Chalmers Open Digital Repository

Välkommen till Chalmers öppna digitala arkiv!

Här hittar du:

  • Studentarbeten utgivna på lärosätet, såväl kandidatarbeten som examensarbeten på grund- och masternivå
  • Digitala specialsamlingar, som t ex Chalmers modellkammare
  • Utvalda projektrapporter

Forskningspublikationer, rapporter och avhandlingar hittar du i research.chalmers.se

 

Enheter i Chalmers ODR

Välj en enhet för att se alla samlingar.

Senast inlagda

Bild (thumbnail)
From Logs to Insights: Exploring User Behavior in RobotStudio
(2025) Li, Jinxuan; Tang, Shijia
Understanding how users interact with software is essential—not only for designing intuitive interfaces but also for building meaningful, behavior-driven test scenarios. In this study, we explore user behavior in ABB RobotStudio by analyzing a large dataset of backend log files, each recording detailed event traces from real usage sessions. To approach this problem from multiple angles, we employ three methods. First, we apply N-gram analysis to examine what users do, what events tend to occur together, and in what order, providing us with a window into common behavioral patterns. Second, we construct first-order Markov chains to model the likelihood of transitioning from one type of event to another, capturing dynamics user actions. Third, we use clustering and N-gram to investigate whether different types of event sequences naturally emerge. This helps us uncover whether there are distinct features and recurring patterns of usage across clusters. Together, these three methods reveal both structural and sequential aspects of how users interact with RobotStudio. We find that certain behaviors repeat consistently across users and sessions, while others are more context-dependent. These insights offer practical value: they can support user-centered UI improvements and help generate test cases that more closely mirror real-world workflows.
Bild (thumbnail)
Framework Insights and Automated Attestation for Software Supply Chain Security
(2025) Falk, Elias; Svensson, Emil
The escalating threat of software supply chain attacks necessitates robust security measures; however, current guidance is fragmented across numerous, often overlapping, frameworks. This thesis addresses this challenge through a dual approach. First, it conducts a systematic comparative analysis of five prominent software supply chain security frameworks - ESF, S2C2F, SCVS, SLSA, and an academic SOK taxonomy - by decomposing their 284 guidelines into 1,321 atomic, actionable statements. These statements were then thematically labeled and semantically compared to define framework scope, identify consensus areas, reveal gaps, and highlight specialized strengths. The analysis found ESF to be the most comprehensive, while S2C2F, SCVS, and SLSA offer significant depth in specific niches, such as consumption, component verification, and build integrity, respectively. This underscores that no single framework is universally optimal. Second, this research develops and evaluates a Proof-of-Concept (PoC) system to demonstrate the feasibility of automating compliance attestation. The PoC automatically verifies a targeted subset of decomposed guidelines for selected open-source projects, embedding cryptographically signed conformance attestations - including claims, evidence, and targets - directly within a Software Bills of Materials (SBOM). A companion visualization tool enables human inspection and signature verification of these enriched SBOMs. A feasibility study confirmed the viability of this endto- end process, showcasing a practical pathway for integrating verifiable compliance into the software development lifecycle. Ultimately, this work provides a clearer map of the current guidance landscape and demonstrates a practical path to embedding verifiable compliance, advancing the automation and trustworthiness of software supply chain security.
Bild (thumbnail)
Fine Tuning a Large Language Model for Tactical Decision Making in Level 3 Autonomous Trucks
(2025) Zhao, Yifan; Wang, Mengyuan
This thesis investigates whether a Large Language Model (LLM) can be adapted to serve as the tactical brain of a Level-3 autonomous truck through supervised fine-tuning (SFT). We first generated highway driving scenarios in the SUMO simulator, pairing each coded scenario with high-level maneuvering decisions, which include ACC set speed, time gap, lane change intent, generated by a powerful LLM. The resulting scenario-decision pairs constitute a domain-specific dataset that captures a variety of safety-critical interactions between a self-propelled truck and surrounding traffic. Three open-source modelsMeta-Llama-3.1-8B, Qwen 2.5-14B, and DeepSeek-R1-Distill-Llama-8B-are then fine-tuned with Low-Rank Adaptation (LoRA). A modular control stack separates the LLMs high-level reasoning from a low-level Intelligent Driver Model (IDM) that executes longitudinal and lateral motion, mirroring real-world practice. Evaluation of SUMO episodes showed that fine-tuning improved the quality of decisions. All models improve the achieve a high success rate. Despite the fact that the fine-tuned LLMs achieved a high success rate, we discovered that the LLMs does not fully learn a perfect set of driving strategies. The LLMs does not completely learn the truck’s lane changing strategy. As a result, the LLMs behaved somewhat clumsily in some scenarios. After fine-tuning, some unsafe decisions were eliminated, which confirms the improvement of logical consistency. The models also generate concise natural language rationales, improving the interpretability and compliance of the system. This study shows that when equipped with a tailored driving dataset and efficient LoRA fine-tuning, a modestly sized LLM can provide a degree of safe, efficient, and interpretable but not perfect tactical decisions for self-driving trucks.
Bild (thumbnail)
Bild (thumbnail)
Optimization of Fan Blade Design Using CFD and Reinforcement Learning
(2025) Ganla, Eeshan; Jakka , Sai Srinivasa Manideep; Koganti , Naga Sai Ramu; Olofsson, Aron; Olsson , Jakob
The design of fan blades has undergone significant advancements over the past century. However, it is important to consider whether conventional design approaches may unintentionally constrain the range of possible blade geometries. This paper investigates the potential of integrating Machine Learning and CFD to further improve and accelerate the fan blade design process. To achieve this, the study focuses on two main tasks: developing an accurate numerical model of a centrifugal fan in STAR-CCM+ and creating a Reinforcement Learning (RL) framework which implements the STAR-CCM+ model to optimize the fan blade geometry. CFD simulations were performed using the k − ω SST solver, and a mesh convergence study was performed. The RL framework for blade optimization was based on the Deep Q-Network (DQN) algorithm, implemented in Python using Pytorch. The CFD Model validation was carried out by comparing the performance curve obtained from STAR-CCM+ simulations with the manufacturer’s fan curve. The results indicate that the developed model accurately predicts the flow field generated by the fan. The static pressure rise across the fan serves as the primary performance metric for evaluating design improvements. The Reinforcement Learning (RL) approach successfully produced new and improved blade designs in each iteration. However, none of the generated designs outperformed the original fan blade. Despite this, the approach shows strong potential for improving blade design given more time and computational resources. For future studies, additional methods can be incorporated to better evaluate blade design. For instance, investigating other Reinforcement Learning methods or alter the current environment to reduce its design constraints.