An Integration Framework to Support Data Sharing and Process Tracking in Intelligent Asset Management Systems

DAYDREAMS, which completed its activities in May 2023, is a project within Shift2Rail’s 3rd Innovation Programme (IP3). The project’s overall objective was to advance the integration and use of data and artificial/human trustworthy intelligence for Intelligent Asset Management Systems (IAMS) in the railway domain. The adoption of an integrated Intelligent Asset Management System is highly relevant in the digitalisation process of railway companies, and different solutions can be considered according to specific needs [2]. The DAYDREAMS project focused on a set of maintenance scenarios proposed by three railway maintenance stakeholders to implement intelligent asset management solutions such as (i) prescription based on asset status forecasting (prescriptive analytics, PA), (ii) optimisation of maintenance intervention plans considering different metrics and constraints (multi-objective optimisation, MOO), (iii) contextdriven interfaces to enhance human decision-making (human-machine interface, HMI).

This paper reports on the design and demonstration of an IAMS Support Integration Framework to enable the integration of different digital artifacts for an intelligent maintenance scenario in a multi-stakeholder environment. In this direction, the complementary objectives of the framework are: (i) to enhance data sharing to support better the reuse of solutions developed by different stakeholders, e.g., a prediction model for a specific type of asset may be reused, and (ii) to introduce a trusted way of tracking and auditing the adoption of digital solutions within maintenance process, e.g., to identify users and software components that failed to predict a breakdown. A prototype, implementing the IAMS Support Integration Framework for the three DAYDREAMS scenarios, is discussed and evaluated to demonstrate the proposed solution.

The SHIFT2RAIL¹ Innovation Program 3 promoted the development of solutions for intelligent maintenance in the context of railway asset management [9]. In this framework, DAYDREAMS tries to make a step forward (leveraging the knowledge and results of IN2RAIL, IN2DREAMS, IN2SMART and IN2SMART2 projects²), proposing a holistic approach where the information on the current and future status of the assets (e.g., actual faults to repair and probability of news faults) together with the estimated information on the maintenance (e.g., costs and duration), are exploited to deliver options to decisionmakers on how to schedule, improve, and reduce the impact of the maintenance with the purpose to improve the infrastructure maintenance management.

The implementation of such a solution requires the collection of data from different data sources and the implementation of specific software components for data analytics (PA, MOO) and visualisation (HMI). Moreover, an effective deployment in an integrated system is needed for users to adopt it within maintenance operations. A significant challenge is associated with multiple stakeholders involved in maintenance processes, e.g., asset manufacturers, maintenance supervisors, and maintenance teams. Moreover, the definition of digital solutions supporting asset management introduces a set of additional stakeholders to be considered, e.g., sensor manufacturers, vendors of data collection solutions, and data scientists developing models for prediction and prescription. Each stakeholder may contribute or use different digital artifacts and should be made accountable for its involvement in the overall maintenance process. Finally, other actors often have similar needs for implementing intelligent maintenance solutions but adopt different or non-interoperable software solutions, thus slowing down the digitalisation process.

To streamline and enhance the integration, two fundamental sets of guidelines were considered in the design of the proposed framework. On the one hand, the FAIR principles (Findability, Accessibility, Interoperability, Reusability) defined in the context of research data management [10], can be adapted to consider different digital artifacts, from software components [6] to machine learning models [4]. The principles emphasize: (i) the importance of structured metadata for facilitating the findability of digital artifacts, (i) the publication of digital artifacts online for accessibility, (iii) the adoption of best practices and standards to facilitate the interoperability of the solutions, and (iv) the need for precise documentation and legal terms of access and usage to support reusability. On the other hand, the Machine Learning Operations (MLOps) guidelines [7] were considered to promote automation in the development and deployment cycle of software components for prediction, prescription and optimisation.

Adopting a shared catalogue of digital artifacts [5] allows actors belonging to different roles to publish and manage descriptions of web services, datasets, and software components promoting their findability and reusability. In the DAYDREAMS project, we focused on how to effectively implement a shared catalogue for intelligent asset maintenance by identifying the relevant digital artifacts and associated requirements. Moreover, while many works in the literature focus on platforms for the development of machine learning models [1], we focused on how to enable the deployment and integrated execution of such models and related software components to support the end users. Finally, considering previous results from the IN2DREAMS project on the usage of blockchain technologies for auditability purposes in the railway domain [8], we define a seamless solution to keep track of operations associated with an intelligent maintenance scenario.

To design a solution supporting the implementation of Intelligent Asset Management Systems, we focused first on the elicitation of the technical requirements associated with it. Integration means first of all the identification or development of the relevant digital artifacts (e.g., datasets, machine learning models, software components) and the availability of a runtime environment for the deployment. Then, it means allowing proper communication between components, which have been developed independently, and guaranteeing their integrated execution according to well-defined processes, i.e., sequence of operations. Finally, the execution of processes must be monitored to ensure that all the stakeholders involved in developing and using the components acted appropriately.

The architecture designed for the IAMS Integration Support Framework, shown in Fig. 1, is based on three macro-functionalities defined to support the elicited requirements: (i) a shared catalogue to enhance the findability and reusability of digital artifacts (IAMS Shared Catalogue); (ii) a runtime area for the integrated deployment and execution of artifacts associated with intelligent maintenance scenarios (IAMS Runtime Area); and (iii) a blockchain-based system to handle the monitoring and tracking of artifacts, processes and operations performed by users (IAMS Process Tracking).

Stakeholders can store and describe in the shared catalogue their digital artifacts, as well as Intelligent Maintenance Packages (IAM Package) defined to group artifacts for integrated execution. The catalogue supports the configuration of metadata profiles for the consistent description of artifacts, and lifecycle processes defining their governance (e.g., process to request/grant access to an artifact). Additionally, it implements programmatic access to the artifacts, allowing stakeholders to develop advanced functionalities (e.g., creation or update of artifacts at runtime). We identified four basic digital artifact types: Dataset, Software Components, Machine Learning Model, and Maintenance Plan. We defined a metadata schema for each of them relying on the generic ADMS-AP specification [3] and on more specific controlled vocabularies targeting each artifact type. The runtime environment enables stakeholders to deploy software components needed for the integrated execution of IAM Packages in a secure and integrated environment. Software components should be containerised³ and expose well-documented interfaces (e.g., HTTP API documented through OpenAPI⁴) to facilitate their deployment and integration with other components. Finally, the tracking functionality allows stakeholders to monitor the lifecycle of digital artifacts, tracking it together with the runtime operations performed by users. Each lifecycle and runtime interaction is transparently registered as an operation, and a blockchain-based infrastructure is leveraged to provide additional trust among the involved stakeholders. This information can also be saved within the shared catalogue as an Intelligent Maintenance Session (IAM Session) artifact and exported for auditing and reproducibility purposes, ensuring transparency and accountability in managing digital artifacts.

Bild vergrößern

The IAMS prototype, developed by DAYDREAMS and shown in Fig. 2, demonstrates the implementation of the IAMS Integration Support Framework and its application considering three intelligent maintenance scenarios proposed by railway stakeholders. The digital artifacts are described in the IAMS Shared Catalogue by different users, and an IAM Package is defined for each scenario grouping together the relevant artifacts. Each IAM Package comprises real datasets provided by railway stakeholders to address a proposed maintenance scenario, innovative software components and machine-learning models for PA e MOO developed by the project, and a dedicated HMI designed by UI experts. The users can access the catalogue through a dedicated web interface and access all the metadata of the digital artifacts. All the lifecycle operations associated with digital artifacts are tracked and can be visualised through the catalogue. The containerised software components are automatically deployed in the IAMS Runtime Area in isolated sub-networks for each IAM Package. The IAMS Prototype functionalities are implemented transparently by a set of dedicated components, i.e., without requiring a modification of the integrated PA and MOO components. The IAMS Gateway handles the authentication and redirects the requests made by users interacting with the HMI. Additional functionalities enabling the tracking of runtime operations and the implementation of advanced integration with the shared catalogue are implemented by dedicated Package Handler components. In the IAMS prototype, we demonstrated the automatic update of the metadata of a machine learning model every time a training operation is performed, as well as the usage of the shared catalogue to store and retrieve maintenance plans generated and modified by different stakeholders. The IAMS Process Tracking component receives all the tracked operations and registers them through the blockchain-based infrastructure to ensure trust among various stakeholders interacting with the system. The user can retrieve the list of operations performed in each IAM Session and trace back all steps taken during the runtime interaction with the digital artifacts in an IAM Package. All the operations returned to the user are validated through the blockchain, and an alert is produced if the validation is unsuccessful.

Bild vergrößern

The implemented prototype successfully supported the demonstration of the final project results within the three practical use cases for railway maintenance considered. It demonstrated the sharing of digital artifacts through structured descriptors and a common catalogue, the integrated deployment of the software components for end-users to visually inspect and interact with maintenance predictions and prescriptions, and the adoption of a blockchain-based solution to track and audit operations in a trusted way.

Moreover, we performed an initial assessment of the prototype by showcasing its capabilities to six stakeholders from three distinct railway-related companies specialized in infrastructure management and intelligent maintenance solutions. This assessment aimed to gather their insights through interviews regarding the technical feasibility and business potential of implementing this solution within their organizations. The participating stakeholders recognized the benefits of integrating similar functionalities into their operations to enhance the reusability of digital artifacts and foster trust among involved parties. All stakeholders indicated their interest in potentially adopting a similar solution in the near future, with an average rating of 3.5 out of 4. Notably, one stakeholder highlighted the importance of applying these principles to external entities (e.g., external service/product providers) and various units and geographical divisions within large corporations. Regarding potential enhancements, the stakeholders stressed the need to ensure the tool’s user-friendliness for individuals with diverse technical backgrounds and to meet stringent security requirements for handling proprietary and confidential digital artifacts.

The paper presented an IAMS Integration Support Framework to foster the sharing and reuse of digital artifacts for different intelligent maintenance scenarios, and to enable process tracking mechanisms guaranteeing trust in a multistakeholder environment. The discussion of the IAMS Prototype, developed within the DAYDREAMS project, demonstrated the application of the proposed framework to a set of practical intelligent maintenance scenarios involving different digital artifacts and processes. The implemented prototype and its evaluation by stakeholders highlight how the proposed framework can effectively support the integration of digital artifacts for a generic intelligent maintenance scenario and the potential interest for adoption by companies.