Wurl, A. M. (2021). A data analysis and maintenance framework for planning tasks of railway infrastructures [Dissertation, Technische Universität Wien]. reposiTUm. https://doi.org/10.34726/hss.2022.100321
E194 - Institut für Information Systems Engineering
Number of Pages:
Data Analytics; Data Integration; Industrial Data Quality; Data Analysis; Robust Regression; Feature Selection; Predictive Asset Management; Obsolescence Management; Digital Companion; Railway Infrastructure
The increasing extension of railway networks worldwide entails highly complex settings of signaling systems. For maintenance planning tasks, decision support approaches are required as analyzing data of differently structured sources may be extremely time consuming. Although existing data analytics techniques have proven to be able to extract and analyze large amounts of data in different fields of application, in the railway domain more advanced approaches are needed to tackle an interplay of complex industrial data settings. This includes aggregating data from different formats, identifying ambiguities, and visualizing most important information. To support maintenance decisions, a modern approach has to be able to process data in a coherent and consistent manner while minimizing the time required for interventions. In this industrial thesis, we present a procedural approach that combines new techniques for processing configuration and operational data of signal systems to support decision-making in maintenance tasks. Since configuration and operation data reveal different data structures and formats, in the first step the approach ensures high quality data integration into a data warehouse. As storing duplicate or contradicting information may have business-critical effects, an interactive technique provides an efficient process where the user resolves ambiguous data classifications. Once data is stored in the data warehouse, information of hardware components and its properties, i.e., features, can be used as variables. This allows the following technique to build a regression model to estimate the quantity of components based on a set of input features, but also ensures the identification of relevant features related to a hardware component. The resulting regression model is combined with a stochastic model to predict the number of hardware components needed for existing and planned systems in the future. Instead of showing plain prediction results, we propose advanced visualizations to support technical engineers to quickly grasp all important information including the uncertainty of the prediction. Extending the mere predictions, we propose the concept of a digital companion which prescriptively recommend maintenance actions in system configurations. All our findings have been evaluated in continuous collaboration with experts of the railway domain. Ultimately, the techniques developed have been integrated into the railway business which confirms the relevance and usefulness of our work.