Abstract
The railway is one of the most critical transportation infrastructures in the world. Efficient facility management (FM) of railway tracks is essential to ensure the safety, reliability, and cost-effectiveness of railway networks. However, the current FM practices face significant challenges due to fragmented data systems, reactive maintenance strategies, and the lack of interoperability between diverse data sources. These limitations often lead to delays, increased operational costs, and safety risks. Additionally, traditional data management standards, such as the Industry Foundation Classes (IFC) schema, fail to address the specialized, dynamic requirements of railway track FM. This research proposes an integrated framework for optimizing railway track FM, achieved through three interconnected studies that aim to overcome the current limitations by utilizing openBIM, extending the IFC schema, and incorporating deep learning models. These three studies together contribute to the development of a unified decision-support framework that improves the efficiency and effectiveness of railway track FM.Railway track FM involves complex and data-intensive tasks, such as regular inspections, maintenance, and resource allocation. However, these activities are hindered by the fragmentation of data, reliance on manual inspections, and a lack of integration across different FM systems. Thus, the first step of this research addresses the challenge of fragmented data and the reactive nature of maintenance by proposing an openBIM-based decision-support conceptual framework. The study begins by identifying key limitations in current FM practices through case studies and interviews with FM stakeholders. A significant gap identified is the lack of a unified digital platform to integrate data from automated inspections, historical records, and real-time monitoring systems. To address the identified gaps, the study proposes a conceptual framework that combines these disparate data sources into a single, interoperable platform. The framework supports real-time monitoring, proactive maintenance planning, and enhanced decision-making by integrating defect data, resource allocation records, and environmental factors. The proposed conceptual framework was validated through focus group interviews and comparative reviews, demonstrating its ability to improve maintenance efficiency, reduce costs, and enhance safety by shifting from reactive to predictive maintenance strategies.
Moreover, although openBIM and the IFC schema have been widely applied in building and construction projects, their use in railway track FM has been limited. In this case, the second step of this research addresses the shortcomings of the current IFC schema, which lacks the capability to capture the dynamic and specialized data required for railway track FM. A conceptual model, using Unified Modeling Language (UML), is developed to classify FM data into four core aspects: Component, Action, Resource, and Operation (CARO). The proposed extension of the IFC schema incorporates new entities and property sets to better represent track defects, maintenance actions, resource allocation, and operational conditions. The extended schema was validated through a case study on a high-speed railway segment, demonstrating that it enhances data integration, supports more efficient maintenance planning, and improves interoperability across various stakeholders. This study's contribution strengthens the framework by providing a standardized, interoperable data model for railway track FM.
Furthermore, accurate prediction of track deterioration is crucial for optimizing maintenance schedules and minimizing disruptions. Traditional models for predicting track degradation often fail to account for complex relationships among operational, environmental, and inspection data. To address the limitations of existing track deterioration prediction models, the third step of this study proposes a hybrid deep learning model that combines Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and an attention mechanism to predict the Track Quality Index (TQI), a key indicator of track condition by segment. The model integrates historical inspection data, maintenance logs, environmental factors and structural information to capture both spatial and temporal dependencies in track deterioration. The methodology was applied to real-world railway track data, and the results demonstrated that the hybrid model significantly outperforms traditional approaches, offering more accurate predictions. This study contributes to the framework by providing an advanced machine learning-based tool for proactive maintenance, enhancing the overall decision-support system.
The three steps of this study presented in this research collectively contribute to the development of an integrated decision-support framework for optimizing railway track FM. The first study establishes an openBIM-based platform for data integration and proactive decision-making. The second study extends the IFC schema to accommodate the dynamic and specialized needs of railway track FM, ensuring better interoperability and data management. The third study introduces a deep learning model for predicting track deterioration, which enhances maintenance scheduling and resource allocation. Together, these studies provide a comprehensive, data-driven approach to improving railway track FM, shifting from reactive to predictive maintenance strategies, and optimizing resource allocation. Future work should focus on integrating other emerging technologies such as digital twins, reinforcement learning methods for FM plan optimization and LLM-based models for decision support to further enhance the capabilities of railway track FM systems.
| Date of Award | 7 Aug 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Jung In KIM (Supervisor) & Xin LI (Supervisor) |