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 publicerade

  • Bild (thumbnail)
    Multimodal Retrieval Augmented Generation via Embedding Space Alignment
    (2026) Albers, Oskar
    Retrieval-augmented generation (RAG) systems answer questions by first retrieving supporting evidence and then generating a response conditioned on that evidence. In visually rich document collections, such as Swedish municipal PDF archives, relevant information is often spread across text, maps, diagrams, and photographs. Many deployed pipelines are built around strong text embedding models like E5, and fully migrating to a new multimodal retriever can be costly and disruptive. This thesis investigates whether visual retrieval can be added to an existing E5- based text pipeline by aligning image representations to its embedding space, instead of re-embedding the corpus with a unified multimodal model. Concretely, a neural “bridge” is trained to project SigLIP image embeddings into the E5 text space, using a staged procedure (MAPS) that gradually shapes the target space, leverages paired question–image and question–text examples, and learns calibrated joint ranking between images and text. A visual-question-answering style benchmark is constructed from Swedish municipal documents, where each natural-language question is paired with the images or text passages from which it is answerable. On a corpus of 235,653 documents, 901,571 text chunks, and 427,843 images, naïve combinations of E5 and SigLIP scores fail to retrieve relevant images, yielding essentially no high-ranked image hits. The learned bridge, in contrast, raises image-only retrieval performance from single-digit to high-20% top-ranked hit rates and achieves comparable gains in a joint image–text setting, while while largely preserving text retrieval beyond the very top rank. These results suggest that embedding-space alignment could be a practical strategy for extending text-centric RAG infrastructure with visual search over document images, without replacing the underlying text encoder or index.
  • Bild (thumbnail)
    Moving HSB Living Lab: A Reuse Evaluation Framework for a Preliminary Assessment of the Reuse Potential of Building Products
    (2026) Bäckström, Alexandra; Alshara, Zainab
    This thesis investigates how to make a preliminary reuse evaluation of used building components, decoupled from a receiving project. In order to test the evaluation a modular residential building is used as a case study. The objectives of the project are to define how different parameters affect the reuse potential of the building products, and to use this knowledge to compare reuse scenarios. Due to existing research gaps on how to evaluate relevant reuse parameters, a framework is developed, suggesting what parameters are relevant, how to evaluate them and how to aggregate the results. The project highlights the great complexities of evaluating and reusing building compo nents, drawing insights from both literature and stakeholder interviews. The first part of the outcome of this project include the first iteration of a preliminary Reuse Evaluation Framework, designed to guide future reuse practices. The evaluation is based on grad ing various product properties to acquire scores for the key criteria Recovery potential, Environmental impact and Market readiness. Thereafter weights are assigned to each criteria, based on the preferences of the decision maker, giving a final reuse potential. The second part of the outcome include an inventory of the case study, as well as the application of the framework through the evaluation of selected building components in two different reuse scenarios. The studied scenarios are: 1. reusing the whole modules of the building (relocating the building); 2. reusing the individual building products. Based on the assumptions made in the study, the results show that reusing individual building products give higher reuse potential for most evaluated products, as compared with reusing them as part of the modules. This is in part due to the low market readiness and adaptability of the modules. Ultimately, which scenario is preferable also depends on circumstances that are outside of the scope of the evaluation framework, such as site-based building regulations. However, the results can indicate what to consider and what direction to take in a reuse project.
  • Bild (thumbnail)
    Multi-Layered In-Vehicle Communication Architectures
    (2026) El Hage, Ali; Qian, Mattias
    Modern bus platforms contain an increasing number of electronic control units (ECUs) distributed over long physical distances. This creates challenges for the in-vehicle communication architecture, since long cable routes, increasing traffic demands, and stricter timing requirements may reduce communication robustness and increase end-to-end delay. Controller Area Network (CAN) is widely used in vehicle systems due to its robustness, simplicity, and established industrial use. However, in long vehicle platforms, CAN-based communication may approach practical limitations related to cable length, signal integrity, and transmission delay. This thesis investigates how a multi-layered in-vehicle communication architecture can be modeled and optimized by combining different communication technologies in a common bus architecture. The considered technologies include CAN, Ethernet, and AUTBUS. The objective is to determine where different technologies should be used and where converters should be placed in order to balance total system cost and communication delay. A mixed-integer linear optimization model is developed to assign communication technologies to discretized bus segments and to determine the required gateway placements. The model includes cable costs, gateway costs, gateway delays, transmission delays, propagation delays, ECU interface constraints, and a maximum consecutive CAN length constraint. Since cost and delay are conflicting objectives, an epsilon-constraint method is used to generate Pareto-optimal architecture alternatives. A secondary fine-tuning model is also introduced to refine boundary positions after the discrete topology has been selected. In addition to the mathematical model, physical gateway latency measurements are performed using a CANcase, CANalyzer, and CAPL scripts. The measurements show that gateway latency is not constant, but varies depending on transmission interval and network traffic. The measured latency is therefore used as a representative modelling parameter rather than an exact deterministic value. The results show that optimization framework can identify meaningful trade-offs between cost and delay in a long bus communication architecture. The Pareto front provides several feasible design alternatives, ranging from low-cost solutions with higher delay to more expensive architectures with lower delay. The fine-tuning step further reduces cost for selected Pareto points, especially in the higher-cost region of the frontier. Overall, the thesis demonstrates that mixed-technology communication architectures can provide a structured approach for improving timing performance in long vehicle platforms, provided that gateway behavior, protocol constraints, and implementation complexity are carefully considered.
  • Bild (thumbnail)
  • Bild (thumbnail)
    Chemical Looping Gasification of Biomass in a 20 kWth pilot plant
    (2026) Landgren, Lina
    The increasing demand for carbon neutral fuels to reduce carbon emissions highlights the need for novel, efficient, and scalable processes for converting low grade biomass into bio fuels. This study investigates the influence of operating conditions on biomass gasification for methanol production using Chemical Looping Gasification (CLG). Experiments were conducted in a 20 kWth pilot plant using ilmenite as the oxygen carrier. The effects of air supply, operating temperature, and gasifying agent composition on process performance were investigated using two different biomass fuels: Waste Wood type B (WWB) and Microalgae (MA). An increased air-to-fuel ratio resulted in a higher degree of combustion, which reduced the concentrations of syngas components (CO and H2) and consequently decreased syn gas production. A decrease in temperature reduced the fuel conversion rate, resulting in a greater fraction of carbon being converted in the Air Reactor rather than the Fuel Reactor. Comparing the achieved gas equilibrium and the theoretical according to the water gas shift equilibrium indicates a temperature dependent catalytic effect the final gas composition. An increased fraction of CO2 in the gasifying agent did not significantly affect fuel reactivity or conversion. However, it increased the amount of carbon introduced into the system and consequently increased the fractions of CO and CO2 in the product gas. Comparison of the two biomass fuels indicates that WWB provides more stable operation and produces lower concentrations of pollutants in the form of NOx than MA. WWB therefore appears to be the more favorable fuel under the investigated conditions.