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

Anti-Aging Accelerators - A study on the effects of GPU degradation in AI applications
(2026) Forssén, Björn
By using a model based on first principles, it is possible to argue about the sustainability effects of aging AI accelerator hardware by relaxing some constraints. But first order principles have its limits and despite this effort to reason regarding the replacement of AI accelerator hardware and significant remains to be explored. This work presents some common considerations for reasoning about sustainability with computer hardware, some of the considerations being based on current trends are based on the situation at the time of writing. Then a model based on first principles is explained and is used to reason about replacement of AI accelerator hardware, this is combined with software simulation of faulty hardware in-place of faulty hardware. Finally, we present some of the considerations learned from testing the first-order model and conclude on some suggestions for further work to study the sustainability of AI accelerators.
Learning the intrinsic value of theorems - Estimating usefulness of theorems with neural networks
(2026) Vallander, Johan
We investigate whether neural networks can learn some notion of usefulness/interestingness of theorems, or as we chose to call it “intrinsic value”. This we define as the ability to take part in the deduction of other (sufficiently “beautiful” ) theorems. To study this we first devise a definition of what constitutes a “beautiful” theorem. We then construct a symbolic system which, starting from a set of axioms, randomly deduces new theorems from existing ones. Using this symbolic system we gather a lot of beautiful theorems, and consequently theorems with intrinsic value. We experiment with different metrics of intrinsic value to find out which works best. We then use this metric together with the collected theorems to train neural networks to classify a theorem as useful or not. We find, using MPNN and DAG-LSTM architectures, that this is possible. We also find that we can optimize the discovery of beautiful theorems with the aid of these trained neural networks.
Secure Attestation Framework for Intelligent Transportation Systems - Advancing Swarm Attestation Through the Application of Homomorphic Hashing
(2026) Rengaraj, Abirami; Alihodzic, Imad
Internet of Thing (IoT) systems, and in particular intelligent transportation systems, are becoming ever increasingly large and complex. Simultaneously, such systems increasingly coming under attack from malicious parties. To aid in monitoring, and ensuring, the security of such system, one prominent solution is remote attestation (RA) - a process by which trusted entities can determine, and attest, the integrity of un-trusted devices. Whilst contemporary research has lead to the creation of numerous RA schemes, each have their own shortcomings. To address these shortcomings, this thesis proposes a new attestation scheme - FLASH: Fast Lightweight Attestation using Scalable Homomorphic Hashing. As the name suggests, this lightweight attestation scheme utilizes homomorphic hashing to aid in the aggregation of attestation results, reducing computational costs, and enabling efficient scaling across large systems. FLASH is comprised of three separate algorithms, namely two separate on-demand algorithms, and one self-attestation algorithm, all built using a common homomorphic hashing library. To verify the performance and usability of FLASH, proof of concept implementations of the on-demand algorithms have been developed and tested using real-world hardware, with the timing results then subsequently being applied to large-scale self-attestation simulations. Furthermore, result analysis of FLASH testing allowed for the evaluation of both worst-case behavior for both the PoC and simulation, as well as the creation of performance estimates in large scale networks. Together, these findings confirm that FLASH achieves its design goal of being fast, lightweight, and scalable, providing an efficient attestation framework suitable for large-scale IoT and intelligent transportation systems.
Improve Vehicle Efficiency Accuracy Through Virtual Sensors - A Comparative Study of MLP, LSTM, and Transformer Architectures with and without Physics-Informed Constraints for Fuel Consumption Prediction
(2026) Jiang, Chuyi; Huang, Yingtian
In modern heavy-duty vehicles, measuring fuel consumption relies on advanced flow meters, which are expensive and challenging to install. Volvo Group currently employs an empirical calculation model based on predefined coefficients, but this thesis explores the potential of replacing the flow meter with machine learning models. The primary objective is to develop predictive models that outperform the existing baseline. To this end, Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Transformer architectures are evaluated to determine the most suitable model for this scenario, while the impact of incorporating Physics-informed Neural Networks (PINNs) on prediction performance is also investigated. Using historical field test data, six models were trained and evaluated. All trained models demonstrated superior performance compared to the baseline. The outcomes of this thesis work have the potential to be embedded in the Electronic Control Unit (ECU) system for future field tests, providing a practical and cost-effective alternative to the physical flow meter.
Development and Tuning of a Finite Element Average Human Hand Model: To support research and development of healthcare products for medtech applications
Maskova, Elin
The human hand is an essential part in our daily lives and a vital sense in how we feel and interact with the environment around us. However, ergonomic discomfort and pain remains prevalent in workplace and healthcare settings. To tackle these challenges, computer models of anatomical human hands can provide valuable insight into hand structure and object interaction that can support medical product development. This thesis documents the development of an anatomically based finite element (FE) average human hand model designed for applications within research and development for MedTech products. The developed model incorporates derived skin geometry from magnetic resonance imaging (MRI) with open-source skeletal components to match a statistical average hand size. The meshing strategy focuses on a hexahedral model which can aid in future morphing of the geometry. A tetrahedral model was also constructed but only for a comparative study against the hexahedral model. Different constitutive models were investigated for the skin and soft-tissue behaviours, primarily consisting of viscoelastic and hyperelastic material models. Joints were modelled using a kinematic approach, using constraint-based rigid wire connections to enable biofidelic positioning. Parameter tuning was performed with published experimental data that tested the finger pulp compression at selected loading rates of 0.1 and 0.3 mm/s. Simulation results depicted similar mechanical behaviour as experimental tests but an additional dataset would be required to fully validate the model. The study outlines the steps taken along with the models used to construct a feasible anatomical human hand FE model for product interactions. Although the current model has its limitations, a single size and simplified anatomical structure (no muscles, ligaments or tendons modelled), it lays a good foundation for future morphing capabilities and a broad analysis for healthcare product investigation and development.