Scholarlink

<p>Reliable characterization of subsurface conditions, such as depth to bedrock, lithological variability, and the presence of fractures or faults, is critical for designing safe and durable foundations. However, geophysical methods remain underutilized in geotechnical design due to limited interdisciplinary collaboration, a general lack of awareness among engineers, and the absence of standardized integration frameworks. Hence, this study addresses these gaps by integrating geophysical and geotechnical data with machine learning (ML) techniques to enhance subsurface integrity modeling for foundation designs. Conducted in a granitic terrain in Minden, Pulau Penang, Malaysia, the research utilized electrical resistivity tomography, induced polarization, seismic refraction tomography, and standard penetration test (SPT-N) data acquired along two transects. The findings revealed critical subsurface anomalies—such as fractures, weathered zones, and lithological transitions—that pose potential risks to structural stability. Three major lithological layers were delineated: saturated clayey/silty topsoil extending to the residual profile; a transitional zone comprising weathered (clayey/silty) and fractured rock; and partially weathered to hard bedrock. ML algorithms applied included K-nearest neighbors (R² = 0.930 training, 0.905 testing), simple linear regression (R² = 0.898 training, 0.905 testing), principal component analysis, and K-means clustering (Silhouette score = 0.6688, residual sum of squares of 30.87 at K<sub>m</sub> = 3), all demonstrating strong predictive performance. The integrated framework proved effective for delineating subsurface lithology and assessing geomechanical properties, offering a scalable, data-driven strategy for cost-effective, high-resolution foundation design. This approach supports improved engineering decisions, reduces reliance on extensive invasive testing, and highlights the transformative potential of ML-assisted geophysical–geotechnical investigations in infrastructure development.</p> Graphical Abstract <p>This study presents a machine learning (ML)-assisted modeling approach that integrates Seismic Refraction Tomography (SRT), Electrical Resistivity Tomography (ERT), Induced Polarization (IP), and Standard Penetration Test (SPT-N) data to evaluate subsurface integrity in the granitic terrain of Minden, Penang Island, Penang, Malaysia. Geophysical surveys (SRT, ERT, and IP) and borehole SPT-N data were collected along two transects in the study area (see Panels 1 and 2). These datasets were integrated, and the individual geophysical models and borehole logs (Panel 3) were used to delineate subsurface lithologies and weak zones. Supervised ML techniques, including K-Nearest Neighbors (KNN) at K<sub>n</sub> = 5 and Simple Linear Regression (SLR) (Panel 4), accurately predicted subsurface properties, as validated by actual versus predicted plots and corresponding performance metrics summarized in Panel 5. In contrast, the unsupervised K-means clustering algorithm (Panel 6) effectively classified the subsurface into three distinct lithological clusters, supported by density plots and average data points within each cluster. Additionally, the relationship between resistivity and chargeability was modeled and validated (Panel 7), reinforcing the reliability of the geophysical interpretation. Overall, the study demonstrates a robust, ML-enhanced framework for evaluating subsurface integrity in complex geological settings.</p>

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

Integrated Machine Learning Modeling of Seismic, Electrical Resistivity, Induced Polarization, and SPT-N Data for Subsurface Integrity Assessment in Granitic Terrain

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • Mbuotidem Dick,
  • Andy Anderson Bery,
  • Adedibu Sunny Akingboye,
  • Joseph Gnapragasan,
  • Nordiana Mohd Muztaza,
  • Nur Azwin Ismail,
  • Syarawi Muhammad Husni Sharoni,
  • Babangida Ahmed Mohammed

摘要

Reliable characterization of subsurface conditions, such as depth to bedrock, lithological variability, and the presence of fractures or faults, is critical for designing safe and durable foundations. However, geophysical methods remain underutilized in geotechnical design due to limited interdisciplinary collaboration, a general lack of awareness among engineers, and the absence of standardized integration frameworks. Hence, this study addresses these gaps by integrating geophysical and geotechnical data with machine learning (ML) techniques to enhance subsurface integrity modeling for foundation designs. Conducted in a granitic terrain in Minden, Pulau Penang, Malaysia, the research utilized electrical resistivity tomography, induced polarization, seismic refraction tomography, and standard penetration test (SPT-N) data acquired along two transects. The findings revealed critical subsurface anomalies—such as fractures, weathered zones, and lithological transitions—that pose potential risks to structural stability. Three major lithological layers were delineated: saturated clayey/silty topsoil extending to the residual profile; a transitional zone comprising weathered (clayey/silty) and fractured rock; and partially weathered to hard bedrock. ML algorithms applied included K-nearest neighbors (R² = 0.930 training, 0.905 testing), simple linear regression (R² = 0.898 training, 0.905 testing), principal component analysis, and K-means clustering (Silhouette score = 0.6688, residual sum of squares of 30.87 at Km = 3), all demonstrating strong predictive performance. The integrated framework proved effective for delineating subsurface lithology and assessing geomechanical properties, offering a scalable, data-driven strategy for cost-effective, high-resolution foundation design. This approach supports improved engineering decisions, reduces reliance on extensive invasive testing, and highlights the transformative potential of ML-assisted geophysical–geotechnical investigations in infrastructure development.

Graphical Abstract

This study presents a machine learning (ML)-assisted modeling approach that integrates Seismic Refraction Tomography (SRT), Electrical Resistivity Tomography (ERT), Induced Polarization (IP), and Standard Penetration Test (SPT-N) data to evaluate subsurface integrity in the granitic terrain of Minden, Penang Island, Penang, Malaysia. Geophysical surveys (SRT, ERT, and IP) and borehole SPT-N data were collected along two transects in the study area (see Panels 1 and 2). These datasets were integrated, and the individual geophysical models and borehole logs (Panel 3) were used to delineate subsurface lithologies and weak zones. Supervised ML techniques, including K-Nearest Neighbors (KNN) at Kn = 5 and Simple Linear Regression (SLR) (Panel 4), accurately predicted subsurface properties, as validated by actual versus predicted plots and corresponding performance metrics summarized in Panel 5. In contrast, the unsupervised K-means clustering algorithm (Panel 6) effectively classified the subsurface into three distinct lithological clusters, supported by density plots and average data points within each cluster. Additionally, the relationship between resistivity and chargeability was modeled and validated (Panel 7), reinforcing the reliability of the geophysical interpretation. Overall, the study demonstrates a robust, ML-enhanced framework for evaluating subsurface integrity in complex geological settings.