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.

Visar 1 - 2 av 2

Senast publicerade

Data Driven Insights in Perioperative Workflows
(2026) Ahlgren, Karl; Stoopendahl, Love
Operating room (OR) workflows are characterized by complex material flows, strict time constraints, and high coordination demands. Disruptions in perioperative ma terial preparation, particularly during the picking phase performed by OR nurses, represent an important but underexplored source of inefficiency. This thesis investigated how material-related OR workflow data can be used to ana lyze workflow behavior, identify disruptions and inefficiencies, and generate insights relevant for decision support. The study was exploratory and based on shadow ing data from a Swedish hospital, complemented by semi-structured interviews and, where necessary, synthetic data. The work was conducted in collaboration with Mölnlycke Health Care. The findings suggest that disruptions in the picking process are closely connected to information fragmentation, unclear material availability, reliance on tacit knowl edge, and changes in the surgical schedule. The quantitative analysis illustrated how variables such as picking duration, interruptions, waiting time, information search, perceived complexity, and staff experience could describe workflow varia tion. However, due to the small sample size and data limitations, the results should be interpreted as exploratory and indicative, rather than statistically generalizable. Process mapping, visualization, and process mining demonstrated how material centric workflows can be made more visible and interpretable for different stakehold ers. The thesis contributes a framework for understanding perioperative material workflows and highlights the need for high-quality structured data to support fu ture workflow analysis, stakeholder-adapted visualization, and clinically meaningful decision support.
Development of Accelerated Life Testing Method for Bone Conduction Devices
(2026) Azadi, Sam; Larsson Rosén, Viktor
This thesis investigates the development of an accelerated lifetime testing (ALT) methodology for Cochlear’s bone conduction hearing devices (BCDs). An ALT is a testing method in which a product is exposed to higher-than-normal stress conditions to accelerate ageing and failure mechanisms. The purpose is to evaluate long-term reliability and predict product lifetime within a significantly shorter testing period. The work was motivated by Cochlear’s need for a robust, data-driven ALT methodology. The implementation of this framework provides significant engineering and business value by reducing testing costs, accelerating time-to-market for future products, and supporting verification of regulatory compliance standards. Several theoretical ALT methods and models were evaluated. The aim was to develop theoretical acceleration models together with practical experimental methods capable of identifying long-term performance degradation caused by age-related failure modes in adhesives, nylon thread-locking materials, and piezoelectric components. An ALT concept based on thermal and electrical overstressing was selected. Individual overstress limits for each component were then determined through tailored practical experiments designed around the specific failure mechanisms of each component. A total of 16 experiments were conducted, with a total sample count exceeding 80. The experiments identified safe accelerated operating ranges of 2-5 VRMS and 40-70 °C. Due to time constraints, obtaining long-term ALT degradation results was outside the scope of this thesis. However, validation tests was initiated using conservative inputs of 4.3 VRMS at 70 °C to verify the functionality and stability of the developed test setup. The developed ALT framework contributes to more reliable predictions of long-term actuator performance and supports earlier validation of future design changes. Lastly, the methodology is highly scalable, allowing multiple actuator variants or models to be tested simultaneously.
Camera-Based Home Rehabilitation Exercise Monitoring: A Technical Evaluation Against Optical Motion Capture
Zhao, Tingting
Home-based rehabilitation often relies on patients performing prescribed exercises independently, without continuous supervision from a physiotherapist. A digital solution that can monitor the exercise performance and provide feedback to the patients would be valuable to improve the follow-ups and support patient empower ment. Camera-based markerless pose estimation may provide a practical and low cost way to monitor exercise quality in such settings. This thesis investigates the fea sibility of using a single RGB camera and MediaPipe Pose for selected rehabilitation oriented exercises. Three exercises were evaluated: single-leg stance, sit-to-stand, and mini-squat. RGB videos were recorded using an iPhone 13, while reference motion data were col lected with a Qualisys optical motion capture system. MediaPipe Pose was used to extract body landmarks from the videos, and exercise-specific metrics were com puted, including trunk orientation, normalized pelvis displacement, squat depth, movement timing, and knee flexion. To enable comparison with MediaPipe, the three-dimensional motion-capture data were projected onto corresponding frontal or sagittal analysis planes before matched metric definitions were applied. The results show that the MediaPipe-based pipeline could estimate several selected metrics with small to moderate errors under controlled condition 1 setup which is camera-to-participant at 3 m with normal room lighting. Trunk-orientation and normalized displacement metrics were generally more stable than two-dimensional knee-flexion estimation. The results also showed that performance depended on the exercise, metric definition, participant, and recording setup. Camera distance and lighting affected the metrics differently, and repeatability was generally stronger within the same session than across different days. These findings indicate that MediaPipe-based rehabilitation monitoring should be interpreted at the metric level rather than as a uniformly accurate motion-analysis solution. A rule-based feedback prototype was also developed to illustrate how pose-derived metrics could be translated into patient-facing feedback and therapist-facing session review outputs. Overall, the findings suggest that MediaPipe Pose can be a use ful low-cost component for selected home rehabilitation monitoring tasks, provided that the exercise, camera view, and metric definitions are carefully chosen. Further validation with larger and more realistic datasets is required before clinical or real home deployment.
Thermal Modelling of Battery Characterization Techniques
(2026) Enochsson, Anton; Lindstad, Simen
The goal of this master’s thesis was to calibrate a calorimeter in a vacuum environment and quantify its potential benefits compared to air. The motivation for the project was to record the heat from a cylindrical cell with high accuracy. Peltier elements were used to record the heat flow of the samples’ cubical enclosure, each side with a heat sink to maintain a stable cold-side reference. The initial testing proved unstable and problematic to replicate, motivating four iterations of the setup. Vacuum conditions proved stable, yet time-consuming, requiring 6 hours to reach a steady state, compared to under 2 hours in air. The sensitivity of the Peltier elements was found to be 50-100 mW, yet the full system could only achieve a stable calibration coefficient from 1 W, found to be 19.93 and 19.85 W/V for air and vacuum, respectively. The main challenges with the calorimeter in vacuum were the high correlation to the surrounding lab temperature and the resulting radiation onto the setup. The added isolation proved to be inefficient. The heat sink’s surface was heated up by radiation, corrupting the heat flow signal. The recorded voltage was corrected using a reference cell and baseline voltage. The digital-twin model reached a 7% deviation, which helped locate the parasitic convection through the power cables, accounting for around 21% of the applied power. It also quantified 170-190 mW of radiative heating onto the heat sinks and 30-80 mW of conductive heat from the plastic stand. Ultimately, the vacuum environment proved challenging with no significant benefit over air. Although convective heat transfer was successfully removed, further changes could potentially improve the sensitivity and stability.
Interpreting Machine Learning Models using Conditional Counterfactual Generation
(2026) Martinsson, Samuel
With the rapid development and application of complex machine learning models, the need to interpret the internal processes of such models have become increasingly relevant. In this thesis, a novel method for interpreting black box machine learning models is proposed, where an autoencoder is used to generate reconstructions of data to visualize in an interpretable way what patterns a model has learned to detect. The method is first shown to work for a simple constructed problem, being able to interpret a model that has learned to predict the mean of an underlying normal distribution from samples. It is then evaluated for a more complex problem, where a model has learned to classify the existence of disease in images from the CheXpert dataset of X-ray images. It is demonstrated that naively implementing the method to interpret this model leads to the autoencoder generating adversarial patterns to trick the model, instead of showing the an interpretable explanation of what the model has learned. To mitigate this issue, the thesis explores adding an additional model in the latent space of the conditional autoencoder and demonstrates that this can provide a certain degree of interpretability. Because of this, the method shows promise for interpreting black box models and with further research it might become viable for practical use.