Â鶹´«Ã½

Author: Ira Hursti
Supervisor: Jussi Leveinen
 

Author: Ahmed Salama
Supervisor: Jussi Leveinen
Advisor(s): Dr. Tigist Chernet
Collaborative partner: Geological Survey of Finland

Abstract:

Titanium (Ti) and vanadium (V) are two critical commodities with a wide variety of applications. The AVANTIS project aims to reduce the European Union (EU) dependency on external suppliers for these critical commodities through developing selective blasting that can reduce the energy demand in the following crushing and grinding phases. This study aims to evaluate the impacts of the test blasts with variable configurations of powder factor and delay time on micro-fracturing and mineral liberation, while analysing the petrology of the study area to examine its influence on mineral liberation. The study area is the Metsa malmi outcrop of the Otanma ki mine, which was one of the global suppliers of vanadium in the 20th century.

Fourteen core samples and three drilling powder samples were collected prior to the blasts for mineralogical studies and to evaluate the pre-blast status of the micro-fracturing. Lump samples and powder samples were collected from the blast sites for assessment of micro-fracturing and mineral liberation successively. The micro-fracturing was studied by X-ray Computed Tomography (XCT). The XCT-processed samples were 2 core samples and 11 lump samples representing 8 blasts: 4 blasts categorised as a high powder factor group, while the other 4 were low powder factor group. The mineral liberation was studied by Micro-X-ray Fluorescence (μ-XRF), and each blast sample was sieved into two size fractions (2585–400 μm) and (400–63 μm) to evaluate the mineral liberation for each size fraction.

The blast samples have a wider fracture thickness range than the core samples. The low powder factor group had a wider range of fracture thickness. The long-delay-time blasts tended to generate thicker fractures in each group than the short delays. However, there was a statistically significant relation between the number of fractures and the powder factor. The finer fraction showed a higher degree of mineral liberation than the coarse one. Ilmenite showed a higher tendency for mineral liberation than magnetite.

Generally, the blasts induced new fractures and widened the existing ones. The delay time influences the thickness per fracture, while the powder factor influences the number of fractures and the sum of fractures thickness. The XCT and μ-XRF produce good results for quantification; however, they are time-consuming. Therefore, more resources should be allocated for producing more data population to strengthen the statistical analysis and its findings.

Author: Sneha Bhaskaran
Supervisor: Wojciech Sołowski
Advisors: Jussi-Pekka Matilainen and Naum Shpata
Collaborative partner: Sumitomo SHI FW Energia Oy

Abstract:

This study investigates the influence of soil–structure interaction (SSI) on the seismic response of an industrial boiler structure, with emphasis on how variations of soil stiffness, and seismicity level, and seismic design code formulations influence the seismic design of the structure. In this comparative study, a typical boiler structure is hypothetically placed at six sites: three in the United States designed accordance with ASCE 7-22 and three in Italy designed using 2nd generation (2024) of EN 1998. The site-specific design spectra were developed using data from ASCE Hazard Tool and EFEHR Hazard Map. SSI was incorporated primarily through inertial interaction by estimating the flexible-base period and developing SSI-adjusted response spectra and base shear calculation. In addition, Plaxis 2D simulations were performed to assess the adequacy of code-based SSI design approaches compared with detailed nonlinear numerical modelling

The results indicate that SSI effects are negligible for stiff soils but become increasingly significant for soft soils under higher seismic demand. In ASCE 7-22, period elongation and base shear reductions are minimal for stiff soils but become more substantial as soil stiffness decreases, particularly in higher seismicity regions. Calculations based on the second-generation EN 1998 show similar overall trends; however, they predict substantially larger period elongations and base shear reductions for soft-soil conditions at high-seismicity sites. The PLAXIS 2D simulations confirmed the general trend of base shear reduction under flexible-base conditions and provided additional insights into stress redistribution, bending moments, and deformation mechanisms that are not captured by code method.

The study concludes that although both codes adopt similar fundamental principles for SSI implementation, difference in shear modulus reduction factors and formulations significantly influence the predicted seismic demand. In particular, the 2nd generation EN 1998 approach yields significantly larger SSI-induced reductions than ASCE 7-22, especially for soft soils at high seismic location. The numerical simulations further demonstrate that advanced nonlinear modelling provides a more detailed assessment of SSI effects than simplified code procedures. These findings highlight the importance of carefully verifying SSI-based reductions when designing industrial boiler structures, particularly in soft-soil, high-seismicity locations.

Author: Dure Kamla Siddiqui
Supervisor: Professor Sanandam Bordoloi
Advisors: Professor Shankar Deka, David Gaharia
Collaborative partner: LabMind, Sweden

Abstract:
This thesis develops and evaluates a machine learning-based decision-support method for early Swedish geotechnical soil classification. Soil classification in laboratory work depends on visual description, sample handling, and professional judgement. In many routine cases, this process is reliable. However, uncertainty can increase when soils are fine-grained, mixed, organic, or layered. At the early stage of assessment, the available information may also be limited, while more detailed laboratory results may come later. This creates a need for a data-based method that can give an early classification suggestion and show when the result should be treated with caution.

The study aims to test whether a hybrid machine learning model can support early soil classification using available laboratory data. The dataset consists of Swedish laboratory soil records with detailed descriptions, including main soil type, secondary components, layering, organic content, and special indicators. These descriptions are grouped into broader soil classes for machine learning. The proposed method combines laboratory soil images with selected physical index properties, including depth, average density, and water content. A convolutional neural network is used for the image data, while a tabular model is trained on the selected physical properties. The two outputs are combined using late fusion. Possible sulphide soil, dry crust, and minor organic content are handled as separate model-based indicators rather than main soil classes.

The results show that machine learning can be used as an early support method for Swedish geotechnical soil classification. The hybrid model achieved better classification performance than the image-only and tabular-only models, showing that soil images and physical properties provide complementary information. The model still showed confusion between some visually and physically similar soil groups. This supports the need to show uncertainty rather than report only one fixed label. For this reason, conformal prediction is used to return prediction sets when the model is less confident. The tool is therefore proposed as a support method that can suggest likely soil classes, highlight uncertain cases, and make early laboratory classification more transparent.