<p>The conversion of old underground galleries into opencast mines creates significant geotechnical challenges related to deformation behaviour, slope performance, and deformation control under surface mining conditions. To address these challenges, a field-informed deformation-response evaluation framework integrating finite element modelling, statistical analysis, IoT-based field monitoring, and machine learning–based surrogate prediction is proposed. A field investigation was conducted at a large opencast mine operated by Singareni Collieries Company Limited in South India. Geometric parameters influencing the conversion process, including partition thickness, pillar width, gallery height, slope angle, and berm width, were collected and used as input for finite element simulations. A total of 3,240 simulation cases were analysed using ANSYS Workbench under systematically varied geometric conditions. The results indicate that partition thickness, gallery height, and pillar width are the dominant parameters influencing deformation response above underground galleries. Increasing partition thickness and pillar width reduced deformation due to improved stiffness and load redistribution effects, whereas increasing gallery height and steeper slope conditions increased deformation because of higher stress concentration and reduced confinement within the overburden strata. Statistical regression analysis confirmed these trends and identified partition thickness as the most influential parameter. To capture nonlinear deformation-response behaviour and cross-parameter interaction effects, a Random Forest regression model was developed. The model achieved high prediction accuracy, with training and testing <InlineEquation ID="IEq1"><EquationSource Format="TEX">\(R^2\)</EquationSource></InlineEquation> values of 0.998 and 0.997, respectively, whereas independent IoT-based field-validation observations obtained from 72 monitored locations achieved an <InlineEquation ID="IEq2"><EquationSource Format="TEX">\(R^2\)</EquationSource></InlineEquation> value of 0.93 under practical mining conditions. Feature importance analysis further identified partition thickness (79.05%), gallery height (9.40%), and pillar width (8.63%) as the dominant parameters governing deformation behaviour. The trained Random Forest model also significantly reduced computational effort, generating deformation-response predictions in less than one second compared with approximately 8–15 min required for repeated finite element simulations. The proposed framework provides a computationally efficient approach for comparative deformation-response evaluation and preliminary engineering decision support within the investigated mining conditions.</p>

Integrated numerical and machine learning framework for deformation-based assessment of slopes over old underground galleries in opencast mines

期刊论文
出版模式：

开放获取
发表日期：

2026年7月18日

查看全文

Kumar Dorthi,
Neelima Bayyapu,
Ram Chandar Karra

摘要

The conversion of old underground galleries into opencast mines creates significant geotechnical challenges related to deformation behaviour, slope performance, and deformation control under surface mining conditions. To address these challenges, a field-informed deformation-response evaluation framework integrating finite element modelling, statistical analysis, IoT-based field monitoring, and machine learning–based surrogate prediction is proposed. A field investigation was conducted at a large opencast mine operated by Singareni Collieries Company Limited in South India. Geometric parameters influencing the conversion process, including partition thickness, pillar width, gallery height, slope angle, and berm width, were collected and used as input for finite element simulations. A total of 3,240 simulation cases were analysed using ANSYS Workbench under systematically varied geometric conditions. The results indicate that partition thickness, gallery height, and pillar width are the dominant parameters influencing deformation response above underground galleries. Increasing partition thickness and pillar width reduced deformation due to improved stiffness and load redistribution effects, whereas increasing gallery height and steeper slope conditions increased deformation because of higher stress concentration and reduced confinement within the overburden strata. Statistical regression analysis confirmed these trends and identified partition thickness as the most influential parameter. To capture nonlinear deformation-response behaviour and cross-parameter interaction effects, a Random Forest regression model was developed. The model achieved high prediction accuracy, with training and testing \(R^2\) values of 0.998 and 0.997, respectively, whereas independent IoT-based field-validation observations obtained from 72 monitored locations achieved an \(R^2\) value of 0.93 under practical mining conditions. Feature importance analysis further identified partition thickness (79.05%), gallery height (9.40%), and pillar width (8.63%) as the dominant parameters governing deformation behaviour. The trained Random Forest model also significantly reduced computational effort, generating deformation-response predictions in less than one second compared with approximately 8–15 min required for repeated finite element simulations. The proposed framework provides a computationally efficient approach for comparative deformation-response evaluation and preliminary engineering decision support within the investigated mining conditions.