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
Channel models for optical transmission systems with polarization dependent losses (PDL)
(2025) Minoshma, Meena
Polarization Dependent Loss (PDL) presents a important transmission impairment in coherent
optical communication systems, originating from components such as wavelength
selective switches and reconfigurable optical add-drop multiplexers. This thesis develops
advanced channel models to characterize the accumulation of PDL across multi-span optical
systems, addressing signal propagation, noise interactions, and capacity constraints.
Utilizing Jones matrix formulations and singular value decomposition, the study models
individual PDL elements and extends the approach to multi-span configurations through
recursive signal evolution, incorporating additive white Gaussian noise (AWGN) from
amplifiers.
Statistical analysis reveals sub-linear PDL growth following Maxwellian distributions,
confirmed via Monte Carlo simulations. Noise properties exhibit PDL-induced anisotropy
in covariance matrices, quantified by eigenvalue ratios, underscoring performance degradation
in long-haul networks. Capacity limits for Gaussian signals indicate losses in
high-PDL scenarios.
An adapted capacity-achieving scheme, inspired by recent advancements, employs a
universal precoder with Linear Minimum Mean Square Error Successive Interference
Cancellation (LMMSE-SIC) to transform channels into scalar AWGN subchannels,
reducing signal-to-noise ratio (SNR) penalties. Simulations demonstrate enhanced
performance compared to standard multiple-input multiple-output (MIMO) approaches,
with notable reductions in outage losses.
These methodologies provide insights for enhanced system design for reliable long-haul
communication. Future investigations could explore nonlinear effects and insertion loss
variations, to enable more competitive optical networks in the future.
Large Language Model Empowered Deep Reinforcement Learning to Support Prescriptive Maintenance
(2025) Luo, Jinxia
To enhance the reliability and efficiency of complex industrial systems, it is critical to adopt approaches that transcend traditional reactive and preventive maintenance methods. Prescriptive maintenance, by leveraging advanced artificial intelligence (AI) techniques to recommend and optimize specific maintenance actions before failures occur, has emerged as a vital strategy for modern industries to reduce downtime and operational costs. In this context, Deep Reinforcement Learning (DRL) has demonstrated significant potential in prescriptive maintenance tasks, enabling autonomous optimization of maintenance strategies through self-learning and adaptive decision-making. However, in real-world factory maintenance scenarios, DRL models frequently face challenges like cold-start problems, imprecise judgments, and a lack of flexibility in complicated dynamic contexts. These challenges prevent them from being widely used in complicated and highly dependable systems.
To address these limitations, this paper proposes an innovative smart maintenance framework that effectively combines cutting-edge Large Language Models (LLMs) with DRL approaches. The framework leverages the powerful capabilities of LLMs in domain knowledge comprehension, semantic reasoning, and knowledge transfer to convert complicated, real-time, structured maintenance data into high-dimensional state representations that DRL models can easily use. The LLMs provide the DRL agents with rich previous knowledge and decision support by automatically extracting important attributes and identifying possible failure patterns.
We conducted a case study at a lab-scale drone manufacturing plant, and the results showed that the proposed framework significantly improved the accuracy of smart maintenance decisions. Compared with previous studies[1], the average reward of traditional deep reinforcement learning algorithms was 0.65, while after introducing large language models, the average reward steadily increased to 0.82. In addition, fluctuations in the early stages of learning were markedly reduced, resulting in a smoother overall training process. Furthermore, we designed multiple sets of cross-experiments with different DRL algorithms and LLM models, all of which outperformed those in previous studies[1], further validating the effectiveness and generalizability of the proposed framework.
The LLM-DRL integrated smart maintenance framework presented in this study not only refines and improves smart maintenance decision-making but also aligns with Industry 4.0’s broader objectives of operational efficiency, flexibility, and sustainability. Our research provides new theoretical and technological perspectives for advancing maintenance intelligence, enabling greater autonomy, improved generalization, and broader industrial applications.
Generalizability of Representation Learning on Downstream Tasks
(2025) Levinsson, Anton; Wang, Ziyuan
We study the theoretical generalizability of representations learned by contrastive learning by analyzing their performance in downstream linear regression tasks. To quantify the quality of these learned features, we provide rigorous proofs for the worst-case and expected downstream performances under specified assumptions. These results offer theoretical results for analyzing the quality of features by looking at downstream tasks. To empirically verify the theory, we conduct experiments on both simulated and real-world data, following time-contrastive learning (TCL) strategies, which solve the problem of nonlinear independent component analysis (nonlinear ICA). In the real-world setting, we construct nonlinear source separation tasks using mixed audio signals from different instrument categories. Then, we propose some specific downstream tasks to verify our theories using our learned features from the TCL method and compare with the observed features. The results show that most of the defined downstream tasks are located in the confidence level of the expected performance and are bounded by the worst case, which indicates that our theory aligns with the real-world setting.
Designing Infotainment Interfaces to Enhance Electric Taxi Drivers Charging Understanding
(2025) Sjöblad, Sofia; El Jabaoui, Adam
As the shift toward battery electric vehicles (BEVs) accelerates, electric taxi drivers are facing new challenges, with long operating hours having a direct impact on the number of times the vehicle needs to be charged. The achievement of optimal charging is influenced by a variety of factors, such as the state of charge of the battery, the temperature, and the charging capacity, which are not necessarily common knowledge for many drivers. This thesis investigates how in-car user interfaces can better support electric taxi drivers in understanding and managing the charging process to achieve efficient fast charging.
A user-centered research approach was used to explore the barriers to efficient fast BEV charging. The process involved literature reviews, interviews with taxi drivers and technical experts, surveys, and benchmarking of existing vehicle interfaces. Personas, user journey maps, and a requirement list were developed to ground the design process in real world needs. Based on this foundation, initial concepts were conceived through collaborative brainstorming and their feasibility was evaluated with experts. These concepts were further developed into low-fidelity prototypes and refined in three design iterations, incorporating user feedback from both electric taxi drivers and regular BEV drivers.
Through the findings, eight design recommendations were formulated aimed at guiding the development of user interfaces designed for BEV charging. These include principles such as providing contextual and transparent information, reducing cognitive load by unifying and prioritizing key charging metrics, and ensuring interface flexibility. Although the proposed solutions were tailored for electric taxi drivers, the insights reflect broader challenges relevant to the larger BEV user base.
This thesis contributes to research on the user experience of electric vehicles, highlighting how interface design can enable drivers to make informed charging decisions. The resulting recommendations can serve as a foundation for future infotainment interface development, both in commercial and private BEV contexts.
Engineering an Efficient Implementation of Pagh’s Algorithm for Sparse Matrix Multiplication
(2025) Ge, Shuhao; Kliemann, Paul
This thesis explores approximate matrix multiplication through the algorithm proposed by R. Pagh in Compressed Matrix Multiplication (2013) and its implementation developed by J. Andersson and M. Karppa. We identify memory bandwidth as the primary performance bottleneck and implement support for handling sparse inputs and outputs.
We further introduce several optimizations to improve cache utilisation. While fitting the input data into the L3 caches of our CPU did not improve L3 hit rate, it did lead to the improvement in the L1 hit rate compared to the dense implementation. Our experimental evaluation demonstrates modest yet consistent runtime improvements, particularly on large-scale and sparse inputs, while retaining the mathematical guarantees of the original implementation.
Despite these optimisations, our implementations remain outperformed by the stateof- the-art libraries on all tested real-world datasets. However, our modifications reduce the asymptotic runtime of the compression step and significantly decrease memory usage when processing sparse matrices.
Based on our findings, we suggest potential directions for future research, including further optimization strategies. These should primarily focus on reducing memory traffic, as improvements in memory locality alone are unlikely to yield substantial gains.