Towards TBM Automation: On-The-Fly Characterization and Classification of Ground Conditions Ahead of a TBM Using Data-Driven Approach

Pre-tunneling exploration for rock mass classification is a common practice in tunneling projects. This study proposes data-driven approach that allows classification. Two machine learning (ML) models, namely random forest (RF) and extremely randomized tree (ERT), are employed to classify the conditions encountered Pahang-Selangor Raw Water Tunnel Malaysia using tunnel boring (TBM) operating parameters. Due imbalance of classes distribution, an oversampling technique was used obtain balanced training dataset unbiased ML models. A five-fold cross-validation tune model hyperparameters validation-set evaluation. ERT achieved overall accuracy 95%, while RF 94% accuracy, rightly classifying conditions. The result shows proposed has potential identify correctly ground TBM, which early problem detection on-the-fly support system selection based on identified condition. study, part ongoing effort towards developing reliable models could be incorporated into TBMs, approaches ahead TBM allow construction problems.