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
Optimization of estimating intrinsic parameters of a camera using a robot arm
(2025) Sudhir, Shilpa
Eye-tracking technology is applied across industries such as automotive, aerospace,
and human-computer interaction, where precise gaze estimation relies on the accurate
calibration of eye-tracking cameras. In this work, a robot arm is used to perform
the intrinsic calibration of eye tracking cameras. The motivation behind using a
robotic arm for the intrinsic calibration of an eye-tracking camera instead of a manual
procedure stems from several key advantages related to precision, repeatability,
and efficiency. Control over the camera poses in front of the checkerboard leads to a
more accurate and efficient calibration process. This eliminates the challenges and
inconsistencies associated with manual camera movements and opens the door for
automating and scaling the calibration process with minimal human involvement.
The robotic arm ensures a level of consistency and reliability that would be hard to
achieve manually, making it a compelling choice for eye-tracking camera calibration,
especially in high-precision applications.
The Design and Performance Evaluation of a Double-three-phase Inverter - DC-Link Thermal Evaluation and Lifetime Estimation
(2025) Yodwong, Jedsada
This thesis presents the design, implementation, and experimental evaluation of a double–three–phase traction inverter based on silicon carbide (SiC) power modules for heavyduty
electric vehicle applications. The work focuses on achieving high power density and robust thermal performance through co-design of the DC-link capacitor bank and laminated busbar structure. Two alternative DC-link layouts were investigated: Case I, with capacitors mounted above the busbar, and Case II, with capacitors in direct contact with the housing.
A comprehensive methodology combining analytical modeling, MATLAB/Simulink simulations, and laboratory testing was employed. Open-loop SVPWM tests validated inverter
functionality, while thermal performance was assessed under RMS ripple currents of 156 Arms and 300 Arms at 10 kHz under 2 thermal conditions: with and without liquid cooling.
Results show that liquid cooling significantly reduces busbar temperatures (up to 67% in Case I), whereas capacitor hot-spot remain dominated by internal thermal paths, limiting
improvements to 18–21%. Lifetime analysis, based on datasheet models, experimental data, and MATLAB/Simulink model, indicates that capacitor lifetime is highly sensitive to temperature. Under ideal assumptions, the lifetime at 55 ◦C ambient was estimated at 24.8 years, whereas experimentally adjusted scenarios ranged from 8.7 to 13.1 years. Liquid cooling improved lifetime by up to 50% at high ripple currents compared to the worst case, emphasizing the importance of accurate thermal modeling and integrated cooling strategies for next-generation traction inverters. The findings underscore the importance of integrated electro–thermal design for reliability in high-power SiC-based traction inverters. Future work includes closed-loop control implementation, and EMI/EMC characterization.
Is This Data Point In your Training Set? Similarity-based Inference Attacks: Performance Evaluation of Range Membership Attacks to Audit Privacy Risks in Machine Learning Models
(2025) Nawrin Nova, Sifat; Shubham, Saha
Membership Inference Attacks (MIAs) pose a serious threat to the privacy of machine learning (ML) models by determining whether a specific data point was used during model training. A recent and powerful variant, the Range Membership Inference Attack (RaMIA), assesses privacy risks over a range of semantically similar data points. However, its practical application is limited by a high query overhead, as it requires querying the target model for every sample in the range. This thesis proposes and evaluates a novel approach designed to overcome this limitation by combining range queries with group testing principles to reduce the number of queries sent to the target model without losing the attack performance and making the attack more stealthy. Instead of testing every sample, this method first groups similar data points based on their extracted features and then queries only a small number of strategically chosen representatives.
All the experiments are conducted on the CIFAR-10 dataset, comparing its performance against the standard RaMIA baseline. The results demonstrate that RaMIA with group testing successfully reduces the number of queries by 84% in a setting of 50 augmentations. This work reveals that even minor enhancements in query design and decoding strategy can lead to substantial gains in auditing. Moreover, we provide practical recommendations for tuning key hyperparameters and integrate our attack into the LeakPro framework for reproducibility and broader adoption in privacy auditing of ML models.
Field measurements of airborne sound insulation; Differences in results for methods prescribed in current standards at different degrees of furnishing and a possible new approach
(2025) von Schmalensee, Marcus
There is a widespread belief among acousticians that the airborne sound insulation
measured between empty rooms systematically results in lower values than when
furnished. No studies investigating the phenomenon were found and since inspection
measurements are mostly performed in empty rooms, the difference, if it exists,
would need to be quantified and addressed. Since regulations in Sweden, in many
cases, require measurements down to lower frequencies than in other countries, and
the theoretical framework on which existing methods are based is generally less
valid at lower frequencies, there is reason to believe that the problem could be more
common in Sweden.
In addition, a variety of measurement methods are described in current standards,
of which only a few are regularly used. No studies have been found where the
methods are systematically tested and compared. Methods based on measurements
and further processing of impulse responses, using deterministic signals, are very
useful in certain situations but are, to the author’s knowledge, almost never used.
This is probably mainly because methods based on manual scanning paths with a
hand-held analyzer are more time-efficient in field measurements.
The main objective of this study is to identify whether furnishing leads to systematic
deviations in measured sound insulation. This is achieved by testing and comparing
several of the methods described in the current standards. A possible explanation
for the error could be that the sound pressure levels are measured with the loudspeaker
positioned in the source room, while the reverberation times are measured
with the loudspeaker in the receiving room. There is reason to believe that the
differences in the sound fields for these two cases reduce the compatibility between
these measurements, especially at low frequencies.
Since methods based on impulse responses measured with swept-sine signals provide
superior signal-to-noise ratios that far exceed the other methods, the reverberation
time can possibly be measured without moving the loudspeaker from the source
to the receiving room. If that would be possible, both the sound level differences
between the rooms and the reverberation time, i.e. all measurements required to
evaluate the sound insulation, could be measured from a single sine sweep. If a
foldable measurement stand was developed, which quickly could spread out a number
of microphone positions in a space, field measurement times could be drastically
reduced. However, the method where the reverberation time is measured with the
loudspeaker in the source room was shown to perform significantly worse than default
methods. Nevertheless, the method can potentially be used for simpler survey
measurements, without further development. Systematic differences, between furnished
unfurnished rooms, were demonstrated for the weighted standardized
level differences. However, contrary to what was expected, the differences do not
arise as a result of deviations for the lower 1/3-octave bands. Rather, the differences arise
as a result of deviations for the 1/3-octave bands 200 Hz, 250 Hz and
for some higher frequency bands. Hence, the results do not support the suspicion
that measurements according to the Swedish regulations, which includes frequency
bands below 100 Hz, result in systematic deviations greater than what would be the
case in other countries.
Acoustic Source Localization for an Indoor Pass-By Measurement System; A Beamforming Approach Using a One-Dimensional Sparse Microphone Array
(2025) Seger, Gustav; Sehic, Semir
Accurate source localization of acoustic sources is critical for vehicle noise analysis. This
thesis examines the possibility of using sparse microphone arrays, which are part of an indoor
pass-by measurement rig, as acoustic cameras in a near-field scenario. Four different
techniques were implemented, validated and comparatively analysed. Three beamforming
techniques, Delay and Sum (DS), Minimum Variance Distortionless Response (MVDR) and
Functional Beamforming (FBF). The fourth technique is a post-processing algorithm, A
Deconvolution Approach for the Mapping of Acoustic Sources (DAMAS). Each technique was
evaluated based on spatial resolution, robustness to noise, array element imperfections and
performance on real measurement data.
The DS beamformer demonstrated robustness to array element sensitivity variations and
placement errors, but was limited under low Signal-to-Noise Ratio (SNR) conditions due to
its broad mainlobe and elevated side-lobe levels. MVDR and FBF, both implemented using
Cross Spectral Matrices (CSM), offered improved directional accuracy and noise suppression
capabilities. FBF introduced a tunable order factor which enhances control of side-lobe
suppression and mainlobe width, making it particularly useful as a pre-processing step for
DAMAS. The DAMAS algorithm was implemented as a post-beamforming processing tool,
offering improved spatial resolution, although its performance was based heavily on the quality
of the beamformer input.
Validation included both simulations and real measurements conducted inside a semi-anechoic
pass-by test rig, using stationary vehicle signals. All techniques successfully localized tonal
sources across a frequency range of 40 − 1600 Hz. However, the spatial resolution remained
constrained by array aperture size and wavelength. Ground reflections introduced interference
patterns that reduced accuracy at certain frequencies, mimicking the acoustic behaviour
expected in a road-like environment.
The thesis concludes that combining MVDR and FBF for initial localization and DAMAS
for source separation provides a flexible approach for acoustic source localization. The importance
of accurate propagation and time delay modelling, calibration and signal processing
was highlighted as a key factor to achieving reliable beamforming results in vehicle acoustic
pass-by measurements