https://research.chalmers.se/ 2026-05-14 21:57:01 Visar de 30 senaste forskningspublikationerna för Chalmers https://research.chalmers.se//Images/chalmers_bldmrk.jpg Electrification of passenger vehicles is an important measure to decarbonize the transport sector. An efficient introduction of electric vehicles (EVs) requires an understanding of how the charging of EVs impacts the electricity system and if, and to what extent, smart charging strategies, including vehicle to grid (V2G) services, can support the electric grid in the future energy systems. The aim of this thesis is to characterize the flexibility of smart charging including V2G by analyzing the real‑world driving, parking, and charging patterns obtained from logged EVs. The analysis is based on data collected from 394 privately owned EVs and survey responses from their owners. The results reveal substantial flexibility potential for smart charging from several perspectives. However, the findings also highlight important limitations that must be carefully considered when estimating flexibility or implementing flexible charging into energy system models.  Using the lower state of charge (SOC) threshold for charging decisions and the SOC when charging ends, the flexible battery capacity range is calculated to be 59% on average. The aggregated SOC for all logged EVs is within 60%–80% throughout the entire logging period. The results show that charging is needed in fewer than half of the days in a week for more than 73% of weeks, regardless of the attributes of the EV owners, including commuter category and battery capacity. Furthermore, EVs are charged more frequently than the minimum number of charging events required per week. Thus, there is potential for charging in a way that is flexible in time depending on, for example, grid congestion or spot prices. This is particularly the case for non‑commuters with large‑battery EVs. The amount of time that EVs are plugged in for smart charging differs by more than a factor of two if one assumes that EVs are plugged in whenever they are parked at home and that EVs are plugged in only when they charge during the parking event. Thus, careful consideration of which of these assumptions is appropriate is essential when estimating the availability of EVs for smart charging, as the choice can significantly affect the outcomes. Installation of chargers at workplaces can increase the number of grid-connected EVs at places other than the home location, although very few EVs need to charge at workplace to fulfill their driving demand. Incentives to promote plug‑in behavior at the home location can, therefore, be prove to be cost-effective at increasing the number of grid-connected EVs. This flexibility at home is exploited by EV owners with hourly electricity contracts through selecting charging times when the spot price is lower than the daily average. In this thesis, the logged EVs are clustered into three, five, and eleven clusters, resulting in groups with distinct characteristics. When the number of clusters is increased from three to five, a cluster with a low probability of parking at home during the night and a cluster maintaining a high SOC are added to the three clusters. When the number of clusters is extended to eleven, some clusters exhibit combinations of characteristics that are not present in the case with five clusters, including clusters with extreme values. Clusters with characteristics that diverge from typical commuter or non-commuter patterns are obtained. Licentiatavhandling 2026-05-14 18:27:45 https://research.chalmers.se/publication/552119 2026-05-14T18:27:45Z The processes and factors that underpin the development of effective heave pressure (EHP) at the base of deep excavations in soft soils have been numerically quantified for two idealised soil-structure systems, i.e. a building unit cell and a tunnel geometry, using the Finite Element Method. Charts were developed exploiting dimensional analysis to estimate the impact of normalised time between the end of excavation and the completion of the restraining structure at the base, on the emerging magnitude of EHP for several scenarios where the excavation geometry, ground profile, relative stiffness and retaining wall length were varied. The results of the analyses were in good agreement with available data from physical model tests and the monitoring data of a deep excavation in Gothenburg, Sweden. Complementary analyses of site-specific background settlements and water table levels demonstrate that the charts developed are conservative. The results of this study can, within the limitations of the scenarios studied, readily be used for estimations of EHP in preliminary design stage, and as a complement to the detailed, project-specific analyses. Artikel i vetenskaplig tidskrift 2026-05-14 11:53:32 https://research.chalmers.se/publication/542646 2026-05-14T11:53:32Z Boundary Value Analysis (BVA) is a software testing technique that targets inputs at the transitions between different program behaviors, where faults are most likely to occur. Automating this process, known as Boundary Value Exploration (BVE), requires not only finding input pairs that lie on opposite sides of these transitions, but finding enough of them, across different behavioral regions, to give testers a complete picture of where a program's behavior changes and why. This thesis advances BVE toward a search that is broader, more general across input types, and more interpretable to the testers.To achieve this, this thesis focuses on unit-level functions and introduces a search-based framework that systematically seeks boundary candidates that are not only sharp but also spread across a wide range of behavioral regions, so that the search covers many different kinds of transitions rather than converging on the most extreme ones. Building on this foundation, a deeper limitation is addressed: existing approaches require search operators to be hand-crafted for each input type, confining BVE mostly to numeric domains. An agentic, LLM-driven framework that autonomously generates its own exploration strategies removes this bottleneck, extending BVE to functions with non-numeric inputs. Finally, since discovering boundaries is only part of the challenge, a mixed methods study with software professionals investigates whether LLMs can generate natural-language explanations for discovered boundaries, and what properties such explanations need to be used in practice.The results show that a broader and more varied set of boundary behaviors can be discovered through diversity-aware search, and that adaptive, LLM-driven exploration generalizes successfully to input types that other existing automated black-box BVE approaches cannot handle. Software professionals found LLM-generated explanations useful overall, though the study also reveals that trust in such explanations is fragile and that correctness and consistency are prerequisites for adoption rather than desirable extras.Together, these contributions suggest that the path toward practical BVE lies not in optimizing a single dimension of the search, but in balancing quality with diversity, and automation with human understanding. The findings point to diversity as an explicit and configurable testing goal, to LLMs as a practical mechanism for scaling BVE across input domains, and to explanation as a prerequisite for adoption rather than an optional enhancement. Licentiatavhandling 2026-05-13 18:31:40 https://research.chalmers.se/publication/552118 2026-05-13T18:31:40Z Universities are increasingly adopting strategies to realise internationalisation at home. These strategies include ensuring that all students develop international awareness and intercultural competence, regardless of their ability to participate in physical mobility. This shift reflects a recognition that intercultural competence is essential for working life and that underutilised international opportunities exist within domestic campuses, for instance through collaborative projects that bring together local and international students. Engineering education has some of the highest proportions of international students, yet research suggests that engineering educators are comparatively reluctant to engage with internationalisation at home. This thesis, containing four articles, investigates the ways in which intercultural pair and group work can contribute to fulfilling the internationalisation goals in engineering education. Intercultural pair and group work involve students from different cultural backgrounds working together on set tasks where they can find collaboration challenging on a personal and professional level. The thesis takes a qualitative, multilevel approach, and explores both student and teacher perspectives. Using an integration model which highlights both human and structural factors, the research spans individual experiences, classroom practices, and wider institutional structures. This research offers three key contributions to internationalisation at home and engineering education in a European context. Firstly, I suggest a multilevel approach to operationalising intercultural group work, taking into account all levels from classroom to national goals. Secondly, I highlight the teacher’s important dual role as both structuring the context (and being influenced by it) and facilitating human relationships. Thirdly, I propose a move beyond a binary approach of “home” and “international” students towards a more integrated, reflective, and context‑sensitive understanding of identity, experience and belonging. Overall, I argue that intercultural group work can meaningfully support internationalisation at home when embedded within relevant pedagogical and institutional frameworks. Doktorsavhandling 2026-05-13 16:19:34 https://research.chalmers.se/publication/551869 2026-05-13T16:19:34Z Scientific discovery is an active process of designing, testing, and improving theories about the natural world. Automating this process is a grand challenge for 21st century science. This thesis examines scientific inquiry as it relates to machine learning, offering contributions to knowledge representations and reasoning frameworks, demonstrated in systems biology.Systems biology is an integrationist approach to biological science, meaning organisms are treated as complex systems whose behaviour is dictated by the interaction of their constituent parts. Eukaryotic organisms are extremely complex, and research progress in systems biology can be slow. Recent advances in robotics and artificial intelligence (AI) offer great opportunity for automating scientific discovery in this field. Using the model organism Saccharomyces cerevisiae (baker’s yeast), this thesis explores: the philosophical motivations for automation in biological research; knowledge models and hypotheses in systems biology; and computational models of metabolism.The first main contribution is a first-order logic framework for modelling cellular physiology, which enables abduction of hypotheses for improvement of knowledge models, using the automated theorem prover (ATP) iProver. The second contribution is an ontology for describing theory changes and hypotheses in a semantic and storage-efficient manner. The third main contribution is an application of graph neural networks (GNNs) to learn knowledge graph embeddings grounded in empirical data and ontology structures. The final contribution is an end-to-end demonstration of autonomous hypothesis generation and experimentation, with hypotheses modelled using ontology terms to support large language model (LLM) agents and human scientists.These contributions demonstrate the power of knowledge graphs for autonomous scientific discovery. This thesis also argues that scientific discovery is better modelled as supervised learning—specifically active learning for AI scientists—than reinforcement learning; mapping concepts from machine learning algorithms to the domain produces systems that align with established scientific values, leading to improved theories. Doktorsavhandling 2026-05-13 16:18:36 https://research.chalmers.se/publication/552088 2026-05-13T16:18:36Z Statistical mechanics provides a broad theoretical framework for modeling molecular and biological systems, but in practice, the underlying dynamical equations are often not analytically tractable. Consequently, molecular simulations have been the workhorse of statistical mechanics for the past seventy years. However, with the advent of machine learning and generative models, these data-driven methods are starting to complement simulations by providing fast surrogates in scenarios where direct simulation is prohibitively expensive.In this thesis, we discuss how generative models can enhance traditional simulation methods in both equilibrium and nonequilibrium settings. In equilibrium sampling with continuous normalizing flow-based Boltzmann generators, likelihood evaluations scale unfavorably with system size. We show how this issue can be alleviated, demonstrating speedups of up to 100 times on Lennard-Jones systems.Nonequilibrium settings encompass a wider range of systems. We briefly discuss some of the generative modeling methods appropriate in this setting and present an extension of implicit transfer operator models to nonautonomous domains. By combining flow map matching with a physically grounded short-time inductive bias, we accurately model both long- and short-time behavior of nonautonomous systems.This work concludes with a discussion of the broader role of generative machine learning methods in computational statistical mechanics, pointing out new applications and possible future research directions. Licentiatavhandling 2026-05-13 16:05:07 https://research.chalmers.se/publication/552115 2026-05-13T16:05:07Z Background: Regular, detailed reporting on population health by underlying cause of death is fundamental for public health decision making. Cause-specific estimates of mortality and the subsequent effects on life expectancy worldwide are valuable metrics to gauge progress in reducing mortality rates. These estimates are particularly important following large-scale mortality spikes, such as the COVID-19 pandemic. When systematically analysed, mortality rates and life expectancy allow comparisons of the consequences of causes of death globally and over time, providing a nuanced understanding of the effect of these causes on global populations. Methods: The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2021 cause-of-death analysis estimated mortality and years of life lost (YLLs) from 288 causes of death by age-sex-location-year in 204 countries and territories and 811 subnational locations for each year from 1990 until 2021. The analysis used 56 604 data sources, including data from vital registration and verbal autopsy as well as surveys, censuses, surveillance systems, and cancer registries, among others. As with previous GBD rounds, cause-specific death rates for most causes were estimated using the Cause of Death Ensemble model—a modelling tool developed for GBD to assess the out-of-sample predictive validity of different statistical models and covariate permutations and combine those results to produce cause-specific mortality estimates—with alternative strategies adapted to model causes with insufficient data, substantial changes in reporting over the study period, or unusual epidemiology. YLLs were computed as the product of the number of deaths for each cause-age-sex-location-year and the standard life expectancy at each age. As part of the modelling process, uncertainty intervals (UIs) were generated using the 2·5th and 97·5th percentiles from a 1000-draw distribution for each metric. We decomposed life expectancy by cause of death, location, and year to show cause-specific effects on life expectancy from 1990 to 2021. We also used the coefficient of variation and the fraction of population affected by 90% of deaths to highlight concentrations of mortality. Findings are reported in counts and age-standardised rates. Methodological improvements for cause-of-death estimates in GBD 2021 include the expansion of under-5-years age group to include four new age groups, enhanced methods to account for stochastic variation of sparse data, and the inclusion of COVID-19 and other pandemic-related mortality—which includes excess mortality associated with the pandemic, excluding COVID-19, lower respiratory infections, measles, malaria, and pertussis. For this analysis, 199 new country-years of vital registration cause-of-death data, 5 country-years of surveillance data, 21 country-years of verbal autopsy data, and 94 country-years of other data types were added to those used in previous GBD rounds. Findings: The leading causes of age-standardised deaths globally were the same in 2019 as they were in 1990; in descending order, these were, ischaemic heart disease, stroke, chronic obstructive pulmonary disease, and lower respiratory infections. In 2021, however, COVID-19 replaced stroke as the second-leading age-standardised cause of death, with 94·0 deaths (95% UI 89·2–100·0) per 100 000 population. The COVID-19 pandemic shifted the rankings of the leading five causes, lowering stroke to the third-leading and chronic obstructive pulmonary disease to the fourth-leading position. In 2021, the highest age-standardised death rates from COVID-19 occurred in sub-Saharan Africa (271·0 deaths [250·1–290·7] per 100 000 population) and Latin America and the Caribbean (195·4 deaths [182·1–211·4] per 100 000 population). The lowest age-standardised death rates from COVID-19 were in the high-income super-region (48·1 deaths [47·4–48·8] per 100 000 population) and southeast Asia, east Asia, and Oceania (23·2 deaths [16·3–37·2] per 100 000 population). Globally, life expectancy steadily improved between 1990 and 2019 for 18 of the 22 investigated causes. Decomposition of global and regional life expectancy showed the positive effect that reductions in deaths from enteric infections, lower respiratory infections, stroke, and neonatal deaths, among others have contributed to improved survival over the study period. However, a net reduction of 1·6 years occurred in global life expectancy between 2019 and 2021, primarily due to increased death rates from COVID-19 and other pandemic-related mortality. Life expectancy was highly variable between super-regions over the study period, with southeast Asia, east Asia, and Oceania gaining 8·3 years (6·7–9·9) overall, while having the smallest reduction in life expectancy due to COVID-19 (0·4 years). The largest reduction in life expectancy due to COVID-19 occurred in Latin America and the Caribbean (3·6 years). Additionally, 53 of the 288 causes of death were highly concentrated in locations with less than 50% of the global population as of 2021, and these causes of death became progressively more concentrated since 1990, when only 44 causes showed this pattern. The concentration phenomenon is discussed heuristically with respect to enteric and lower respiratory infections, malaria, HIV/AIDS, neonatal disorders, tuberculosis, and measles. Interpretation: Long-standing gains in life expectancy and reductions in many of the leading causes of death have been disrupted by the COVID-19 pandemic, the adverse effects of which were spread unevenly among populations. Despite the pandemic, there has been continued progress in combatting several notable causes of death, leading to improved global life expectancy over the study period. Each of the seven GBD super-regions showed an overall improvement from 1990 and 2021, obscuring the negative effect in the years of the pandemic. Additionally, our findings regarding regional variation in causes of death driving increases in life expectancy hold clear policy utility. Analyses of shifting mortality trends reveal that several causes, once widespread globally, are now increasingly concentrated geographically. These changes in mortality concentration, alongside further investigation of changing risks, interventions, and relevant policy, present an important opportunity to deepen our understanding of mortality-reduction strategies. Examining patterns in mortality concentration might reveal areas where successful public health interventions have been implemented. Translating these successes to locations where certain causes of death remain entrenched can inform policies that work to improve life expectancy for people everywhere. Funding: Bill & Melinda Gates Foundation. Artikel i vetenskaplig tidskrift 2026-05-13 15:49:02 https://research.chalmers.se/publication/541901 2026-05-13T15:49:02Z Background: Accurate assessments of current and future fertility—including overall trends and changing population age structures across countries and regions—are essential to help plan for the profound social, economic, environmental, and geopolitical challenges that these changes will bring. Estimates and projections of fertility are necessary to inform policies involving resource and health-care needs, labour supply, education, gender equality, and family planning and support. The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2021 produced up-to-date and comprehensive demographic assessments of key fertility indicators at global, regional, and national levels from 1950 to 2021 and forecast fertility metrics to 2100 based on a reference scenario and key policy-dependent alternative scenarios. Methods: To estimate fertility indicators from 1950 to 2021, mixed-effects regression models and spatiotemporal Gaussian process regression were used to synthesise data from 8709 country-years of vital and sample registrations, 1455 surveys and censuses, and 150 other sources, and to generate age-specific fertility rates (ASFRs) for 5-year age groups from age 10 years to 54 years. ASFRs were summed across age groups to produce estimates of total fertility rate (TFR). Livebirths were calculated by multiplying ASFR and age-specific female population, then summing across ages 10–54 years. To forecast future fertility up to 2100, our Institute for Health Metrics and Evaluation (IHME) forecasting model was based on projections of completed cohort fertility at age 50 years (CCF50; the average number of children born over time to females from a specified birth cohort), which yields more stable and accurate measures of fertility than directly modelling TFR. CCF50 was modelled using an ensemble approach in which three sub-models (with two, three, and four covariates variously consisting of female educational attainment, contraceptive met need, population density in habitable areas, and under-5 mortality) were given equal weights, and analyses were conducted utilising the MR-BRT (meta-regression—Bayesian, regularised, trimmed) tool. To capture time-series trends in CCF50 not explained by these covariates, we used a first-order autoregressive model on the residual term. CCF50 as a proportion of each 5-year ASFR was predicted using a linear mixed-effects model with fixed-effects covariates (female educational attainment and contraceptive met need) and random intercepts for geographical regions. Projected TFRs were then computed for each calendar year as the sum of single-year ASFRs across age groups. The reference forecast is our estimate of the most likely fertility future given the model, past fertility, forecasts of covariates, and historical relationships between covariates and fertility. We additionally produced forecasts for multiple alternative scenarios in each location: the UN Sustainable Development Goal (SDG) for education is achieved by 2030; the contraceptive met need SDG is achieved by 2030; pro-natal policies are enacted to create supportive environments for those who give birth; and the previous three scenarios combined. Uncertainty from past data inputs and model estimation was propagated throughout analyses by taking 1000 draws for past and present fertility estimates and 500 draws for future forecasts from the estimated distribution for each metric, with 95% uncertainty intervals (UIs) given as the 2·5 and 97·5 percentiles of the draws. To evaluate the forecasting performance of our model and others, we computed skill values—a metric assessing gain in forecasting accuracy—by comparing predicted versus observed ASFRs from the past 15 years (2007–21). A positive skill metric indicates that the model being evaluated performs better than the baseline model (here, a simplified model holding 2007 values constant in the future), and a negative metric indicates that the evaluated model performs worse than baseline. Findings: During the period from 1950 to 2021, global TFR more than halved, from 4·84 (95% UI 4·63–5·06) to 2·23 (2·09–2·38). Global annual livebirths peaked in 2016 at 142 million (95% UI 137–147), declining to 129 million (121–138) in 2021. Fertility rates declined in all countries and territories since 1950, with TFR remaining above 2·1—canonically considered replacement-level fertility—in 94 (46·1%) countries and territories in 2021. This included 44 of 46 countries in sub-Saharan Africa, which was the super-region with the largest share of livebirths in 2021 (29·2% [28·7–29·6]). 47 countries and territories in which lowest estimated fertility between 1950 and 2021 was below replacement experienced one or more subsequent years with higher fertility; only three of these locations rebounded above replacement levels. Future fertility rates were projected to continue to decline worldwide, reaching a global TFR of 1·83 (1·59–2·08) in 2050 and 1·59 (1·25–1·96) in 2100 under the reference scenario. The number of countries and territories with fertility rates remaining above replacement was forecast to be 49 (24·0%) in 2050 and only six (2·9%) in 2100, with three of these six countries included in the 2021 World Bank-defined low-income group, all located in the GBD super-region of sub-Saharan Africa. The proportion of livebirths occurring in sub-Saharan Africa was forecast to increase to more than half of the world's livebirths in 2100, to 41·3% (39·6–43·1) in 2050 and 54·3% (47·1–59·5) in 2100. The share of livebirths was projected to decline between 2021 and 2100 in most of the six other super-regions—decreasing, for example, in south Asia from 24·8% (23·7–25·8) in 2021 to 16·7% (14·3–19·1) in 2050 and 7·1% (4·4–10·1) in 2100—but was forecast to increase modestly in the north Africa and Middle East and high-income super-regions. Forecast estimates for the alternative combined scenario suggest that meeting SDG targets for education and contraceptive met need, as well as implementing pro-natal policies, would result in global TFRs of 1·65 (1·40–1·92) in 2050 and 1·62 (1·35–1·95) in 2100. The forecasting skill metric values for the IHME model were positive across all age groups, indicating that the model is better than the constant prediction. Interpretation: Fertility is declining globally, with rates in more than half of all countries and territories in 2021 below replacement level. Trends since 2000 show considerable heterogeneity in the steepness of declines, and only a small number of countries experienced even a slight fertility rebound after their lowest observed rate, with none reaching replacement level. Additionally, the distribution of livebirths across the globe is shifting, with a greater proportion occurring in the lowest-income countries. Future fertility rates will continue to decline worldwide and will remain low even under successful implementation of pro-natal policies. These changes will have far-reaching economic and societal consequences due to ageing populations and declining workforces in higher-income countries, combined with an increasing share of livebirths among the already poorest regions of the world. Funding: Bill & Melinda Gates Foundation. Artikel i vetenskaplig tidskrift 2026-05-13 15:48:18 https://research.chalmers.se/publication/541958 2026-05-13T15:48:18Z With an ageing population, healthcare systems are facing increasing numbers of patients and shrinking resources. Hospital at Home – a service delivery model moving hospital-level care to patients’ homes – is suggested to help address these issues by increasing hospital capacity while maintaining or increasing quality of care. However, the model is a complex innovation which is notoriously challenging to implement and scale in public healthcare. It has been suggested that different logics of change – mechanical, ecological, and social – can be combined to address these challenges. Mechanical logic denotes a focus on standardisation, linearity, and replicability, ecological logic emphasises emergence, interdependence, and adaptiveness, and social logic focuses on identifying and explaining social mechanisms such as people’s behaviour and interpretations.This thesis uses an action research approach to longitudinally study the implementation and scaling of Hospital at Home in a large hospital in Sweden, including interview data from 59 clinicians, patients, and managers, and a systematic literature review. It confirms that the aforementioned logics can be used in combination in order to support innovation processes. Multiple ways in which this can be achieved are described, for example by maintaining tensions conducive to continued innovation or by avoiding Catch-22 situations. The thesis confirms and extends views in extant innovation literature as well as suggests a convergence point between service ecosystem literature and learning-based approaches to innovation in organisations, enabling integration and cross-fertilisation of insights. Additionally, it provides a developed version of an analytical framework for practitioners who seek to support ongoing innovation processes. Licentiatavhandling 2026-05-13 15:46:33 https://research.chalmers.se/publication/552116 2026-05-13T15:46:33Z Background: Detailed, comprehensive, and timely reporting on population health by underlying causes of disability and premature death is crucial to understanding and responding to complex patterns of disease and injury burden over time and across age groups, sexes, and locations. The availability of disease burden estimates can promote evidence-based interventions that enable public health researchers, policy makers, and other professionals to implement strategies that can mitigate diseases. It can also facilitate more rigorous monitoring of progress towards national and international health targets, such as the Sustainable Development Goals. For three decades, the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) has filled that need. A global network of collaborators contributed to the production of GBD 2021 by providing, reviewing, and analysing all available data. GBD estimates are updated routinely with additional data and refined analytical methods. GBD 2021 presents, for the first time, estimates of health loss due to the COVID-19 pandemic. Methods: The GBD 2021 disease and injury burden analysis estimated years lived with disability (YLDs), years of life lost (YLLs), disability-adjusted life-years (DALYs), and healthy life expectancy (HALE) for 371 diseases and injuries using 100 983 data sources. Data were extracted from vital registration systems, verbal autopsies, censuses, household surveys, disease-specific registries, health service contact data, and other sources. YLDs were calculated by multiplying cause-age-sex-location-year-specific prevalence of sequelae by their respective disability weights, for each disease and injury. YLLs were calculated by multiplying cause-age-sex-location-year-specific deaths by the standard life expectancy at the age that death occurred. DALYs were calculated by summing YLDs and YLLs. HALE estimates were produced using YLDs per capita and age-specific mortality rates by location, age, sex, year, and cause. 95% uncertainty intervals (UIs) were generated for all final estimates as the 2·5th and 97·5th percentiles values of 500 draws. Uncertainty was propagated at each step of the estimation process. Counts and age-standardised rates were calculated globally, for seven super-regions, 21 regions, 204 countries and territories (including 21 countries with subnational locations), and 811 subnational locations, from 1990 to 2021. Here we report data for 2010 to 2021 to highlight trends in disease burden over the past decade and through the first 2 years of the COVID-19 pandemic. Findings: Global DALYs increased from 2·63 billion (95% UI 2·44–2·85) in 2010 to 2·88 billion (2·64–3·15) in 2021 for all causes combined. Much of this increase in the number of DALYs was due to population growth and ageing, as indicated by a decrease in global age-standardised all-cause DALY rates of 14·2% (95% UI 10·7–17·3) between 2010 and 2019. Notably, however, this decrease in rates reversed during the first 2 years of the COVID-19 pandemic, with increases in global age-standardised all-cause DALY rates since 2019 of 4·1% (1·8–6·3) in 2020 and 7·2% (4·7–10·0) in 2021. In 2021, COVID-19 was the leading cause of DALYs globally (212·0 million [198·0–234·5] DALYs), followed by ischaemic heart disease (188·3 million [176·7–198·3]), neonatal disorders (186·3 million [162·3–214·9]), and stroke (160·4 million [148·0–171·7]). However, notable health gains were seen among other leading communicable, maternal, neonatal, and nutritional (CMNN) diseases. Globally between 2010 and 2021, the age-standardised DALY rates for HIV/AIDS decreased by 47·8% (43·3–51·7) and for diarrhoeal diseases decreased by 47·0% (39·9–52·9). Non-communicable diseases contributed 1·73 billion (95% UI 1·54–1·94) DALYs in 2021, with a decrease in age-standardised DALY rates since 2010 of 6·4% (95% UI 3·5–9·5). Between 2010 and 2021, among the 25 leading Level 3 causes, age-standardised DALY rates increased most substantially for anxiety disorders (16·7% [14·0–19·8]), depressive disorders (16·4% [11·9–21·3]), and diabetes (14·0% [10·0–17·4]). Age-standardised DALY rates due to injuries decreased globally by 24·0% (20·7–27·2) between 2010 and 2021, although improvements were not uniform across locations, ages, and sexes. Globally, HALE at birth improved slightly, from 61·3 years (58·6–63·6) in 2010 to 62·2 years (59·4–64·7) in 2021. However, despite this overall increase, HALE decreased by 2·2% (1·6–2·9) between 2019 and 2021. Interpretation: Putting the COVID-19 pandemic in the context of a mutually exclusive and collectively exhaustive list of causes of health loss is crucial to understanding its impact and ensuring that health funding and policy address needs at both local and global levels through cost-effective and evidence-based interventions. A global epidemiological transition remains underway. Our findings suggest that prioritising non-communicable disease prevention and treatment policies, as well as strengthening health systems, continues to be crucially important. The progress on reducing the burden of CMNN diseases must not stall; although global trends are improving, the burden of CMNN diseases remains unacceptably high. Evidence-based interventions will help save the lives of young children and mothers and improve the overall health and economic conditions of societies across the world. Governments and multilateral organisations should prioritise pandemic preparedness planning alongside efforts to reduce the burden of diseases and injuries that will strain resources in the coming decades. Funding: Bill & Melinda Gates Foundation. Artikel i vetenskaplig tidskrift 2026-05-13 15:42:11 https://research.chalmers.se/publication/542122 2026-05-13T15:42:11Z Background: Understanding the health consequences associated with exposure to risk factors is necessary to inform public health policy and practice. To systematically quantify the contributions of risk factor exposures to specific health outcomes, the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2021 aims to provide comprehensive estimates of exposure levels, relative health risks, and attributable burden of disease for 88 risk factors in 204 countries and territories and 811 subnational locations, from 1990 to 2021. Methods: The GBD 2021 risk factor analysis used data from 54 561 total distinct sources to produce epidemiological estimates for 88 risk factors and their associated health outcomes for a total of 631 risk–outcome pairs. Pairs were included on the basis of data-driven determination of a risk–outcome association. Age-sex-location-year-specific estimates were generated at global, regional, and national levels. Our approach followed the comparative risk assessment framework predicated on a causal web of hierarchically organised, potentially combinative, modifiable risks. Relative risks (RRs) of a given outcome occurring as a function of risk factor exposure were estimated separately for each risk–outcome pair, and summary exposure values (SEVs), representing risk-weighted exposure prevalence, and theoretical minimum risk exposure levels (TMRELs) were estimated for each risk factor. These estimates were used to calculate the population attributable fraction (PAF; ie, the proportional change in health risk that would occur if exposure to a risk factor were reduced to the TMREL). The product of PAFs and disease burden associated with a given outcome, measured in disability-adjusted life-years (DALYs), yielded measures of attributable burden (ie, the proportion of total disease burden attributable to a particular risk factor or combination of risk factors). Adjustments for mediation were applied to account for relationships involving risk factors that act indirectly on outcomes via intermediate risks. Attributable burden estimates were stratified by Socio-demographic Index (SDI) quintile and presented as counts, age-standardised rates, and rankings. To complement estimates of RR and attributable burden, newly developed burden of proof risk function (BPRF) methods were applied to yield supplementary, conservative interpretations of risk–outcome associations based on the consistency of underlying evidence, accounting for unexplained heterogeneity between input data from different studies. Estimates reported represent the mean value across 500 draws from the estimate's distribution, with 95% uncertainty intervals (UIs) calculated as the 2·5th and 97·5th percentile values across the draws. Findings: Among the specific risk factors analysed for this study, particulate matter air pollution was the leading contributor to the global disease burden in 2021, contributing 8·0% (95% UI 6·7–9·4) of total DALYs, followed by high systolic blood pressure (SBP; 7·8% [6·4–9·2]), smoking (5·7% [4·7–6·8]), low birthweight and short gestation (5·6% [4·8–6·3]), and high fasting plasma glucose (FPG; 5·4% [4·8–6·0]). For younger demographics (ie, those aged 0–4 years and 5–14 years), risks such as low birthweight and short gestation and unsafe water, sanitation, and handwashing (WaSH) were among the leading risk factors, while for older age groups, metabolic risks such as high SBP, high body-mass index (BMI), high FPG, and high LDL cholesterol had a greater impact. From 2000 to 2021, there was an observable shift in global health challenges, marked by a decline in the number of all-age DALYs broadly attributable to behavioural risks (decrease of 20·7% [13·9–27·7]) and environmental and occupational risks (decrease of 22·0% [15·5–28·8]), coupled with a 49·4% (42·3–56·9) increase in DALYs attributable to metabolic risks, all reflecting ageing populations and changing lifestyles on a global scale. Age-standardised global DALY rates attributable to high BMI and high FPG rose considerably (15·7% [9·9–21·7] for high BMI and 7·9% [3·3–12·9] for high FPG) over this period, with exposure to these risks increasing annually at rates of 1·8% (1·6–1·9) for high BMI and 1·3% (1·1–1·5) for high FPG. By contrast, the global risk-attributable burden and exposure to many other risk factors declined, notably for risks such as child growth failure and unsafe water source, with age-standardised attributable DALYs decreasing by 71·5% (64·4–78·8) for child growth failure and 66·3% (60·2–72·0) for unsafe water source. We separated risk factors into three groups according to trajectory over time: those with a decreasing attributable burden, due largely to declining risk exposure (eg, diet high in trans-fat and household air pollution) but also to proportionally smaller child and youth populations (eg, child and maternal malnutrition); those for which the burden increased moderately in spite of declining risk exposure, due largely to population ageing (eg, smoking); and those for which the burden increased considerably due to both increasing risk exposure and population ageing (eg, ambient particulate matter air pollution, high BMI, high FPG, and high SBP). Interpretation: Substantial progress has been made in reducing the global disease burden attributable to a range of risk factors, particularly those related to maternal and child health, WaSH, and household air pollution. Maintaining efforts to minimise the impact of these risk factors, especially in low SDI locations, is necessary to sustain progress. Successes in moderating the smoking-related burden by reducing risk exposure highlight the need to advance policies that reduce exposure to other leading risk factors such as ambient particulate matter air pollution and high SBP. Troubling increases in high FPG, high BMI, and other risk factors related to obesity and metabolic syndrome indicate an urgent need to identify and implement interventions. Funding: Bill & Melinda Gates Foundation. Artikel i vetenskaplig tidskrift 2026-05-13 15:39:34 https://research.chalmers.se/publication/542312 2026-05-13T15:39:34Z Open Science practices are shaping both science and policymaking. This thesis examines the visions of Open Science and their enactments through four empirical cases. It provides an understanding of what Open Science means in terms of infrastructures: in cases where Open Science practices exhibit infrastructuring efforts, where they reconfigure existing infrastructures, or where such infrastructuring efforts are not sustained. In this thesis, the overarching visions of Open Science are identified as participation, public benefit, and transparency, and these visions are interpreted in relation to the erosion of public trust in science, the commercialisation of science, and the replication crisis. Alongside the emergence of these crises, new digital sharing technologies have led to the positioning of Open Science practices, such as open-source software, Citizen Science, and Open Data, as imagined and prescribed solutions to these crises.Building on this framework, this thesis has four main objectives, each pursued through a case study with its own set of methods: first, it examines infrastructuring efforts in a cluster of published articles to investigate how transparency is prescribed as a solution to the replication crisis through a mixed-methods approach. Second, it examines the limitations of participatory knowledge-making initiatives situated outside science by comparing four citizen observatories through interviews. Third, through retrospective participant observation and document analysis, it addresses policymakers’ visions of participation in Tehran and their efforts to improve digital participatory tools. Lastly, it investigates what a public good Open Science might entail by studying the use of open-source tools in a water infrastructure through ethnographic visits and interviews.This thesis concludes that Open Science practitioners and advocates, aim to enact reform in science and policy through the implementation of Open Science practices. However, relying on technical approaches will not address the changes they aim to achieve. Furthermore, infrastructures play a critical role in enabling or constraining Open Science visions. Doktorsavhandling 2026-05-13 14:16:20 https://research.chalmers.se/publication/552096 2026-05-13T14:16:20Z Design decisions in healthcare and educational facilities have long-term consequences for everyday work. During design review, end-users are typically expected to evaluate architectural proposals using conventional formats such as 2D drawings, which require mental reconstruction of three-dimensional space and may constrain the use of their operational knowledge. Immersive virtual reality has therefore gained attention as an alternative format enabling full-scale exploration of proposed designs. However, studies in real-world projects remain limited, and it is unclear how immersive formats influence cognitive load, identification of design issues, and end-users' ability to communicate their operational knowledge. This matters because facilities that fail to reflect end-user work practice can compromise operational efficiency, safety, and well-being for years after construction. To address this need, this thesis investigates how immersive virtual environments can support end-user involvement, including Co-Design, during building design review. The research draws on four empirical studies in ongoing healthcare and educational building projects involving end-user groups such as building occupants, facility planners, and client representatives. It combines qualitative analyses of design review sessions with quantitative measures of cognitive load and issue identification, and draws on Cognitive Fit Theory, Collaborative Virtual Environment research, and participatory design theory.The findings show that immersive virtual environments enable end-users to explore proposed designs at full scale and assess spatial layouts in relation to operational work practices. Immersive review supported identification of workflow- and layout-related design issues across design phases and was most effective when combined with conventional 2D drawings and overview-based formats. Interactive features such as multi-user interaction, object interaction, and multi-scale views further supported communication of operational knowledge within representations of the facilities in focus.The thesis makes three contributions. First, it provides empirical evidence that immersive virtual environments can support higher levels of end-user involvement, including Co-Design, when interactive features such as multi-user collaboration and object interaction are available. Second, it extends Cognitive Fit Theory into collaborative design review, arguing that individual cognitive fit is a necessary precondition for higher collaborative involvement. Third, it offers practical implications for combining immersive and conventional formats to support end-user involvement during design review, helping ensure that facilities support the daily work of building occupants. Doktorsavhandling 2026-05-13 14:02:16 https://research.chalmers.se/publication/552075 2026-05-13T14:02:16Z Electric vehicle charging faces challenges of high infrastructure costs and low utilization. Shared charging among fleet operators offers a sustainable alternative. This study formulates a collaborative scheduling problem in which two companies coordinate charging to minimize their individual costs while achieving efficient and equitable outcomes. A bi-objective optimization framework is developed, proposing the Balanced Bounding Box Method (B3M) to generate a representative subset of globally optimal solutions with substantially reduced computational effort. Cooperative bargaining is then applied to derive an actionable final decision from the efficient frontier. Numerical results show that this framework maintains frontier integrity while cutting computation time. Beyond improving decision efficiency, the study offers insights into how transparent and equitable solution selection can sustain long-term collaboration among operators. The framework provides practical guidance to improve charger utilization and reduce system costs, supporting more sustainable use of existing infrastructure. Artikel i vetenskaplig tidskrift 2026-05-13 13:56:46 https://research.chalmers.se/publication/229b94c8-5685-466b-a0be-6fe0e1cd9c86 2026-05-13T13:56:46Z Artificial intelligence (AI) in chemical engineering has moved from promise to practice: physics-aware (gray-box) models are gaining traction, reinforcement learning complements model predictive control (MPC), and generative AI powers documentation, digitization, and safety workflows. Near-term value arises where AI augments, rather than replaces, process system engineering (PSE) practice (e.g., through soft sensing and surrogate models), while autonomous operations, fully automated hazard and operability (HAZOP) analysis, and large-scale mechanistic discovery remain largely at the research stage. The decisive bottleneck is reliable deployment: AI models must be treated like any other engineered system, with validation, monitoring, and governance aligned with emerging frameworks such as the EU AI Act and NIST risk management framework (RMF). With incubator labs, open benchmarks, and retooled education pipelines, AI can become a safe, reliable, and sustainable co-worker in the process industries within years. Reviewartikel 2026-05-13 13:18:28 https://research.chalmers.se/publication/0ab3278b-8fac-4d06-aed0-259e5f129b8d 2026-05-13T13:18:28Z Industrial multiphase reactors remain among the most challenging systems to model due to their complexity, multiscale coupling, and persistent uncertainties in turbulence, interphase transport, and constitutive closures. While traditional approaches combining first-principles physics, empirical correlations, and numerical pra have enabled substantial progress, fundamental limitations persist. This perspective outlines how advances in artificial intelligence (AI), high-performance computing, and, eventually, quantum computing (QC) can steer multiphase modeling toward industry-ready predictive capability with an accuracy unthinkable today. AI enables more generalizable, physics-constrained closures, while graphics processing units (GPUs) and exascale platforms already enable industry-scale simulations at unprecedented fidelity. Although QC is a longer-term prospect, hybrid quantum–classical approaches offer pathways to address complexities beyond classical limits. These developments promise to transform modeling workflows and engineering practice, with direct implications for scale-up, reliability, sustainability, and cost reduction. We highlight key research priorities, including multiphase-aware turbulence models, AI-assisted closures, hybrid solvers, computing architectures, and rigorous verification, validation, and uncertainty quantification. Reviewartikel 2026-05-13 13:18:07 https://research.chalmers.se/publication/550229 2026-05-13T13:18:07Z Bubble dynamics in bubbling fluidized bed reactors govern heat and mass transfer rates, mixing uniformity, and overall process efficiency, but remain challenging to predict accurately. This study develops a hybrid modeling framework to model bubble growth along bed height by integrating empirical correlations with data-driven corrections using Physics-Informed Neural Networks (PINNs) and Universal Differential Equations (UDEs). Experimental data used for training, which were obtained for Geldart Group B particles with various particle size distributions (PSDs), are characteristically noisy with non-normal bubble diameter distributions. The optimal black-box data-driven NN fails to capture the known monotonic bubble growth with bed height. In contrast, the PINN, embedding the Hilligardt-Werther empirical correlation, enforces physical monotonicity and robustness despite the noisy data. The UDE further couples this empirical correlation with a learned neural correction, yielding a physically interpretable, data-calibrated model. Results show that the UDE preserves the physical trend while adaptively compensating for empirical correlation discrepancies, with the data-driven neural correction contributing 20-35% of the total trend. The largest data-driven correction occurs for the broadest PSD case, which is expected because such conditions likely lie outside the calibration range of the empirical correlation. Importantly, this also demonstrates the ability of the hybrid framework to extend empirical models beyond their original scope. In the absence of first-principles-based descriptions, the hybrid (gray-box) approach presented here reconciles physical underpinnings with data-driven flexibility, offering a more reliable, interpretable, and generalizable framework for modeling bubble growth in practical fluidized beds. Artikel i vetenskaplig tidskrift 2026-05-13 13:17:47 https://research.chalmers.se/publication/551431 2026-05-13T13:17:47Z This study provides unique insights into the dynamics of bubble deformation and breakup under turbulent flow conditions, utilizing both experimental measurements and high-resolution simulations and unveils information that has been previously unattainable with current methods. The simulations are rigorously validated against experimental data obtained under identical hydrodynamic conditions, and enable analyzes of the interfacial dynamics, breakup time scales, daughter size distributions, and internal flow mechanisms, crucial for advancing future model development. Overall, the dynamic deformation and statistical data show very strong agreement with experimental measurements and reveal an inherent stochastic behavior of bubble breakup due to turbulent interactions. For the first time, details of the internal flow mechanism during bubble breakup have been resolved, revealing development of flow velocities up to 30 times greater at the bubble neck compared to the mean bubble velocity. Analysis reveals that the characteristic internal redistribution flow occurs within a fraction of a millisecond, necessitating a temporal resolution of 20,000 frames per second. The development of an accelerating internal flow is quantified throughout the process until a sudden termination of the flow occurs due to the rapidly shifting balance of stresses at the interface. This ultimately leads to the breakup and formation of unequal sized daughter fragments, approximating a U-shaped distribution, with consistent results in both experimental and simulation data. Evidence suggests that bubble breakup at higher Weber number can form satellite fragments like what is known from droplet breakup, but these are likely beyond the resolution capabilities of the most advanced experimental setups documented in single bubble breakup literature. Consequently, simulations offer a more comprehensive understanding of bubble dynamics, surpassing current experimental capabilities due to their superior temporal and spatial resolutions and the absence of complications from light reflection and refraction at interfaces. The details and quantifications presented in this study are anticipated to contribute significantly to the development of refined breakup kernels. Artikel i vetenskaplig tidskrift 2026-05-13 13:17:05 https://research.chalmers.se/publication/545829 2026-05-13T13:17:05Z The ubiquity of heat exchangers, coupled with the urgent need to augment energy efficiency in the green transition of industries, underscores the importance of optimizing the flow-field design to maximize heat transfer. Inspired by nature-evolved transport networks (e.g., tree branches), this study explores 3D three-level trifurcating pipe networks with varying branching angles (20° - 65°) as alternatives to conventional vertical pipes. Three key conclusions are highlighted. Firstly, steep temperature increases at junctions lead to distinctly different heat transfer and flow behaviors in the middle level among the geometries. Secondly, the relationship between the Re-normalized thermal performance factor (TPF) is non-monotonic with respect to angle, with the 36° model giving the highest TPF/Re value. Thirdly, the superior performance of the 36° model is associated with the lowest mean normalized turbulent viscosity (µt/Re) and highest mean normalized vorticity (ΩD/U), suggesting the flow is dominated by coherent rotational strcutures rather than chaotic, dissipative turbulence. These coherent vortices could be leveraged - by judiciously mimicking the 36° configuration - to further enhance thermal performance. Furthermore, the difference in turbulent viscosity between the outer and central pipes in the top level is the least for the 36° model, indicating enhanced uniformity. These findings offer insights for designing efficient, nature-inspired heat exchangers. Artikel i vetenskaplig tidskrift 2026-05-13 13:16:25 https://research.chalmers.se/publication/548685 2026-05-13T13:16:25Z Digitalization in chemical engineering enhances data accessibility and allows leveraging advancements in machine learning to develop predictive reactor models. Nevertheless, purely data-driven machine learning models lack fundamental physical and chemical principles, thereby limiting their interpretability and generalization capabilities. Adhering to the principles of scientific machine learning, the integration of fundamental constraints can overcome these limitations. In this study, we propose incorporating conservation laws as a soft constraint in the reactor model. These laws, which are not known a priori, are automatically discovered from data and integrated with neural ODEs. By assessing the null space of the dependent variables within the datasets, these laws are identified, thereby enhancing model generalization. The findings show that embedding physics-regularization not only improves generalization, especially when data is scarce, but also enhances robustness. Furthermore, we demonstrate that when sampling fails to capture key dynamics, the model can still accurately predict the evolution of individual species concentrations. This finding suggests that information loss resulting from inadequate sampling can be effectively mitigated by using the proposed method. By introducing noise into the training and validation datasets, we show that the methodology remains robust and consistently outperforms benchmarks across all different levels of noise studied. In data-rich scenarios where generalization is less of a concern, pre-training the model reduces the total computational time by 68 percent. Thus, the implementation of suitable strategies for reactor modeling can significantly improve accuracy, robustness, and computational efficiency. Artikel i vetenskaplig tidskrift 2026-05-13 13:16:08 https://research.chalmers.se/publication/547456 2026-05-13T13:16:08Z Distillation is one of the most widely used separation techniques in chemical processes, this separation method accounts for a significant portion of global energy consumption. Process intensification strategies have led to alternative technologies such as Heat Integrated Distillation Columns (HIDiC), these units can significantly reduce energy consumption. The goal of this study is to develop a CFD model to analyze momentum, heat and mass transfer in the separation of propylene/propane system in a Top HIDiC. The initial and boundary conditions are based on data from Aspen plus process simulations. Ranz-Marshall approaches are used for the representation of the heat transfer between phases and resistance in the gas phase. The mass transfer coefficient was calculated using Higbie’s model implemented via UDF. The Peng-Robinson thermodynamic equilibrium model was implemented to estimate equilibrium ratio at the interface by UDF. The developed CFD model was used to evaluate three pairs of internal rectification stage plus external stripping stage, that are thermally integrated. These stages were selected as representative of the column. Phases distribution, temperature and vapor phase Murphree efficiency were obtained as results, showing the potential of CFD models to analyze new and structurally complex units. Artikel i vetenskaplig tidskrift 2026-05-13 13:15:34 https://research.chalmers.se/publication/547806 2026-05-13T13:15:34Z This study investigates single droplet breakup from a theoretical perspective and addresses whether breakup in turbulent flows can be studied using highly-resolved simulations. Transient and three-dimensional turbulent flow simulations are performed to investigate if the apparent stochastic outcome from the droplet breakup can be predicted. For a given turbulent dissipation rate the breakup events were simulated for various detailed turbulence realizations. For this purpose, a well-characterized system widely used for kernel development is utilized to validate the simulations with respect to the key characteristics of stochastic breakup, including droplet deformation time, the number of fragments, and the specific breakup rate. The statistical validations show very good agreement with all the quantitative properties relevant to the breakup dynamics. Necklace breakup is also observed in line with patterns found in experiments. Evidence is found that the rate of energy transfer is positively correlated with higher order fragmentation. This can allow development of more accurate breakup kernels compared to the ones that only relies on the maximum amount of energy transfer. It is concluded that the simulation method provides new data on the stochastic characteristics of breakup. The method also provides a means to extract more details than experimentally possible since the analysis allows better spatial and temporal resolutions, and 3D analysis of energy transfer which provides better accuracy compared to experimental 2D data. Artikel i vetenskaplig tidskrift 2026-05-13 13:12:22 https://research.chalmers.se/publication/512109 2026-05-13T13:12:22Z This work provides an in-depth understanding of different breakup mechanisms for fluid particles in turbulent flows. All the disruptive and cohesive stresses are considered for the entire turbulent energy spectrum and their contributions to the breakup are evaluated. A new modeling framework is presented that bridges across turbulent subranges. The model entails different mechanisms for breakup by abandoning the classical limitation of inertial models. The predictions are validated with experiments encompassing both breakup regimes for droplets stabilized by internal viscosity and interfacial tension down to the micrometer length scale, which covers both the inertial and dissipation subranges. The model performance ensures the reliability of the framework, which involves different mechanisms. It retains the breakup rate for inertial models, improves the predictions for the transition region from inertia to dissipation, and bridges seamlessly to Kolmogorov-sized droplets. Artikel i vetenskaplig tidskrift 2026-05-13 13:11:15 https://research.chalmers.se/publication/511847 2026-05-13T13:11:15Z Foundation models and Large Language Models (LLMs) have strong general capabilities. They can understand language, reason across different tasks, generate code, and solve problems in areas where they were not specifically trained. This broad capability makes them powerful starting points for real-world AI systems. However, for high-stakes domains, such as automotive software engineering, AI systems must do more than provide plausible answers. They must follow domain rules, respect data structures, handle operational constraints, and produce reasoning that experts can check and trust. This creates a gap between general capability and domain-specific reliability.This thesis argues for Domain-Centric AI: the design of AI systems that are generalizable across domains, adaptable to target domains, and able to reason reliably within specific operational domains. These levels build on one another. Generalization provides the model-level foundation. Adaptation aligns this foundation with a target domain. Domain-specific system design then enforces the rules, workflows, and constraints needed for reliable use.The thesis explores this progression through four papers. The first paper surveys meta-learning methods for domain generalization and shows how models can become more robust to unseen domains. The second paper extends this perspective to vision-language models by introducing latent domain prompt learning for domain generalization. The third and fourth papers focus on industrial LLM-based agent systems for automotive software release analytics. They demonstrate how general LLM capabilities can be embedded in multi-agent and pipeline designs to support informed decision-making. Together, the studies show reliable AI in domain-specific settings can be designed by combining a flexible model core with a constraining system around it. The core model must generalize across unseen domains. The surrounding system must enforce domain logic. By bridging general AI capabilities with structured domain-specific context, Domain-Centric AI can improve robustness, reduce manual effort, and support more reliable decision-making in safety-critical workflows. Licentiatavhandling 2026-05-13 11:39:57 https://research.chalmers.se/publication/552092 2026-05-13T11:39:57Z The objective of domain generalization (DG) is to enable models to be robust against domain shift. DG is crucial for deploying vision-language models (VLMs) in real-world applications, yet most existing methods rely on domain labels that may not be available and often ambiguous. We instead study the DG setting where models must generalize well without access to explicit domain labels. Our key idea is to represent an unseen target domain as a combination of latent domains automatically discovered from training data, enabling the model to adaptively transfer knowledge across domains. To realize this, we perform latent domain clustering on image features and fuse domain-specific text features based on the similarity between the input image and each latent domain. Experiments on four benchmarks show that this strategy yields consistent gains over VLM-based baselines and provides new insights into improving robustness under domain shift. Paper i proceeding 2026-05-13 11:35:15 https://research.chalmers.se/publication/552091 2026-05-13T11:35:15Z More compact wastewater treatment is attractive for handling increased loads and stricter effluent requirements.The moving bed biofilm reactor (MBBR) can achieve high carbon and nitrogen removal capacity at a smallfootprint but is challenged to also reach efficient enhanced biological phosphorus removal (EBPR). This pilotstudy evaluated a new approach to promote EBPR and high nitrogen removal in continuous flow MBBRs byoperating with alternating flow direction and intermittent aeration, thus enabling the cyclic anaerobic andaerobic/anoxic conditions required for polyphosphate accumulating organisms (PAOs). The MBBRs utilised anovel bio-based support material with external biofilm growth for improved mass transfer and were fed prefilteredmunicipal wastewater. To increase the availability of readily biodegradable carbon for nutrientremoval, the influent was supplied with volatile fatty acids (VFA), mimicking fermentation of sludge from primaryfiltration. The 460-day long study demonstrated stable nitrogen removal and EBPR. The presence of PAOsin the biofilm was demonstrated in batch tests as well as by microbial analysis, with Ca. Phosphoribacter, Tetrasphaeraand Ca. Accumulibacter being detected in high relative abundances. Both aerobic and anoxic phosphateuptake were observed, indicating denitrifying PAO activity. The VFA addition had a strong impact on theEBPR, which increased when directing the VFA to the anaerobic phases, compared to dosing VFA continuously.With VFA dosing in the anaerobic phases, nitrogen and phosphorus removal were 84 ± 5% and 68 ± 18%,respectively, demonstrating the possibilities with this novel process for future full-scale installations. Artikel i vetenskaplig tidskrift 2026-05-13 11:01:12 https://research.chalmers.se/publication/552090 2026-05-13T11:01:12Z We provide a framework for optimizing energy conversion processes in coherent quantum conductors fed by nonthermal resources. Such nonthermal resources, which cannot be characterized by temperatures or electrochemical potentials, occur in small-scale systems that are smaller than their thermalization length. Using scattering theory in combination with a Lagrange multiplier method, we optimize the device’s performance based on the efficiency, precision, or a trade-off between the two at a given output current. The transmission properties leading to this optimal performance are identified. We showcase our findings with the example of a refrigerator exploiting experimentally relevant nonthermal resources, which could result from competing environments or from light irradiation. We show that the performance is improved compared to a device exploiting a thermal resource. Our results can serve as guidelines for the design of energy-conversion processes in future nanoelectronic devices. Artikel i vetenskaplig tidskrift 2026-05-13 10:24:54 https://research.chalmers.se/publication/551941 2026-05-13T10:24:54Z Palladium supported on ceria (Pd/CeOx) has recently emerged as a promising electrocatalyst for the alkaline hydrogen oxidation reaction (HOR) in anion exchange membrane fuel cells. It has been proposed that CeOx provides OH spillover and modulates Pd–H binding, enhancing the reaction kinetics at the key Pd–Ce interface. Herein, we report a method to maximise the Pd–Ce interfacial area by synthesising highly ordered Pd/CeOx inverse opals (IOs) with tunable pore sizes (20–250 nm) directly on glassy carbon electrodes. The resulting IOs exhibit highly ordered pore networks which could be scaled down to the mesoporous regime (<50 nm), and dispersed palladium species, including Pd–O–Ce interfacial sites. Electrochemical measurements reveal a pore size dependence of HOR activity, with IOs fabricated from 104 nm microspheres templates delivering the highest specific activity and strongest enhancement relative to non-templated Pd/CeOx controls. Electrochemically active surface area (ECSA) estimations reveal that larger-pore IOs suffer reduced ECSA likely due to diminished support conductivity associated with thinner ceria interconnections. Increasing the Ce3+ concentration, in an effort to improve conductivity, and increasing relative Pd–O–Ce content do not improve HOR activity, highlighting the need to balance conductivity, Pd and Ce speciation and pore size. The Pd/CeOx IOs remain structurally stable after testing and interestingly, even exhibit improved kinetics after 1000 cycles. This study demonstrates that while the inverse opal architecture is a powerful route to engineer Pd–Ce interfaces, these interfaces are not the only predictor of enhanced HOR. Instead, the inverse opal pore size and interconnect thickness appear to ultimately govern the enhanced HOR kinetics and mass transport. We envision that this fabrication method for inverse opals on complex carbon substrates will allow the design of mesoporous bifunctional catalysts for gas-diffusion electrodes for applications in fuel cells and electrolyzers. Artikel i vetenskaplig tidskrift 2026-05-13 10:20:04 https://research.chalmers.se/publication/552086 2026-05-13T10:20:04Z Carbon-intensive industries account for a quarter of global annual CO2 emissions. Achieving mandated climate targets requires rapid deployment of decarbonization technologies in these industries. Such deployment typically involves substantial upfront investments amidst technical, economic, and policy uncertainties. Consequently, careful selection of decarbonization technologies or a combination thereof, coupled with measures such as process electrification and energy efficiency, is crucial.This thesis presents limitations in existing methodological approaches for comparing decarbonization pathways, spanning systems-, plant-, and site-level considerations. A generalized assessment framework was developed that addresses these limitations, with individual methodological frameworks developed in the appended papers. At the system level, extended boundaries and exergy as a metric were used to compare two CO2 capture technologies with inherently different heat and electricity demands per unit of CO2 captured, considering the perspectives of both plant owners and end-users. At the plant level, an iterative exergy-pinch analysis, combined with techno-economic analysis, was developed to identify promising process modifications in unabated process plants that maximize overall exergy utilization and CO2 avoidance, leading to successive designs towards net-zero emissions. At the site level, a site-specific techno-economic analysis was developed by incorporating quantitative and qualitative site-specific factors expected to influence the choice of decarbonization technologies. Finally, to address deployment barriers for low-emissions hydrogen, an integrated system of complementary production technologies was evaluated using a generalized optimization framework, enabling cost-optimal supply strategies under site constraints and market uncertainties. The frameworks were demonstrated in case studies on bio-CHP in a district heating system, propane dehydrogenation, and a steam cracker plant.The case study results show that integrating amine-based CO₂ capture with industrial heat pumps in bio-CHP plants could enable greater district heat delivery and provide product flexibility across heat, power, and CO₂ emissions. The iterative exergy-pinch analysis applied to the propane dehydrogenation plant identified an unconventional process modification, resulting in a substantial reduction in CO2 avoidance cost (58–70%) compared to CO2 capture from its highly diluted flue gas stream from the unmodified process. The site-specific techno-economic analysis revealed that incorporating site-specific cost factors yields higher avoidance cost estimates than standardized assessments, underscoring the risk of suboptimal technology selection. Finally, the integrated hydrogen production system demonstrated how combining multiple distinct production technologies can reduce costs, improve operational flexibility, and system redundancy. In summary, the generalized assessment framework, combining these individual framework methodologies, provides a comprehensive early-stage indication of the optimal decarbonization pathway for specific industrial sites. Doktorsavhandling 2026-05-13 09:56:40 https://research.chalmers.se/publication/548169 2026-05-13T09:56:40Z The process of developing theories and models and testing them with experiments is fundamental to the scientific method. Automating the entire scientific method then requires not only automation of the induction of theories from data, but also experimentation from design to implementation. This is the idea behind a robot scientist -- a coupled system of AI and laboratory robotics that has agency to test hypotheses with real-world experiments. In this chapter we explore some of the fundamentals of robot scientists in the philosophy of science. We also map the activities of a robot scientist to machine learning paradigms, and argue that the scientific method shares an analogy with active learning. We demonstrate these concepts using examples from previous robot scientists, and also from Genesis: a next generation robot scientist designed for research in systems biology, comprising a micro-fluidic system with 1000 computer-controlled micro-bioreactors and interpretable models based in controlled vocabularies and logic Preprint 2026-05-13 09:20:13 https://research.chalmers.se/publication/552083 2026-05-13T09:20:13Z