Institutional Repository

U n i v e r s i t y of P e r a d e n i y a

The Institutional Repository of the University of Peradeniya is the University's digital gateway to scholarly knowledge and research excellence. Established as the next phase of the Digital Library initiative launched by the University Library in 2011, it preserves, showcases, and provides open access to the intellectual output of the University community.

Through a rich collection of research articles, conference proceedings, theses, dissertations, and other scholarly works, the repository enhances the visibility, accessibility, and long-term preservation of knowledge, extending the global reach and impact of the University of Peradeniya’s academic and research contributions.

Photo by @ITD,Main Library

Communities in DSpace

Select a community to browse its collections.

Now showing 1 - 5 of 5

Recent Submissions

Evaluating out-of-domain performance prediction methods under domain shifts
(University of Peradeniya, Sri Lanka, 2026-08-20) Weerasinghe, P.; Tennakoon, R.; Thangarajah, J.; Herath, D.
Deep neural networks have achieved remarkable success in various applications where the test data follows the same distribution as the training data. However, in many real-world scenarios, this assumption often fails due to domain shifts, where test data referred to as out-of-domain (OoD) data differs from the training distribution. Such shifts can significantly degrade model performance and are particularly pronounced in data-constrained domains such as medical imaging, arising from variations in imaging devices, acquisition protocols, and patient demographics. While numerous techniques have been proposed to address domain shift, their effectiveness remains limited, especially when labeled target data is unavailable. This study investigates algorithms designed to predict OoD performance on unlabeled datasets and evaluates their reliability under different types of domain shifts, specifically diversity shifts (emergence of new patterns) and correlation shifts (changes in feature–label relationships). A systematic benchmarking framework is adopted to evaluate a diverse set of state-of-the-art OoD performance prediction algorithms across datasets exhibiting both shift types. The evaluation includes benchmark datasets such as PACS, CIFAR-100- C, and ColoredMNIST. OoD performance prediction methods Average Threshold Confidence (ATC), Difference of Confidence (DoC), Mandoline, and AugMax are assessed under consistent experimental settings using metrics such as robustness accuracy and mean absolute error. The results reveal significant variability in algorithm performance across shift types, indicating a lack of consistent generalization. Under correlation shift (MNIST to ColoredMNIST), ATC achieves substantially higher robustness accuracy (0.9466) compared to DoC (0.2096), demonstrating its effectiveness in confidence-based estimation. In contrast, under diversity shift (CIFAR-100 to CIFAR-100-C), Mandoline outperforms ATC (0.5164 vs. 0.2655), highlighting its sensitivity to distributional variation. Overall, no single method performs reliably across both shift types, emphasizing limitations of current approaches in complex real-world scenarios. These findings highlight the need for shift-aware or hybrid OoD performance prediction frameworks capable of handling diverse and interacting domain shifts. This work provides a benchmark for future research and offers critical insights for advancing reliable OoD performance estimation.
Enhancing developer productivity through hybrid knowledge graph–LLM integration in software engineering environments
(University of Peradeniya, Sri Lanka, 2026-08-20) Vaishaly, M.; Vasanthapriyan, S.; Rubasinghe, T. D.
Modern software repositories contain highly interconnected artifacts, including source code, documentation, issue reports, pull requests, commits, and dependency information, making repository exploration and software maintenance increasingly challenging. Although Large Language Models (LLMs) provide valuable assistance for software engineering tasks, they frequently lack repository-specific contextual understanding and may generate unsupported or hallucinated responses when reasoning across multiple software artifacts. This study proposes an adaptive hybrid Graph Retrieval-Augmented Generation (GraphRAG) framework that integrates a repository-specific knowledge graph with graph and vector retrieval to generate accurate, explainable, and evidence-grounded repository reasoning. The research addresses the questions of whether graph-grounded retrieval improves repository question answering, reduces hallucinations, supports multi-hop reasoning, and enhances repository exploration efficiency. A mixed-method evaluation is adopted using both a large-scale synthetic software engineering dataset and five real-world open-source repositories: Spring PetClinic, OpenMRS, eShopOnContainers, Apache Airflow, and MLflow. The synthetic dataset is programmatically generated to simulate realistic software development environments by creating interconnected repositories, developers, commits, issues, pull requests, documentation, dependencies, and software components with predefined semantic relationships. This controlled environment provides complete ground truth, enables reproducible experiments, supports scalability analysis, and facilitates systematic ablation studies, while the real repositories validate the framework under practical software engineering conditions. Repository artifacts are parsed using static code analysis and repository metadata extraction, transformed into a Neo4j Knowledge Graph, and integrated with adaptive GraphRAG retrieval. A benchmark comprising 300 manually designed repository reasoning questions is developed across entity retrieval, dependency analysis, issue reasoning, documentation retrieval, pull request analysis, and multi-hop reasoning. Ground-truth answers are generated through validated Cypher queries and manual verification against repository evidence. Experimental evaluation against Plain LLM, Vector RAG, Graph RAG, and the proposed adaptive GraphRAG framework demonstrates improvements with an Exact Match score of 0.842, Precision of 0.891, Recall of 0.876, F1-score of 0.883, ROUGEL of 0.901, BERTScore of 0.928, a hallucination rate of 0.071, multi-hop reasoning accuracy of 0.864, and an average repository task completion time of 41.8 seconds, outperforming the comparison methods. These findings demonstrate that adaptive graph-grounded retrieval provides more reliable, explainable, and scalable repository reasoning while establishing a reproducible evaluation framework for intelligent software engineering assistants.
Dynamic simulation and control of a wind-powered chlor-alkali electrolysis system
(University of Peradeniya, Sri Lanka, 2026-08-20) Rajapaksha, R. B. G. D. N.; Fernando, N.
The chlor-alkali membrane electrolysis process is one of the most electricity intensive operations in the chemical industry, producing chlorine, sodium hydroxide, and hydrogen for numerous industrial applications. Replacing fossil fuel-based electricity with renewable energy sources such as wind power offers significant environmental benefits but introduces power fluctuations that challenge stable electrolyzer operation. This study develops a dynamic mathematical model and control strategy to investigate the feasibility of operating a membrane chlor-alkali electrolyzer under fluctuating wind power while maintaining stable product quality. The research methodology involved developing an electrochemical model based on reversible voltage, activation overpotential, ohmic losses, and the additional resistance effects caused by gas bubble formation at the electrode surfaces. Dynamic mole balance model was also developed for the anode and cathode compartments to predict concentration changes as a result of flow variations with time. The model was implemented in Scilab, while the control strategy was developed in Scilab-Xcos using two PI controllers to regulate the anode brine feed rate and cathode water feed rate. The electrolyzer was operated over an industrial current density range of 3000–6000 A m⁻², representative of flexible operation under variable wind power. Model predictions were validated against published literature data, showing good agreement for steady-state cell voltage and electrolyte concentration trends. Inclusion of the bubble resistance model increased the predicted cell voltage by approximately 2.0 – 2.5 V at 6000 A m⁻², significantly improving agreement with reported operating voltages. The PI controllers maintained the cathode outlet concentration at 32 wt.% NaOH despite rapid current density variations. Controller parameters of Kp = 0.12, Ki = 0.28 for brine flow control and Kp = 74.27, Ki = 7.42 for cathode water flow control provided stable disturbance rejection with minimum steady state error. The results demonstrate that membrane chlor-alkali electrolyzers can be operated flexibly under renewable electricity while maintaining stable product quality. Future work will focus on benchmarking the model against industrial operating data and developing a two-dimensional CFD model using OpenFOAM to capture detailed hydrodynamics, gas bubble behaviour, and species transport, thereby providing more realistic predictions of electrolyzer performance.
Towards a pedagogically aligned blended learning framework for foundation computing education: evidence from behavioural learning analytics of engineering undergraduates
(University of Peradeniya, Sri Lanka, 2026-08-20) Ratnapala, I. P.
Undergraduate engineering students possess diverse skills, necessitating a well-structured foundation for a smooth transition to tech-enhanced learning. However, many blended teaching methods are ad hoc and misaligned with learner readiness, topic difficulty, or learning outcomes. This study proposes a Pedagogically Aligned Blended Course Design Framework for Foundation Computing in engineering programmes. Unlike existing blended learning models that primarily prescribe instructional methods or technology integration, this framework provides a systematic process for aligning learner readiness, topic complexity, cohort feedback, and ILOs with appropriate blended strategies. This alignment enables lecturers to make evidence-informed instructional decisions rather than adopting blended learning practices in an ad hoc manner. By integrating these pedagogical dimensions into a repeatable course design process, the framework provides a practical approach for developing adaptive, scaffolded, and learner-centred Foundation Computing courses in engineering education. The curriculum was organised into six instructional clusters: face-to-face instruction, external online courses, instructional videos, interactive videos, interactive lessons, and annotated learning materials. Formative assessments integrated game-based learning, peer review, and automated programming practices within the Moodle LMS. The framework was implemented in a six-week programme at the Faculty of Engineering, University of Peradeniya, in 2025. After preprocessing and removing missing data and outliers, data from 282 students were analysed using R. The evaluation encompassed pre- and post-assessment analysis, LMS behavioural analytics, and qualitative feedback analysis. Descriptive statistics indicated that pre-assessment scores had a mean of 46% (SD=9.38), whereas post-assessment scores had a mean of 83% (SD=15.29). A paired-samples t-test demonstrated a statistically significant improvement in post-assessment scores (t(280) = 41.07, p < 0.001), with a mean learning gain of 37.08 points and a very large effect size (d = 2.45). Furthermore, a Wilcoxon Signed-Rank test confirmed the significance of the improvement (p<0.001). Correlation analysis revealed weak relationships between basic LMS usage indicators (such as login frequency and time spent) and performance outcomes; however, LMS activity completion, configured across all course activities, exhibited a strong positive correlation with learning gain (r=0.681, p<0.001). Qualitative feedback further underscored students' positive perceptions of interactive activities, personalised guidance, clear navigation, and game-based learning. The findings suggest that effective blended learning environments in engineering education should be developed through pedagogically aligned instructional planning rather than technology-driven implementation. The proposed framework demonstrates how learner readiness, curriculum complexity, and ILOs can be systematically mapped to blended teaching strategies to establish adaptive, evidence based, and engagement-oriented Foundation Computing courses.
Digitalizing natural rubber metrology: a multi-parameter IoT terminal for first-principles DRC measurement and fraud detection
(University of Peradeniya, Sri Lanka, 2026-08-20) Suriyapperuma, H. D.; Manilgama, N. C.; Rukshan, A. D.; Perera, K. A. P. M.; Vimukthi, Y.
According to the Sri Lanka Export Development Board (EDB), the natural rubber industry generates approximately $870 million in annual export earnings, yet systemic measurement inefficiencies contribute to an estimated $26 million (approximately 3% of annual export earnings) in annual industrial losses. Rural collection centers currently rely on manual glass hydrometers (Metrolacs) prone to temperature errors and human manipulation or slow laboratory oven methods that delay farmer payments to bi-monthly cycles. While a prior electronic research prototype has been published, it relies on an empirical regression model requiring extensive pre-calibration data that limits its adaptability across diverse field conditions. This research proposes a compact, field-ready IoT terminal designed to bridge this gap by providing near-instantaneous Dry Rubber Content (DRC) measurements alongside multi-parameter sensing for fraud detection. The methodology involves a cyber physical system centered around an ESP32 microcontroller and high precision load cells to calculate DRC via buoyancy measurement using Archimedes’ principle. Unlike the existing regression dependent model, this physics based approach provides a universal measurement independent of variations in field latex composition. A DS18B20 probe provides real time temperature compensation while a unique relay-isolation hardware architecture resolves galvanic ground loop interference between the co-submerged pH and TDS sensors. The developed system is a fully integrated, hand-held operational terminal that outperforms the accuracy of the traditional glass Metrolac at the point of collection, utilizing an MQTT-based “store-and-forward” protocol to prevent data loss in remote areas with intermittent connectivity. Validation against eight field latex samples from Sri Lankan rubber collection centers showed that our device achieved a mean error of ±0.74% DRC against the ISO 126 laboratory oven standard, compared to ±1.53% for the traditional glass Metrolac against the same standard. The worst-case error was reduced from ±7.75% to ±2.10%. These results demonstrate that the proposed system is a practical and more accurate alternative to the traditional glass Metrolac, offering rubber farmers and collection centers a faster, fairer and more reliable way to measure latex quality. Future work will expand validation across a wider range of latex sources and further improve accuracy through hardware fine-tuning and extended field trials.