MWD data Analysis for Risk Assessment and Process Optimization in Tunneling

In drill and blast operations, the collected Measure While Drilling (MWD) data can be used for analysis. In this study we process MWD data with machine learning (ML) methods to demonstrate how real-time risk assessment can be conducted: by predicting the volume of over-excavation, one can aim in reducing risk, lowering project costs, and minimizing environmental impact. The complexity of MWD data makes it necessary to address several challenges before utilizing the data within ML frameworks, namely the three steps (data pre-processing, feature extraction, and normalization) shall take place before further modelling. Here, we discuss the data preparation process, showing how the output from the correlation analysis impacts ML models accuracy. We will show that information from the raw MWD data is preserved and, even, enriched in the correlation analysis. Such information enrichment allows ML models to discover patterns implicitly related to changes in the rock mass conditions in MWD data. By including correlation analysis into the data preparation pipeline, combining it with encoding strategies, fine-tuning procedures, and careful selection of validation metrics, one can dramatically improve the accuracy of ML models in geotechnical applications.