Periodic Reporting for period 2 - DAYDREAMS (Development of prescriptive AnalYtics baseD on aRtificial intElligence for iAMS)
Berichtszeitraum: 2022-06-01 bis 2023-05-31
According to the International Union of Railways the length of tracks maintained by the European railway sector exceeds 300.000 km operating more than 5 billion train-kilometres and offering services for more than 400 billion passenger-kilometres. A steady increase is expected for the next 30 years making railways a key-asset in the European transportation ecosystem. Railway systems are expected to increase their share in transportation by expanding their geographical reach and deliver innovative and integrated travel solutions for people and goods meeting the highest service standards in terms of safety and security. Although European Railways remain the safest in the world, according to data reported in 2017 by the ERA, there have been on average just over 2000 significant accidents each year on the railways of the EU Member States. The economic impact of these accidents has been estimated in the order of €1.61 billion for 2015.
To address railway safety challenges, UN Member States agreed on a set of measures promoting sustainable development ensuring access to safe, affordable, accessible and sustainable transport systems for all citizens by 2030. Besides the UN, the EC also released a new European Mobility Package setting a target for zero traffic fatalities and severe injuries by 2050. To achieve these goals, enhancement of the existing decision support systems with advanced data analytics and mathematical modelling tools are expected to play a key role. These tools can be used not only to predict future issues but also to provide solutions for preventing and solving them, proposing actions to enhance safety and reducing maintenance costs.
The Shift2Rail JU has established in its Multi Annual Action Plan that for “delivering the capabilities to bring about the most sustainable, cost-efficient, high-performing, time-driven, digital and competitive customer-driven transport mode for Europe,” among other characteristics, intelligent maintenance should be introduced to increase capacity and availability and to reduce maintenance costs. The S2R JU also identifies, among the key enabling technologies, machine learning (ML), artificial intelligence (AI) and big data analytics targeting predictive and possibly prescriptive maintenance in S2R demonstrators.
The overall DAYDREAMS objective was to move forward the integration and use of data and artificial/human trustworthy intelligence together with context-driven HMI for prescriptive Intelligent Asset Management Systems (IAMS) in railway by (i) advancing in maintenance approach towards prescriptive asset management, (ii) improving the decision-making process by developing multi-objective decision optimisation approaches taking into account all implications of IAMS decisions in the railway environment, and (iii) reinforcing the role of the person-in-the-loop by designing and developing advanced context-driven HMIs to allow context- and risk-aware multiple-options decision-making processes.
The HMI will allow the person-in-the-loop to: (i) properly access and visualise predictions/metrics and models, (ii) assess why and how the model predicts something, (iii) Steer models by setting parameters, and (iv) evaluate alternatives using parameter steering and extending this process through speculative execution.
To address railway safety challenges, UN Member States agreed on a set of measures promoting sustainable development ensuring access to safe, affordable, accessible and sustainable transport systems for all citizens by 2030. Besides the UN, the EC also released a new European Mobility Package setting a target for zero traffic fatalities and severe injuries by 2050. To achieve these goals, enhancement of the existing decision support systems with advanced data analytics and mathematical modelling tools are expected to play a key role. These tools can be used not only to predict future issues but also to provide solutions for preventing and solving them, proposing actions to enhance safety and reducing maintenance costs.
The Shift2Rail JU has established in its Multi Annual Action Plan that for “delivering the capabilities to bring about the most sustainable, cost-efficient, high-performing, time-driven, digital and competitive customer-driven transport mode for Europe,” among other characteristics, intelligent maintenance should be introduced to increase capacity and availability and to reduce maintenance costs. The S2R JU also identifies, among the key enabling technologies, machine learning (ML), artificial intelligence (AI) and big data analytics targeting predictive and possibly prescriptive maintenance in S2R demonstrators.
The overall DAYDREAMS objective was to move forward the integration and use of data and artificial/human trustworthy intelligence together with context-driven HMI for prescriptive Intelligent Asset Management Systems (IAMS) in railway by (i) advancing in maintenance approach towards prescriptive asset management, (ii) improving the decision-making process by developing multi-objective decision optimisation approaches taking into account all implications of IAMS decisions in the railway environment, and (iii) reinforcing the role of the person-in-the-loop by designing and developing advanced context-driven HMIs to allow context- and risk-aware multiple-options decision-making processes.
The HMI will allow the person-in-the-loop to: (i) properly access and visualise predictions/metrics and models, (ii) assess why and how the model predicts something, (iii) Steer models by setting parameters, and (iv) evaluate alternatives using parameter steering and extending this process through speculative execution.
During the whole life of the project, the following objectives were achieved:
• Definition of the requirements specific for each scenario (HITACHI, RFI, STIMIO and TRAINOSE) and the related KPIs for the prototype evaluation and model validation;
• Collection of the main datasets for the four scenarios;
• Study and design of the Prescriptive Analytics (PA) and Multi-Objective Optimisation (MOO) models;
• Analysis of the state-of-the-art and design of the HMI to be specialised for each scenario;
• Implementation at intermediate level of the PA and MOO components;
• Definition of the integration guidelines and integration of the developed TRL 4 components into the intermediate (I-REL) prototype;
• Complete evaluation and validation of the I-REL prototype and assessment of the pipeline of the implemented models;
• Definition of the integration guidelines and integration of the developed TRL 5 components into the final (F-REL) prototype;
• Implementation in the system of the Blockchain-based session tracking functionality;
• Complete evaluation and validation of the F-REL prototypes;
• Assessment of the sensitivity and robustness of the IAMS prototype;
• Definition of the way forward for industrial exploitation.
• Definition of the requirements specific for each scenario (HITACHI, RFI, STIMIO and TRAINOSE) and the related KPIs for the prototype evaluation and model validation;
• Collection of the main datasets for the four scenarios;
• Study and design of the Prescriptive Analytics (PA) and Multi-Objective Optimisation (MOO) models;
• Analysis of the state-of-the-art and design of the HMI to be specialised for each scenario;
• Implementation at intermediate level of the PA and MOO components;
• Definition of the integration guidelines and integration of the developed TRL 4 components into the intermediate (I-REL) prototype;
• Complete evaluation and validation of the I-REL prototype and assessment of the pipeline of the implemented models;
• Definition of the integration guidelines and integration of the developed TRL 5 components into the final (F-REL) prototype;
• Implementation in the system of the Blockchain-based session tracking functionality;
• Complete evaluation and validation of the F-REL prototypes;
• Assessment of the sensitivity and robustness of the IAMS prototype;
• Definition of the way forward for industrial exploitation.
The DAYDREAMS objectives have been assessed by validating the proposed solution using the following two-step approach: (i) A first validation at TRL 4 of the approaches (developed prescriptive asset management) using several internal scenarios and one scenario provided by the CFM IN2SMART2 project; (ii) Given the promising results of the developments at TRL4, a further validation of the DAYDREAMS methodologies integrated in a TRL 5 prototype using all the scenarios validated at TRL4, instead of the two that were originally planned as minimum objective. This decision has broadened the scope of the project and therefore allowed a more consistent evaluation and validation of the results. The validation covered both the performances of the prototypes and its trust for future adoption in multi-actors environments.
The validation was carried out by defining evaluation and validation metrics and KPIs: (i) linked to asset management problems to be solved by the involved Infrastructure Managers (IMs) and related baselines, (ii) quantified and measurable; (iii) referred to high-level KPIs defined in the S2R IMPACT2 CFM project, and (iv) useful to address multi-objective optimisation.
DAYDREAMS innovated by proposing a leapfrog step introducing prescriptive analytics and AI-based multi-objective optimisation combined with context-driven HMI for decision support in railway asset management processes. This, combined with increased awareness about the sensitivity and robustness of these decisions, has offered a trustable and trackable system able to be used in a context with heterogeneous vendors and data providers. DAYDREAMS has therefore been able to improve maintenance planning and to offer a more proactive maintenance decision making involving not only the maintenance but also other entities within the organisation like the traffic control and the energy management departments in the process.
The innovation potential surrounding the DAYDREAMS prototype implementation is the provision of a holistic approach and environment that allows the validation and testing of scenarios for asset maintenance, considering the diversity of sources, approaches, and components. The proposed approach has allowed to accelerate the testing of new scenarios and AI models combining existent scenarios and algorithms and being more cost and time sustainable. Therefore, the overall capacity of IMs to learn how to use and adopt AI, collect feedback, and refine and adapt solutions has allowed to improve and to increase the quality of the final design. Furthermore, the user’s trust in AI models and algorithms has been increased as they have been made accessible through HMI leveraging further innovations within the railway context. Moreover, the overall user experience has been improved as IMs are now capable of thoroughly verifying more scenarios in a shorter time, reaching more optimal decisions.
The validation was carried out by defining evaluation and validation metrics and KPIs: (i) linked to asset management problems to be solved by the involved Infrastructure Managers (IMs) and related baselines, (ii) quantified and measurable; (iii) referred to high-level KPIs defined in the S2R IMPACT2 CFM project, and (iv) useful to address multi-objective optimisation.
DAYDREAMS innovated by proposing a leapfrog step introducing prescriptive analytics and AI-based multi-objective optimisation combined with context-driven HMI for decision support in railway asset management processes. This, combined with increased awareness about the sensitivity and robustness of these decisions, has offered a trustable and trackable system able to be used in a context with heterogeneous vendors and data providers. DAYDREAMS has therefore been able to improve maintenance planning and to offer a more proactive maintenance decision making involving not only the maintenance but also other entities within the organisation like the traffic control and the energy management departments in the process.
The innovation potential surrounding the DAYDREAMS prototype implementation is the provision of a holistic approach and environment that allows the validation and testing of scenarios for asset maintenance, considering the diversity of sources, approaches, and components. The proposed approach has allowed to accelerate the testing of new scenarios and AI models combining existent scenarios and algorithms and being more cost and time sustainable. Therefore, the overall capacity of IMs to learn how to use and adopt AI, collect feedback, and refine and adapt solutions has allowed to improve and to increase the quality of the final design. Furthermore, the user’s trust in AI models and algorithms has been increased as they have been made accessible through HMI leveraging further innovations within the railway context. Moreover, the overall user experience has been improved as IMs are now capable of thoroughly verifying more scenarios in a shorter time, reaching more optimal decisions.