http://dspace.dtu.ac.in:8080/jspui/handle/123456789/45 2026-11-07T02:00:33Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/23396 Title: LANDSLIDE SUSCEPTIBILITY ASSESMENT IN SIKKIM USING FREQUENCY RATIO AND SHANNON ENTROPY APPROACH Authors: KUMAR, ISHAAN; SARKAR, RAJU (SUPERVISOR) Abstract: Landslide Susceptibility Assessment (LSA) is crucial for understanding and mitigating landslide risks, especially in vulnerable regions like Sikkim, where complex topography, diverse geological formations, and significant precipitation contribute to instability. This study integrates geospatial analysis with statistical modelling techniques to evaluate the influence of 10 landslide conditioning factors (LCFs), including aspect, rainfall, curvature, elevation, hillshade, roughness, slope, distance to stream and topographic wetness index (TWI) and normalized difference vegetation index (NDVI). Using a landslide inventory of 677 recorded events and GIS-derived thematic layers were used to apply two statistical models: Frequency Ratio (FR) and Shannon Entropy (SE). Landslide inventory is created by dividing the total landslide points into training data (80%) and testing data (20%). The vulnerability map is then created using the 2 models and a comparison is done between them. Then, the map created by the above-mentioned 2 models is validated with the help of a testing data set using the AUC (Area under curve) of the ROC (Receiver Operating Characteristics) curve. The resulting susceptibility map categorizes the state into distinct risk zones, aiding in hazard mitigation and land-use planning. The Frequency Ratio shows AUC of 0.760 and the Shannon Entropy shows the AUC of 0.724. Model validation through AUC- ROC analysis and success prediction rate curves confirm the reliability and predictive accuracy of the applied techniques. This comparative study helps in underscores the importance of combining statistical models with geospatial tools for disaster risk management and infrastructure planning in Sikkim. These maps can be further utilized by the government in mitigation planning. 2026-03-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/23369 Title: BIO-HEALING OF GENERATED CRACKS IN ROCK MASS USING ENZYME Authors: ANAND, ADARSH; Srivastava, Amit Kumar (SUPERVISOR) Abstract: Rock masses in tunnels, slopes, foundations and underground excavations are highly susceptible to crack development caused by blasting, excavation, weathering and repeated mechanical loading, which compromises their strength, permeability and long-term stability. Conventional repair techniques such as cement grouting and epoxy injection are costly, environmentally damaging and often ineffective for fine or inaccessible cracks. This study investigates Enzyme-Induced Carbonate Precipitation (EICP) as a sustainable, bio-based alternative for healing artificially generated cracks in rock specimens. Rock core samples retrieved from a field borehole were characterized for their physical properties, after which artificial cracks were induced and treated with an enzyme healing solution containing urease, urea and calcium chloride. The treated specimens were cured under controlled conditions to allow calcium carbonate precipitation within the crack network. Post-treatment evaluation included measurements of weight change, density, water absorption, crack width closure and Uniaxial Compressive Strength (UCS) recovery, which were compared against untreated cracked specimens. The results demonstrate that EICP treatment produced measurable healing efficiency, with enzyme-induced calcium carbonate deposits sealing crack surfaces, reducing permeability and restoring a significant portion of the original compressive strength. These findings confirm that enzyme-based bio-healing offers a viable, eco-friendly and cost-effective approach for repairing damaged rock masses in geotechnical and underground engineering applications, with potential for further validation over longer curing durations and varied rock types. The other serval application of EICP in tunnel stabilization, Sustainable rock engineering and foundation improvement. 2026-06-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/23368 Title: POND ASH AS A SUSTAINABLE BACKFILL FOR REINFORCED EARTH (RE) WALL Authors: POONIA, PRATHAM; Srivastava, Amit Kumar (SUPERVISOR) Abstract: The rapid expansion of highway infrastructure in India has intensified the demand for granular backfill in reinforced soil wall construction, placing sustained pressure on natural borrow sources and renewing interest in industrial by-products such as pond ash as an alternative fill material. This dissertation investigates whether pond ash can serve as a technically sound, code-compliant and economically viable backfill for reinforced soil walls at a live highway project site on the NH-46 Guna-Biaora section (chainage 118+400), where the native fill is a plastic, SC-classified sandy clayey soil. The investigation proceeds through a comparative geotechnical characterisation of the native soil and pond ash, an assessment of the mechanical and reinforcement-interaction behaviour of pond ash under site-specific loading, an examination of hydraulic and environmental response under seasonal groundwater fluctuation, the identification of field performance monitoring requirements, and an economic and sustainability appraisal of pond ash against conventional borrow material within an Engineering, Procurement and Construction (EPC) project framework. The native soil exhibits a dry density of 1.821–1.829 g/cc, cohesion of 0.03–0.04 kg/cm² and an internal friction angle of 25.3°–25.9°, with a shallow groundwater table at approximately 3.50 m governing drainage and pore-pressure behaviour. By comparison, the pond ash is non-plastic, contains 6.75% fines, and returns a markedly higher friction angle of 31.50° with negligible cohesion, a maximum dry density of 1.344 g/cc at an optimum moisture content of 25.30%, and full compliance with the gradation and strength criteria of IRC:SP:102-2014. Limit-equilibrium stability checks for the reinforced soil wall section return factors of safety of 1.82 against sliding, 2.45 against overturning and 3.10 against bearing failure, with a global factor of safety of 1.68, all exceeding their respective minimum requirements. These results indicate that pond ash offers genuine geotechnical and sustainability advantages — reduced dependence on natural borrow material, beneficial reuse of an industrial residue, and favourable frictional behaviour — but also reveal material properties that differ substantially from the native soil baseline, particularly in density and cohesion. The economic case for pond ash is accordingly found to be conditional rather than assured: within an EPC-mode project framework, any cost or schedule advantage depends on verified field compaction, drainage performance and reinforcement interaction, since unvalidated deviations from design assumptions could erode the anticipated savings. The dissertation concludes that pond ash is a technically promising and potentially sustainable backfill alternative for reinforced soil wall construction, whose safe and economical adoption should be confirmed through calibrated field-scale trials and continued instrumented monitoring prior to full-scale implementation. This thesis addresses the technical, structural, economic, field and sustainability feasibility of using source-specific pond ash as an alternative sustainable backfill material for Reinforced Earth walls, through comparison with conventional soil and compliance with IRC:SP:102-2014. 2026-06-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/23366 Title: AN EXPLAINABLE MACHINE LEARNING FRAMEWORK FOR PREDICTING UNIAXIAL COMPRESSIVE STRENGTH OF ROCK USING INDEX PROPERTIES Authors: ROY, SWAGAT; Srivastava, Amit Kumar (SUPERVISOR) Abstract: The uniaxial compressive strength (UCS) of rock is one of the most important engineering properties used in the design and analysis of tunnels, slopes, foundations, underground excavations, and mining projects. Although the laboratory determination of UCS provides reliable results, it is often time-consuming, expensive, and requires carefully prepared standard specimens. In many practical situations, obtaining suitable rock samples for direct UCS testing is difficult. Consequently, the prediction of UCS using easily measurable index properties has become an important area of research in rock mechanics and geotechnical engineering. This study presents an explainable machine learning framework for predicting the UCS of rocks using four commonly measured index properties: Porosity (%), Schmidt Hammer Rebound (SHR), P-wave velocity (Vp), and Point Load Index (Is50). A dataset of 584 rock specimens, compiled primarily from the open-access ROCK/10/4025 database and supplemented with specimen-level data reported by Abd El Aal (2017), was used for model development following a zero-imputation policy that retained only specimens with complete measurements. Three machine learning algorithms which are Random Forest Regression, Ridge Regression, and Huber Regression were developed and evaluated on an 80:20 train-test split using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The Random Forest Regression model achieved the best performance on the test partition, with an R² of 0.73, an RMSE of 26.19 MPa, and an MAE of 16.96 MPa, outperforming both linear alternatives. To improve the transparency and interpretability of the predictive model, SHapley Additive exPlanations (SHAP) analysis was applied to the trained Random Forest model to quantify the contribution of each index property to the predicted UCS. The explainability analysis, consistent with the Pearson correlation trends observed in the dataset, ranked porosity as the most influential parameter, followed by P-wave velocity, Schmidt Hammer Rebound Number, and Point Load Index. Based on the trained models, an empirical equation for UCS estimation was also developed to provide a practical, rapid engineering tool. iv The developed framework was further validated against ten independently tested laboratory rock specimens that were not part of the training dataset. Since the Point Load Test and the UCS Test are both destructive and cannot be performed on the same specimen, the corresponding Is50 values for the validation samples were estimated theoretically using established empirical correlations. The comparison between laboratory-measured and model-predicted UCS values showed close agreement, with an MAE of 3.09 MPa, an RMSE of 4.00 MPa, and a mean absolute percentage error (MAPE) of 10.32% across the ten specimens. Prediction accuracy was found to be highest for competent, low-porosity rock and comparatively lower for weaker, more porous specimens, where micro-scale heterogeneity has a greater relative influence on strength. As this validation exercise was based on a limited set of ten specimens, it should be regarded as an encouraging preliminary indication rather than conclusive proof of the framework's reliability, and broader validation on larger, independent datasets is recommended before wider adoption. The proposed explainable machine learning framework offers an efficient, accurate, and interpretable approach for estimating the uniaxial compressive strength of rocks. It has the potential to reduce the need for extensive laboratory testing and can assist geotechnical engineers in preliminary site investigations, rock mass characterisation, and engineering design where rapid and reliable UCS estimation is required. 2026-07-01T00:00:00Z