Chalmers Open Digital Repository
Not a member yet
26247 research outputs found
Sort by
Navigation and Localization for Railway Inspection Drone in GPS-denied Environments: An Investigative Study of Modular SLAM Baselines and End-to-End Learning-based Approaches
No full textAutonomous navigation in GPS-denied environments remains a critical challenge
for unmanned aerial vehicles performing infrastructure inspection. This thesis investigates
the feasibility of learning-based navigation for railway-inspection drones
by first constructing and analyzing a state-of-the-art modular baseline and then
evaluating emerging end-to-end paradigms.
A high-performance navigation system combining FasterLIO SLAM with Fast-Planner
trajectory generation is implemented in high-fidelity simulation and used as an
analytical baseline. While the system successfully navigates dense forest environments,
controlled experiments reveal three structural failure modes—SLAM localization
drift, flight-controller tracking limitations, and planner-induced trajectory
constraints—highlighting deeper challenges such as cumulative error propagation
and real-time sensor–compute bottlenecks.
Building on these insights, the thesis conducts an experimentally grounded feasibility
study of three dominant end-to-end learning directions: predictive worldmodel
architectures, self-supervised representation learning pipelines, and visionreinforcement-
learning approaches. By implementing prototype models and stresstesting
their stability and data requirements, the study identifies several infeasible
or unstable directions—such as feature-forecasting models and self-distillation objectives—
and reveals simulator limitations that currently block scalable vision-RL
for UAVs.
Rather than delivering a complete end-to-end navigation system, this work provides
a systematic evaluation of the landscape, clarifies the fundamental obstacles facing
learning-based navigation in GPS-denied environments, and establishes concrete
design requirements and a research roadmap for future PhD-level research
Development of Graphene Related Solid Lubricant Coating
No full textTribological coatings are important for improving performance and service life by
reducing friction and wear in mechanical systems across several industries. Making
composite coatings that balance tribological and electrical properties can improve
mechanical strength, durability, and electrical performance with positive effects on
energy efficiency and material use.
Graphene is known for its good structural, electrical, thermal, and lubrication
properties. One of its derivatives- reduced graphene oxide (RGO), has a similar
structure and properties. In this work, the aim was to make composite coatings
by combining the ability of graphene- related materials with polymers and metals.
RGO- polymer composite coatings were prepared using direct application and dip
coating methods. In parallel, RGO- copper (Cu) multilayer composites were developed
using surface modification, direct application, and electroplating processes.
Among the samples, RGO- Cu coatings showed better overall results. Coatings with
one to five layers were made on steel and copper substrates.
Tribological, electrical, adhesion tests were carried out on the samples and characterization
techniques used to analyze the properties. RGO outer layers on steel
showed good friction performance but were electrically insulating. Cu outer layers
gave better electrical properties but inferior wear properties. RGO layer interfacial
bonding over Cu coating seems weak
From storage to circulation
No full textMany products are retained by households without being used. Instead, these products are kept in storage spaces without serving any direct purpose. When a product has fulfilled its intended purpose at its first owner, it is of essence that it gets back into circulation and not stuck in storage. This is important to make sure resources are used more efficiently, and in turn to create a more resource efficient way of living.
As a complement to the research project “Mining garage gold”, this master thesis aims to investigate which factors trigger, motivate, and prevent households from engaging with, and recirculating, their unused, stored products as well as how these factors are manifested in the process of engagement and recirculation. In addition, the project aims to explore how design can aid and influence households to engage with and recirculate their unused, stored products.
The thesis resulted in an extensive mapping of different triggers, motivators and barriers that exist in the process of engaging with and recirculating unused, stored products. A process flowchart, showcasing how the different factors are manifested in the households’ process of engagement and recirculation of unused, stored products, was also created. This process flowchart includes the choices households make where triggers, motivators and barriers have great influence over the decision-making process. To give examples of how design can influence and aid households in the process of engaging with and recirculating their unused, stored products, a design portfolio was created. This design portfolio consists of eight different high levelled design concepts, targeting different parts of the process. This thesis provides a new, more holistic perspective of what triggers, motivates, and prevents households from engaging with and recirculating their unused, stored products
ML assisted circuit design using active learning
No full textThis thesis explores the use of active learning (AL) to reduce the amount of training
data required for machine learning (ML) models used for circuit design by selective
sampling of the most informative data points. In this study, an uncertainty-based
AL approach was implemented. This method leverages the model’s prediction uncertainty
to selectively sample the most informative data points. A convolutional
neural network (CNN) is trained to predict scattering parameters (S-parameters)
from pixelated representations of 2-port passive microwave circuits. This method
enables the ML model to act as a fast surrogate to electromagnetic (EM) solvers.
The goal in this project is to speed up the training process for the ML models using
AL.
The performance of the AL-based model is compared to a baseline model trained
using random sampling. Evaluation is conducted on a fixed test set, as well as across
different frequency ranges and S-parameter. Results show that AL consistently outperforms
the baseline in terms of root mean square error (RMSE), particularly at
higher frequencies where EM behavior becomes more complex.
Ensemble models were also investigated to assess their potential in improving the
sampling strategy. However, they did not yield better results. Each ensemble run required
over two weeks of computation, limiting further experimentation. An ensemble
of models refers to a collection of multiple individual models whose predictions
are combined to improve overall performance and robustness.
Finally, the models were tested in a design task where a genetic algorithm generated
circuits from targeted S-parameters. The AL model achieved a 32.9% lower mean
RMSE than the baseline when comparing predicted and simulated S-parameters.
These findings highlight AL as a promising approach for improving data efficiency
in ML-based circuit design
Statistical Inference with Auxiliary Information under Block-Structured Missing Data
No full textIn medical research, a common challenge is missing data. Missing data can lead to biased
findings and loss of precision if not handled appropriately. Common methods of
handling missing data are complete-case analysis (CCA), multiple imputation (MI), or
inverse probability weighting (IPW), but these methods have drawbacks. This thesis
aims to compare these methods to the method augmented inverse probability of completecase
weighting (AIPCCW), that is less established but with certain desirable theoretical
properties. AIPCCW is an extension of inverse probability of complete-case weighting
(IPCCW), and utilises information from both participants with fully observed data and
participants with partly observed data. AIPCCW utilises two models, one for the outcome
and one for missingness, where only one model is required to be correctly specified
for AIPCCW to achieve unbiased inference.
This thesis implement and compare AIPCCW, CCA, MI, and IPCCW in different
scenarios through a simulation study and a application study on real-world data. The
scenarios cover unique combinations of sample size, proportion of missing data, levels
of correlation between variables with missing values and with an auxiliary variable, and
different missingness mechanisms.
In our experiments, the AIPCCW method demonstrate a performance in bias and
eRMSE statistically significantly better than CCA, MI, and IPCCW, in certain scenarios,
especially in simulated scenarios with a large proportion of missing data. AIPCCW is
found to significantly improve in scenarios with a higher correlation between the variable
with missing values and the auxiliary variable. On the other hand, the performance of
AIPCCW is found to not outperform CCA, MI, and IPCCW in a majority of the scenarios
that were implemented. AIPCCW performed comparable to CCA and IPCCW on realworld
data in this study, but AIPCCW could potentially perform better on real-world data
if a stronger correlation between the variable with missing data and the auxiliary variable
existed. Owing to these results, the evaluation is inconclusive to whether AIPCCW is
significantly better than CCA, MI, and IPCCW. This thesis concludes that AIPCCW is
a stable method, but does not necessarily recommend it over more common methods.
However, further research is needed
Optimizing Market Engagement: Strategic Models for District Heating Companies’ Participation in Electricity Markets
No full textDistrict heating companies in Sweden are presented with opportunities to participate
in newer electricity markets beyond the spot market, such as intraday and ancillary
markets. However, navigating these markets requires advanced and complex strategies
due to the varying market rules, market timings, and operational constraints of
combined heat and power units. This thesis develops a flexible mixed-integer linear
programming model to optimize multi-market participation for district heating
companies. The model integrates the market rules and all possible operational constraints
to determine profit-maximizing strategies across electricity markets. Simulations
using historical data showed that there is great value in participating in
one additional ancillary market, with increases in profit ranging from 35% to 1100%
depending on season. We also noticed that participation in more than two to three
markets yields less profit increase, but on the other hand an increased complexity
for daily operations in the district heating companies, suggesting that two to three
markets is a balanced amount of participation. Since the technical qualifications for
the ancillary markets are tough, many district heating companies might not have the
opportunity to participate in more than one or two such markets, which strengthens
this result. A comparison with Utilifeed’s baseline model highlights the accuracy of
our model and the added value of incorporating it in Utilifeed’s model. The results
show the importance of enabling district heating companies to navigate the complexity
of multi-market participation, improving profitability while supporting the
grid balancing
Transformer-Based Crystal Structure Generation from OTC and Chemical Composition
No full textMulti-component oxides, composed of three or more elements, offer a vast combinatorial
space of possible structures with tunable properties such as thermal stability,
ion conductivity, and catalytic activity. Exploring this space using traditional trialand-
error methods is time-consuming and expensive.
This thesis investigates the use of a Transformer-based language model to generate
Crystallographic Information Files (CIFs), which encode atomic positions, lattice
parameters, and symmetry elements. The model is trained to learn relationships
between structural features and material properties, allowing it to propose new
CIFs representing potential novel crystal structures based on input descriptors like
oxygen transfer capacity and composition.
The results show that the Transformer model can capture complex structural patterns
and generate valid CIF sequences, demonstrating its potential as a data-driven
tool to accelerate the discovery and design of multi-component oxides
Non-Isolated Battery AC Charging Using CHB Converters - Investigating Leakage Current and Short-Circuit Conditions
No full textThe demand for efficient and compact electric vehicle battery chargers has sparked interest in non-isolated charging topologies that eliminate the need for a transformer. This thesis investigates the use of a 31-level grid-connected cascaded H-bridge converter for direct AC battery charging, a concept that integrates the converter into the battery itself and enables bidirectional power flow without galvanic isolation. A simulation model was developed to evaluate the system’s behavior with respect to leakage current, touch current, and short-circuit conditions under various grid scenarios. The simulations showed excessive touch current compared to electric vehicle safety standards, which was especially high when using zero-sequence injection. This highlights the importance of minimizing parasitic capacitance. The leakage and touch current were found to be dependent on grid voltages, converter voltages, and modulation scheme. An analytical model was developed to describe the leakage current, yielding results consistent with the simulations. An approximate maximum limit for the total parasitic capacitance of the converter was calculated to be 120 nF. Short-circuit simulations highlighted the importance of grid impedance and R/X-ratio in determining let-through energy and protection requirements
Exploring Image-to-Text Visual Search Using Open Source Models
No full textVisual searching refers to the use of visual data, typically images, in order to perform
a search rather than textual input. Most visual search implementations rely
on performing similarity searching over image features, in which a user-submitted
query image is compared against all searchable entries’ features before returning
sufficiently similar results. This thesis explores a different method which utilizes
image descriptions generated by vision-language models instead of image features,
where the descriptions are converted into embeddings in order to match with other
search entries. Evaluation data indicate that the method can provide satisfactory
retrieval performance in addition to maintaining a low search query execution time,
provided that an adequate vision-language model is employed and sufficient server
capacity is available