Tunnel Overbreak Prediction: An Integrated Approach Using 3D Photogrammetry and Machine Learning

Abstract In civil and mining engineering applications, predicting overbreak is crucial to optimise excavation processes and ensure safety. This study boards on a comprehensive exploration of overbreak prediction-based 3D photogrammetry utilising advanced analytical methods such as principal component analysis (PCA), maximum relevance minimum redundancy (mRMR), and machine learning (ML) regression models, enhanced by generative adversarial networks (GAN) for data augmentation. Initially, correlations and multicollinearity amongst geological and operational variables were investigated. Ten geological parameters from tunnel face mapping were analysed to reveal the causative factor of the blasting mechanism between the rock mass’s complex geological parameters and overbreak. Initially, PCA and mRMR facilitated feature extraction and selection, revealing significant variables influencing overbreak. After the reduction of the dimensionality of the input parameter, the research compares the result of the target and comparative models. Moreover, to address the limited availability of tunnel observations for machine learning, the original 210 datasets of input and target parameters were expanded to 1000 datasets using the GAN method. Subsequently, ML regression models, enriched by GAN-augmented datasets, were employed to unravel the impacts of the selected features on overbreak prediction. Augmenting the dataset fivefold via GANs markedly improved ML regression model efficacy, especially for the ANN model, which exhibited a substantial $${R}^{2}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msup> <mml:mrow> <mml:mi>R</mml:mi> </mml:mrow> <mml:mn>2</mml:mn> </mml:msup> </mml:math> increase from 29 to 96.4% and a 68% reduction in MSE to 0.402E-3 when compared to the original dataset. This robust methodology underscores the relevance of comprehensive feature analysis and data augmentation in improving overbreak prediction in excavation projects, thereby contributing substantively to tunnel excavation.