Scholarlink

The accurate prediction of tunnel boring machine (TBM) performance is critical for effective project planning, cost management, and utilization factor estimation. This chapter aims to develop predictive models for the field penetration index (FPI) using post-hoc interpretability (PHI) and explainable artificial intelligence (XAI), considering the main theory of the UT model (University of Tehran, proposed by Hassanpour et al., Tunn Undergr Space Technol 26:595–603, 2011). A database was gathered from Iranian hard-rock tunnelling projects, covering a wide range of lithological and engineering geological units. Data were collected from laboratory tests, daily TBM operating records, and TBM data loggers, resulting in 156 data points recorded per stroke and geological unit. Unsupervised learning (principal component analysis, PCA) and hybrid-optimized supervised learning models within an interpretable machine learning (IML) framework were developed to overcome the black-box limitations of traditional ML approaches. XAI techniques were also applied to assess the results of FPI prediction using a dataset with a 156 × 7 structure, enabling sensitivity analysis for each input variable and classification based on defined engineering geological and lithological units. To mitigate overfitting associated with the small dataset, network-specific optimization algorithms were used to enhance weight updating and model generalization. Both XAI and IML analyses showed strong correlations between the predicted and measured values of FPI, as well as within each classified lithological unit. The hybrid learning approach outperformed conventional training methods, showing superior predictive accuracy and interpretability. The proposed hybrid and interpretable modelling framework provides an effective solution for small-sample, high-dimensional problems in tunnel engineering and addresses the black-box nature of ML models through interpretable and PHI-based analysis. The developed models can be applied to future hard-rock tunnelling projects in similar lithological units.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of Machine Learning (ML)-Based Algorithms in Hard Rock TBM Performance Analysis Using Explainable Artificial Intelligence (XAI)

  • 章节

  • 发表日期：

    2026年

查看全文
  • Hanan Samadi,
  • Jafar Hassanpour,
  • Jamal Rostami

摘要

The accurate prediction of tunnel boring machine (TBM) performance is critical for effective project planning, cost management, and utilization factor estimation. This chapter aims to develop predictive models for the field penetration index (FPI) using post-hoc interpretability (PHI) and explainable artificial intelligence (XAI), considering the main theory of the UT model (University of Tehran, proposed by Hassanpour et al., Tunn Undergr Space Technol 26:595–603, 2011). A database was gathered from Iranian hard-rock tunnelling projects, covering a wide range of lithological and engineering geological units. Data were collected from laboratory tests, daily TBM operating records, and TBM data loggers, resulting in 156 data points recorded per stroke and geological unit. Unsupervised learning (principal component analysis, PCA) and hybrid-optimized supervised learning models within an interpretable machine learning (IML) framework were developed to overcome the black-box limitations of traditional ML approaches. XAI techniques were also applied to assess the results of FPI prediction using a dataset with a 156 × 7 structure, enabling sensitivity analysis for each input variable and classification based on defined engineering geological and lithological units. To mitigate overfitting associated with the small dataset, network-specific optimization algorithms were used to enhance weight updating and model generalization. Both XAI and IML analyses showed strong correlations between the predicted and measured values of FPI, as well as within each classified lithological unit. The hybrid learning approach outperformed conventional training methods, showing superior predictive accuracy and interpretability. The proposed hybrid and interpretable modelling framework provides an effective solution for small-sample, high-dimensional problems in tunnel engineering and addresses the black-box nature of ML models through interpretable and PHI-based analysis. The developed models can be applied to future hard-rock tunnelling projects in similar lithological units.