Welcome to SDU Institutional Repository
Communities
Select a community to browse its collections.
Recent Submissions
Integration of Machine Learning for Enhanced Efficiency in Industrial Systems: Analysis, Forecast, and Perspectives
(SDU University, 2026) Borangazin T.
Contemporary industrial systems exhibit a pronounced dependence on extensive data, thereby requiring sophisticated analytical frameworks to facilitate decisionmaking under conditions of uncertainty. Conventional statistical methodologies are frequently constrained in their ability to encapsulate the non-linear, dynamic, and multi-factorial interactions characteristic of complex industrial ecosystems. This Master’s thesis undertakes an exploration into the integration of advanced machine learning paradigms to augment the efficacy of industrial forecasting and property prediction endeavors, as evidenced by two pertinent case studies. The initial case study undertakes an investigation into the efficacy of supervised learning models for the prediction of international crude oil prices. A suite of algorithms, including linear regression, tree-based ensembles, and recurrent neural networks, were subjected to training and evaluation utilizing historical price data in conjunction with pertinent macroeconomic indicators. The study substantiates that non-linear machine learning models yield superior predictive accuracy relative to traditional baselines, with a notable advantage observed during periods of market volatility. The second case study delves into predicting mineral properties, with a particular emphasis on Mohs hardness, by leveraging tabular datasets that capture chemical composition and physical measurements. The study develops regression models and a basic generative model to evaluate the role of machine learning within materials informatics. The generative model underscores the feasibility of generating synthetic mineral-like feature vectors, representing an early step towards data augmentation and the concept of inverse design. For both case studies, the dissertation outlines a standardized methodology that involves preparing data, crafting features, training models, validating them, and evaluating performance with metrics like MAE, RMSE, and (R2 ) . The findings illustrate how versatile machine learning is for different industrial data types and suggest that data-driven models offer exciting possibilities for integration into current industrial practices.
Digital Competence of Secondary School EFL Teachers: Perceptions, Challenges, and Practices
(SDU University, 2026) Bekaidar N.
Against the backdrop of a national declaration of 2026 as the “Year of Digitalization and Artificial Intelligence” Kazakhstan is carrying out a large-scale reform of its educational system. As for secondary school English as a Foreign Language (EFL) teachers, this change entails a transition from basic Information and Communication Technology (ICT) literacy to advanced, pedagogy-oriented digital competence. The qualitative study examined perceptions, practices and systemic challenges of six EFL teachers in different urban, private and rural settings in Almaty region. This study uses semi-structured interviews and theme analysis to explore how instructors deal with the new “teacher - artificial intelligence (AI) - student” relationship. The research reveals a clear divide in self-efficacy, with many instructors feeling overwhelmed by the pace of change in technology and constant infrastructure problems, while younger teachers in well-resourced workplaces report high levels of confidence and active tool use. Concerns included the rural-urban digital divide, severe time constraints, and lack of on-site pedagogical Information Technology (IT) support, similar to those in the first- and second-order obstacles literature (Ertmer, 1999). The study concludes that for sustainable development of digital competence we should move from the generic ICT training to contextual, practice-based, ‘cyberpedagogy’ models and formation of formal PLCs to encourage peer support and reduce professional isolation. The study recommendations clearly show that success in the national digitalisation policies requires a balance between the provision of hardware and pedagogical support.
A Comparison of Effect of Abstract Textbook and Real-Life–Based Physics Problems on Students’ Attitudes and Perceived Relevance
(SDU University, 2026) Rakhymberdiyev B.
The purpose of this study was to investigate the effects of physics problems set within real-life contexts on students' understanding of physics and the relevance of physics to their lives․ The study used a mixed-method design with 84 students responding to textbook physics problems and five context-based problems, and 52 of them participated the survey with multiple-choice, Likert-scale, and open-ended questions․ Their results in both types of problem solving activities were compared and analyzed to test if there is any difference in overall achievements of students. Thematic analysis was implemented in order to make sense of the qualitative data as results of the open-ended questions. Most studies found that context-based problems had a positive effect on students' interest and relevance to physics․ Student learning‚ as measured by conceptual understanding tests‚ often improved‚ but the effect on performance was more mixed․ Results of the study indicated positive tendency in both problem solving and survey. Due to open-ended questions results, the vast majority of participants posed that realistic problems are deemed more interesting and encourage meaningful learning of physics‚ though this is only the case with a balanced set of problems
Variety of special Tortken algebras
(SDU University, 2026) Baigali M.
This dissertation studies special Tortken algebras arising from the symmetrization of Novikov algebras. The main object is the space of symmetric elements in the free Novikov algebra generated by one element. Using the differential realization of free Novikov algebras, the symmetrized product is written as a ◦ b = (ab) ′ , which allows us to expand elements in the ◦-language into differential monomials and study their linear relations by methods of linear algebra. For the homogeneous component Tn of degree n, we describe the dimensions and obtain dim T1 = dim T2 = 1 and dim Tn = p(n − 2) for n ≥ 3, where p(n) is the partition function. Low-degree components are constructed explicitly, and special attention is given to relations in degrees 6 and 7. The computations suggest a recursive structure for the spaces Tn and contribute to the study of free special Tortken algebras.
Dynamic Obstacle Avoidance in Autonomous Robot Navigation Using Deep Reinforcement Learning
(SDU University, 2026) Parmash A.
Existing indoor navigation systems for autonomous robots have proven their success in geometric mapping and avoidance of obstacles in the local environment, but they lack semantic reasoning capabilities; thus, they cannot comprehend instructions issued in natural language. In contrast, utilizing state-of-the-art vision and language foundation models in the robot control loop would create too much computational delay, making the implementation impractical for the purpose of reactive execution of tasks. This thesis attempts to fill this gap through a two-stage hierarchy. During the initial phase, a mobile robot performs an exploration task within the unknown environment by employing an SLAM pipeline based on LiDAR, while the robot’s on-board VLM performs object detection to build a structured, small global 3D semantic database. During the second phase, the robot takes up any natural-language instructions (for example, “go to the bookshelf”), extracts the coordinates of the target from its memory, and performs autonomous navigation. Here, the navigation process is entirely mapless and does not involve any VLM inference cost overhead. Control at the lower level is done by the robot’s DRL-based policy trained using TD3 algorithms. Assessed under ROS-integrated Gazebo simulation, the proposed architecture attains an accuracy rate of 85% in navigation with a lightweight Moondream VLM, substantially superior to cloud-based GPT-4o mini versions and classical approaches. The trained policy for depth proves to have excellent generalization performance with respect to unknown objects, dense obstacles, and even multiple rooms, as well as a success rate of 80% in evading moving obstacles without the need for further training. Achieving a processing rate of 10 Hz on board, the framework manages to bypass foundation model delay constraints.