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 publicerade
- Evolution Strategies as an Alternative to Reinforcement Learning in De Novo Molecular Generation - Evaluating Distribution Based Gaussian Evolution Strategies in REINVENT vs. Policy Based Reinforcement Learning on the Practical Molecular Optimization Benchmark(2026) Eklöv, MalvaDe novo molecular generation examines how computational methods can propose novel drug candidates by scoring generated molecules against desired properties and updating generation toward higher scoring regions. One approach trains a SMILES based language model and fine tunes it toward such regions. Fine tuning is commonly performed with policy gradient reinforcement learning (RL), as in AstraZenecas highly optimized REINVENT platform. An alternative is evolutionary strategies (ES), where a population of models is created and their parameters are updated to bias generation toward higher scores. This thesis investigates how distribution based ES compares with RL when implemented in REINVENT and evaluated on the Practical Molecular Optimization benchmark. OpenAI-ES shows near competitive performance on scoring and diversity metrics, whereas variance estimating Natural ES methods perform poorly. Variations of fixed variance ES are explored, and a novel sampling technique that biases generation toward high diversity shows promising performance across multiple metrics.
- Data-Driven Requirement Development - From Field Data to Reliability Requirements: Identifying and Analyzing High-Performing Automotive ECUs(2026) Wang, YiminThis thesis investigates a data-driven methodology for reliability-oriented requirement development in automotive electronics. Traditional automotive reliability engineering is primarily failure-focused, relying on warranty analysis and reactive investigation of defective components. However, such approaches provide limited understanding of why certain Electronic Control Units (ECUs) consistently demonstrate strong field reliability performance. The research initially aimed to analyze relationships between ECU operational conditions and field reliability behavior through classical data-driven analysis. However, the required centralized operational-condition dataset was not available within the industrial data environment. Consequently, the study evolved toward a practical industrial screening methodology based on available enterprise engineering and field-quality data resources. The proposed framework integrates multiple industrial datasets, including the KDP Engineering Database (KDP), Quality Follow-Up (QFU) warranty repair records, the Early Warning System (EWS), and procurement-related production volume data. By combining field repair occurrence with market exposure normalization, ECU populations with exceptionally low repair occurrence relative to deployment volume were identified. Selected ECUs subsequently underwent hardware-oriented engineering investigation, including open-lid assessment, Printed Circuit Board (PCB)-level visual inspection, and review of available Design Verification (DV) and Product Validation (PV) documentation. The investigation focused on identifying recurring robustness-related engineering characteristics rather than performing direct failure analysis. Observed features included reinforced PCB mechanical support structures, controlled PCB cleanliness, environmental protection strategies, and evidence of robustness-oriented validation practices. The work demonstrates how multiple industrial engineering and field-quality data sources can be systematically combined to support evidence-based reliability investigation and practical reliability requirement development in automotive electronics.
- Existence of Static Solutions to the Vlasov-Poisson and Einstein-Vlasov Systems as Fixed Points of a Mass-Preserving Algorithm(2026) Lehmann, LudvigThis thesis develops a new method for proving the existence of static, spherically symmetric solutions to the Einstein–Vlasov system by reformulating a masspreserving algorithm as a fixed-point problem. Building on earlier work for the Vlasov–Poisson system, we first extend an existing fixed-point existence proof from isotropic to anisotropic solutions. We then establish the corresponding result for the Einstein–Vlasov system, using a simple isotropic ansatz.
- Large-Scale Noise Simulation in Urban Environments Using a Preconditioned Helmholtz Solver(2026) Sahakyan, MarinaThis thesis studies the finite element solution of the time-harmonic Helmholtz equation for acoustic wave propagation in urban environments. The discretized Helmholtz equation gives rise to large, complex-valued, indefinite linear systems, for which standard iterative methods may converge very slowly. The aim of this work is to implement and evaluate a preconditioned GMRES solver for such systems, with focus on solver performance, parameter choices, and practical memory limitations. The solver uses a two-level restricted additive Schwarz preconditioner with a spectral coarse space, following the RAS/MS-GFEM approach of Ma, Alber, Scheichl, and Zhang. The implementation is first verified on an exact plane-wave problem, then tested on a controlled rectangular box domain, and finally applied to realistic city-domain meshes representing part of Gothenburg. The experiments investigate how mesh resolution, frequency, and the choice of coarse space affect convergence and computational cost. The results show that the two-level coarse correction considerably improves GMRES convergence compared with a one-level method and with plain GMRES without preconditioning. On the box domain, the method gives stable algebraic convergence up to 30 Hz, although the acoustic field is not fully mesh-converged at the higher frequencies. On the city geometry, the 5 Hz case gives the clearest mesh-refinement behaviour, while the 10 Hz, 20 Hz, and 30 Hz cases become increasingly demanding. The largest city runs show that explicit storage of the coarse basis can become memory-limiting. The study therefore shows that the two-level spectral Schwarz preconditioner is promising for large Helmholtz problems in urban acoustics, but that higher-frequency simulations require careful choices of mesh resolution, coarse-space size, and memory-efficient implementation. Keywords: