomes, and related deliverables, the outcomes include the identification of gaps and innovation needs, research directions, and recommendations.
Numerous dissemination activities have been undertaken, encompassing the publication of papers, presentations at both national and international conferences and events, and the organization of workshops. Two datasets from the Proof of Concepts (PoCs) are available on the Zenodo platform for widespread dissemination and exploitation. In support of establishing a robust research network on AI in railways, a dedicated series of workshops, AI4RAILS, has been initiated and successfully conducted.
RAILS also addressed the training of two Ph.D. students to support the research capacity in AI within the rail sector across Europe by involving research institutions with a combined background in both computer science and transportation systems.
The RAILS Consortium consists of four educational organizations. Hence, exploitation of the project's results is expected to foster new project proposals and positively influence both teaching and future research activities within the academic community. The outcomes from the research activities will be seamlessly integrated into academic courses at the Master's degree level and will serve as key topics for PhD or Master's theses. Furthermore, the public datasets made available on Zenodo as part of the project results will not only benefit the project itself but will also be open for broader utilization.
Following the outcomes of the first phase of the project, two pilot case studies were suggested for each technical WP to investigate innovative and strategic applications of AI and a Proof-of-Concept (PoC) has been developed for each case study. The PoCs were conceived as benchmarks to experiment with AI solutions and identify promising directions but also limits and issues to be faced.
The six pilot case studies (and related PoCs) proposed in RAILS are the following:
• WP2 CS1: “Vision-based Obstacle Detection on Rail Tracks”.
• WP2 CS2: “Cooperative Driving for Virtual Coupling of Autonomous Trains”.
• WP3 CS1: “Smart Maintenance at Level Crossings”.
• WP3 CS2: “AI-based Rolling Stock Rostering”.
• WP4 CS1: “Primary Delay Prediction”.
• WP4 CS2: “Incident Attribution Analysis”.
For each of the above case studies, AI-powered approaches have been proposed and results have been produced through the experimental PoCs. They are reported in the project's deliverables.
The overall activities have produced valuable knowledge and pinpointed specific research directions toward intelligent control, autonomous trains, enhanced safety, and optimized operations
Figure 2 provides a mapping between the road-mapping steps, the project outcomes, and related deliverables, the outcomes include the identification of gaps and innovation needs, research directions, and recommendations.
Numerous dissemination activities have been undertaken, encompassing the publication of papers, presentations at both national and international conferences and events, and the organization of workshops. Two datasets from the Proof of Concepts (PoCs) are available on the Zenodo platform for widespread dissemination and exploitation. In support of establishing a robust research network on AI in railways, a dedicated series of workshops, AI4RAILS, has been initiated and successfully conducted.
RAILS also addressed the training of two Ph.D. students to support the research capacity in AI within the rail sector across Europe by involving research institutions with a combined background in both computer science and transportation systems.
The RAILS Consortium consists of four educational organizations. Hence, exploitation of the project's results is expected to foster new project proposals and positively influence both teaching and future research activities within the academic community. The outcomes from the research activities will be seamlessly integrated into academic courses at the Master's degree level and will serve as key topics for PhD or Master's theses. Furthermore, the public datasets made available on Zenodo as part of the project results will not only benefit the project itself but will also be open for broader utilization.
The road-mapping activities performed in RAILS contribute to shaping the future research directions for the design and implementation of smarter railways, including the identification of the needed conceptual shift, current criticalities, and recommendations for safe and effective integration of AI (in particular, Machine Learning and Deep Learning techniques).
Through the assessment of the current state-of-the-art, the methodological approaches and the experiments developed for the PoCs as well as the surveys conducted among railways stakeholders and experts, RAILS defined and provided roadmaps for the following research directions:
- Fully Autonomous Trains in Open Environments;
- Obstacle Detection Through On-board Cameras and Artificial Vision;
- Intelligent Audio-Video Technologies for Non-intrusive Infrastructure Inspection;
- Intelligent Digital Twins (DT) for Predictive Maintenance of Railway Assets;
- Train Delay Prediction Using Graph Embedding;
- Railway Incident Attribution Analysis Using Big Data Analytics;
- Mixed-Reality Technologies to Support AI Testing (Figure 3);
- Datasets sharing for Benchmarking AI Technologies for Railways (Figure 4).
These research outcomes are expected to contribute to future research endeavors and generate new possibilities and ideas for both railway stakeholders and the broader research
community. By delineating research directions for forthcoming projects and emphasizing the need for strengthened collaborations among railway stakeholders in the realm of AI,
RAILS results have the potential to exert a profound socio-economic impact and broader societal implications.
Through the assessment of the current state-of-the-art, the methodological approaches and the experiments developed for the PoCs as well as the surveys conducted among railways stakeholders and experts, RAILS defined and provided roadmaps for the following research directions:
- Fully Autonomous Trains in Open Environments;
- Obstacle Detection Through On-board Cameras and Artificial Vision;
- Intelligent Audio-Video Technologies for Non-intrusive Infrastructure Inspection;
- Intelligent Digital Twins (DT) for Predictive Maintenance of Railway Assets;
- Train Delay Prediction Using Graph Embedding;
- Railway Incident Attribution Analysis Using Big Data Analytics;
- Mixed-Reality Technologies to Support AI Testing (Figure 3);
- Datasets sharing for Benchmarking AI Technologies for Railways (Figure 4).
These research outcomes are expected to contribute to future research endeavors and generate new possibilities and ideas for both railway stakeholders and the broader research
community. By delineating research directions for forthcoming projects and emphasizing the need for strengthened collaborations among railway stakeholders in the realm of AI,
RAILS results have the potential to exert a profound socio-economic impact and broader societal implications.