ABSTRACT
The tunneling industry is exploding with data gathered from various sources, like sensors installed at TBMs, site equipment and periphery, but also from secondary sources like tunnel information models (TIM) and third-party project data. Sensor data is structured data in digital form, used for documentation, analysis and reports, diagnosis of failure and design of future projects. Secondary data is often stored in unstructured form and hence unfit for data science while still highly relevant for predictive analysis. Manual analysis of these vast amounts of data is time intensive and calls for automated data analysis. Data science methods are useful tools to analyze the data and make predictions to increase operational efficiency, improve maintenance, safety aspects and prevent downtimes during the tunneling operation in an automated approach. However, it is also crucial to understand the underlying behavior of the data under different conditions, like changing ground conditions or fault zones, as well as the influence of the human factor. Ensuring high quality and consistency of different data sources becomes a challenging task e.g. for the performance of machine learning models. This paper summarizes the lessons learned on a data science project in TBM tunneling and it describes the benefits and challenges as well as a roadmap for data science in the tunneling industry. It will be crucial to combine the knowledge of different domains, like mechanical engineering, civil engineering, information technology, data engineering and data science for new applications and use cases in TBM tunneling. Further challenges in human resources like an aging technical work force, limited availability of personnel with the necessary skillsets and limited awareness of the required cultural changes, are hampering the progress in the field of data science in TBM tunneling.
