Research Projects

We welcome applications with any project proposals relevant to the SustAI CDT themes. The table below contains projects submitted by SustAI CDT academic staff members, and so they are an indication of the general topics that are of interest and within the remit of the CDT. Please note these are just examples and other projects within the remit of the CDT themes are also very welcome.

Important

When applying, we require you to submit your own original research proposal, even if you are interested in one or more of the projects below. For any queries, please contact the SustAI team at sustai@soton.ac.uk.

Themes

All projects belong to one or more SustAI themes.
Use the search box below to filter projects according to their theme.

TitleProject DescriptionRelevant ThemesSupervisor
An AI-Supported Mechanical Testing Protocol for Accelerating Materials QualificationOur vision is to accelerate high-fidelity validation of new, more durable materials, reducing material usage, increasing resource efficiency, minimising premature failures, and lowering operational costs for high-value manufactured products.

High-temperature materials performance involves a complex interplay of factors, including creep, fatigue, oxidation, and microstructural stability [1]. This critically limits the longevity of many high-value technologies, infrastructures, and engineered products, from electricity generation to transportation and industrial processes. The development cycle for new materials, especially for safety-critical applications, typically spans 10-20 years. This process requires rigorous validation through testing at the lab scale before scaling up to production. However, the efficiency of validation is significantly hindered by existing testing methodologies, which are laborious, time-consuming, and costly [2]. These methods are low-throughput, often evaluating one alloy composition at a time [3].

This project aims to pioneer a data-rich, high-temperature testing protocol by integrating heterogeneous testing (evaluating multiple compositions within a single test) with predictive and generative AI techniques. The goal is to accelerate the qualification of sustainable stainless steels with enhanced durability. The key objectives are:
• Develop a high-throughput testing procedure integrated with full-field strain measurements at high temperature, to efficiently evaluate material performance.
• Create AI models (e.g., an established AI model based on 2-point statistics and principal component analysis techniques [4], as well as a more exploratory AI model called generative adversarial networks [5]) to automatically quantify key microstructural features and their evolutions before and after testing.
• Establish data-driven models with enhanced predictive capabilities to assess material damage sensitivity and recommend the next test to further refine accuracy.

The methodology and associated tasks are described below.

Task 1: We will fabricate a stainless steel sample with varied compositions. Using scanning electron microscopy (SEM), we will capture high-resolution images, focusing on the grain-level microstructural features at 50 distinct points, each representing a unique composition. After subjecting the sample to high-temperature testing for a given time, we will re-image the same areas to observe microstructural changes. These images will be divided into smaller sections, resulting in approximately 50,000 paired images before and after testing.

Task 2: We will run established algorithms to identify regions where cracks or other forms of damage have occurred. Using these data, we will train AI deep-learning methods to predict damage based on the initial microstructure and composition. By analysing how the microstructure evolves, we aim to understand the relationship between composition and durability.

Task 3: we will employ advanced AI techniques, such as generative adversarial networks (GANs), to predict the behaviour of new, untested microstructures. This approach not only accelerates material development but also aligns with sustainable AI practices by minimising the need for extensive physical testing.

AI will streamline the entire material testing process, including the collection and processing of large volumes of microstructure data, ultimately delivering faster and more accurate prognostics. While the focus is on stainless steel, the AI-enhanced end-to-end material testing protocol developed by this project is designed to be adaptable, offering benefits well beyond this material type.

References:
1. Eswarappa Prameela, S., Pollock, T.M., Raabe, D. et al. Materials for extreme environments. Nat Rev Mater 8, 81–88 (2023). https://doi.org/10.1038/s41578-022-00496-z
2. Materials 4.0 – Predicting and Controlling Materials’ Microstructures and Performance, Henry Royce Institute, https://www.royce.ac.uk/collaborate/roadmapping-landscaping/materials-4-0/
3. Couet, A., Integrated high-throughput research in extreme environments targeted toward nuclear structural materials discovery. J Nucl Mater 559, 153425 (2022)
4. Cameron, B.C., Tasan, C.C. Microstructural damage sensitivity prediction using spatial statistics. Sci Rep 9, 2774 (2019). https://doi.org/10.1038/s41598-019-39315-x
5. Khan, A., Lee, CH., Huang, P.Y. et al. Leveraging generative adversarial networks to create realistic scanning transmission electron microscopy images. npj Comput Mater 9, 85 (2023). https://doi.org/10.1038/s41524-023-01042-3		AI for Sustainable Operations and Circular EconomyBo Chen
AI Adoption in Carbon Cap-and-Trade within Supply Chains1. Introduction
Climate change pressures and tightening environmental regulations have led many governments to adopt carbon cap-and-trade systems as a key mechanism. These schemes set emission caps for firms, while enabling trading of carbon allowances to ensure cost-efficient abatement (Xu et al. 2017). Within this context, the adoption of Artificial Intelligence (AI) has emerged as a promising enabler for two critical aspects: (i) From micro perspective, AI helps monitoring and verifying real-time emissions at the firm and supply chain level (Yin et al. 2024), eliminating the potential information asymmetry in supply chains. (ii) From macro perspective, AI can forecast carbon policies and carbon allowances to mitigate market uncertainties (Jazairy et al. 2025).

Based on the above background, the motivation for this study stems from several important challenges. First, while monitoring AI (micro) can eliminate information asymmetries in system, the high-emission firm is less incentivized to implement as benefit might be hurt. Second, forecasting AI (macro) can reduce uncertainties, but its effectiveness depends on model accuracy, which may incur a significant cost. Firms must balance this cost against the benefits of improved accuracy. Third, in competitive supply chains, rivals may adopt AI at different stages, creating strategic asymmetries that influence both market outcomes and environmental performance. These considerations highlight the need for a rigorous analytical study of AI adoption strategies in carbon markets.

Existing literature has studied carbon cap-and-trade in supply chains extensively (Kroes et al. 2012, Huang et al. 2021). There is also growing work on AI adoption in supply chains, primarily focusing on predictive analytics and operational efficiency (Mithas et al. 2022, Jazairy et al. 2025). However, the intersection of AI adoption and carbon market dynamics remains underexplored, particularly with regard to enhancing environmental sustainability in supply chains.

2. Research Aim and Questions
This project seeks to examine how firms in competing supply chains should strategically adopt AI under a carbon cap-and-trade system. The key research questions (RQs) include:

RQ1. What is the optimal AI adoption strategy, i.e., what to adopt (monitoring vs. forecasting), when to adopt (timing), and to what extent (accuracy level)?
RQ2. How does AI adoption change firms’ strategic decisions (e.g., pricing, quantity) in carbon trading and supply chain competition?
RQ3. How is the effectiveness of AI adoption on carbon reduction and supply chain performance under cap-and-trade? What are the implications of AI adoption for the long-term stability of carbon markets?

3. Methodology
The project will adopt game-theoretical approaches to model competing supply chains under cap-and-trade regulation. Firms’ strategies will include whether, when, and how to adopt AI for emission monitoring and policy forecasting. Unique features such as information asymmetry, policy uncertainty, and consumers’ environmental concerns will be captured in model settings. Equilibrium analysis will reveal the conditions under which AI adoption is optimal, and comparative statics will show its impact on emissions, profits, and social welfare.

4. Expected Outcomes
This research contributes by bridging AI adoption and carbon cap-and-trade in the context of competitive supply chains. It advances theory in sustainable operations and provides insights for firms and policymakers on leveraging AI for emissions reduction. The project is expected to yield the following outcomes:

(i) Publishing/submitting 3 research articles that develop innovative analytical models to identify optimal AI adoption strategies under varying considerations (e.g., different roles of AI, market contexts, and consumer behaviors).
(ii) Attending at least 1 international academic conference in operations management domain to present the major findings.
(iii) Producing a policy brief for policy-makers to highlight key policy implications and practical recommendations derived from the study.

References
[1] Huang, X., Tan, T., & Toktay, L. B. (2021). Carbon leakage: The impact of asymmetric regulation on carbon‐emitting production. Production and Operations Management, 30(6), 1886-1903.
[2] Jazairy, A., Shurrab, H., & Chedid, F. (2025). Impact pathways: walking a tightrope—unveiling the paradoxes of adopting artificial intelligence (AI) in sales and operations planning. International Journal of Operations & Production Management, 45(13), 1-27.
[3] Kroes, J., Subramanian, R., & Subramanyam, R. (2012). Operational compliance levers, environmental performance, and firm performance under cap and trade regulation. Manufacturing & Service Operations Management, 14(2), 186-201.
[4] Mithas, S., Chen, Z. L., Saldanha, T. J., & De Oliveira Silveira, A. (2022). How will artificial intelligence and Industry 4.0 emerging technologies transform operations management? Production and Operations Management, 31(12), 4475-4487.
[5] Xu, X., He, P., Xu, H., & Zhang, Q. (2017). Supply chain coordination with green technology under cap-and-trade regulation. International Journal of Production Economics, 183, 433-442.
[6] Yin, Y., Wang, H., & Deng, X. (2024). Real-time logistics transport emission monitoring-integrating artificial intelligence and internet of things. Transportation Research Part D: Transport and Environment, 136, 104426.		AI for Sustainable Operations and Circular EconomyXiaoyan Xu
Large Language Models for Illegal Wildlife Trade MonitoringThe Illegal Wildlife Trade (IWT) of species (flora and fauna) is having a direct impact on endangered species worldwide, putting at risk biodiversity and environmental sustainability. In the plant domain poachers are routinely destroying vast habitats for CITES protected species, such as rare cacti which take 100+ years to grow in the wild, which once gone are irreplaceable. Poachers will traffic illegally obtained wildlife to low regulation third party countries (e.g. China) where wildlife is sold via online auction sites such as Alibaba, eBay and Amazon using euphemisms and trade jargon to avoid online platform detection and takedown of the adverts. Law enforcement (e.g. UK Border Force) need AI to efficiently monitor IWT for target species to allow better intelligence led interventions.

This PhD project will explore how to use Multimodal Large Language Models to perform Information Extraction on auction HTML web pages obtained from web crawling software (e.g. MEMEX crawler). LLMs have recently been shown [1] effective at automatically parsing HTML for information extraction. This PhD will explore human-in-the-loop LLM training ideas (e.g. deep active learning, rationale-based training, multi-agent debating, LLM preference alignment) allowing efficient LLM model fine-tuning through a set of feedback sessions that conservation experts could participate in. The downstream application goal is to develop LLM algorithms that can be used with a crawler to build a IWT intelligence picture about what species are being traded, by whom and where.

An inter-disciplinary project, a conservation expert from Kew Gardens will join the team as an external supervisor to provide IWT expertise. If successful, results can be presented to contacts within the UK's National Wildlife Crime Unit and UK Border Force for potential tool adoption by law enforcement officers.

[1] https://serpapi.com/blog/traditional-parsing-vs-mistral-7b/		AI for the Natural EnvironmentStuart Middleton
Advancing sustainable and resilient marine fisheries for the future using AISustainable development is traditionally viewed in relation to three fundamental pillars that represent the social, economic and environmental domains. This approach has been criticised for failing to acknowledge the intricate and interconnected interactions between the pillars and the dynamic nature of the systems in which they function. Instead, silos are reinforced and governance fragmented(1), leading to failure to address complex challenges(2). To manage natural resources in a more sustainable way a systems-based approach is now needed to identify and capitalise on synergies that may exist between sectors, while minimising or even avoiding trade-offs that result in negative consequences(3).

The long-term overexploitation of marine fish stocks(4) has rendered them poorly able to buffer, adapt and recover from additional stressors and shocks (e.g. policy, economic, climate). Consequently, the fisheries and fishing communities the fish populations support are also imperilled and lack long-term resilience. Acknowledging that fisheries are social-ecological systems(5) in which the fish stocks themselves form the foundations on which economic prosperity and wellbeing of fishing communities depends, future management should adopt a systems dynamics-based approach to enhance sustainability and resilience.

Focusing on marine fisheries, this interdisciplinary project will use AI to investigate how fishing communities responds to a series of shocks so that lessons may be learnt on how resilience of the resource and the people it supports may be reinforced.

(1) Creating a conceptual systems dynamics model of case-study marine fisheries using causal loop and stocks and flows diagrams and where appropriate build a systems model to test assumptions.

(2) Investigating and quantifying fishing activity over time-scales that cover the period of the shock. AI will be used to enhance the efficiency of interrogating large data-sets available, such as that provided by employing: (a) Automated Identification Systems (AIS) that track fishing vessel activity; (b) geospatial remote sensing satellite data of fishing vessel abundance (e.g. operating in less developed regions of the world where AIS data is unavailable); and social media data for inference when direct evidence is unavailable and difficult to obtain (e.g. for very small vessels or recreational boats).

(3) Exploring shifts in catch from available data on landings for specific species and vessel types.

(4) Obtaining information related to the status of specific fish stocks.

(5) Correlating measures of activity and catch / stock status with metrics that document the consequences of shocks.

The study will help improve understanding of how those dependent on marine fisheries to supply food, respond and adapt in the face of systemic shocks, enabling complex feedback loops to be evaluated and trade-offs, synergies and unintended consequences to be identified. Lessons will be learned on how to enhance the resilience of primary resources and the communities that depend on them in the face of ongoing existential threats, such as climate change.

References:
1Bogers, M., Biermann, F., Kalfagianni, A., Kim, R. E., Treep, J. and Vos, M. G.. 2022. The impact of the Sustainable Development Goals on a network of 276 international organizations. Global Environmental Change 76, DOI: 10.1016/j.gloenvcha.2022.102567.
2UN 2023. Halfway to 2030, world ‘nowhere near’ reaching Global Goals, UN warns https://news.un.org/en/story/2023/07/1138777
3Kemp, P. S., Acuto, M., Larcom, S., Lumbroso, D. and Owen, M. 2022. Exorcising Malthusian ghosts: Vaccinating the nexus to advance integrated water, energy and food resource resilience. Current Research in Environmental Sustainability, 4, [100108]. DOI:10.1016/j.crsust.2021.100108).
4Thurstan, R., Brockington, S. & Roberts, C. The effects of 118 years of industrial fishing on UK bottom trawl fisheries. Nat Commun 1, 15 (2010). https://doi.org/10.1038/ncomms1013
5Ostrom, E. 2009. A general framework for analyzing sustainability of social-ecological systems. Science 325, 5939, 419-422. DOI: 10.1126/science.1172133
AI for the Natural EnvironmentPaul Kemp
Using AI to advance environmental monitoring of freshwater ecosystems Freshwater ecosystems or the most degraded and threatened of all ecosystems on the planet(1). They also host a high diversity of vertebrate species, many of which are critically endangered (e.g. cetaceans, reptiles, amphibians, and fish)(2). The high biodiversity and threats from human activity create “biodiversity hotspots”(3).

Threats are posed from a range of anthropogenic activities that include overexploitation for food, habitat degradation and destruction, water pollution, and fragmentation of rivers through the construction of infrastructure (e.g. large dams) required to support increased urbanisation. Technologies and management strategies are developed to mitigate environmental impacts of these activities, such as the construction of fish passes designed to facilitate the negotiation of dams by migratory fish, or screening systems the prevent them entering dangerous areas, such as hydropower turbines.

Unfortunately, the efficiencies of fish passes and screens can be highly variable, ranging from those that work relatively well, to others that totally fail – and in the worst cases can have negative impacts on the species they were designed to protect(4). To advance the next generation of technologies. There is a need to understand why current system fail through effective monitoring of their operations and outcomes. The results can then be used to inform future design. A range of techniques are used to monitor the performance of environmental impact mitigation technology, including the use of video imagery and photography, acoustic cameras when visual techniques are inappropriate (e.g. when dark or under turbid conditions), telemetry used to describe trajectories of the animals themselves (e.g. radio or acoustic tagging), or other techniques such as aerial or satellite imagery to track animal movements. However, such data is often poorly analysed due to ineffective techniques needed to handle big data with considerable amounts of material archived never to be reviewed. AI is likely to disrupt this field and be a major game changer through enhancing the efficiency of current environmental monitoring techniques(5).

Working with the Environment Agency and the water industry this project aims to advance effective monitoring of environmental impact mitigation technologies designed to protect freshwater ecosystems through the use of AI. The key objectives are:

1. Review the range of data collected by the Environment Agency for the purpose of monitoring the effectiveness of environmental impact mitigation technology, such as fish passes or screening systems, or more generic water quality information available to identify pollution events.

2. Prioritise the greatest needs of the regulatory agency to close information gaps in key areas related to protecting freshwater ecosystems.

3. Identify available data that would help address the challenges identified.

4. Build appropriate AI solutions to resolve the data challenges selected to improve UK environmental monitoring programmes operating in freshwater ecosystems.

References

1. https://doi.org/10.1017/S1464793105006950
2. https://doi.org/10.1016/S2542-5196(23)00275-9
3. https://wwf.panda.org/discover/our_focus/freshwater_practice/freshwater_biodiversity_222/
4. https://doi.org/10.1002/9781118394380.ch52
5. https://pml.ac.uk/news/revolutionizing-biodiversity-monitoring-the-power-of-ai-and-new-technologies/
AI for the Natural EnvironmentPaul Kemp
Sustainable Generative AI ModelsProject Summary

This project aims to pioneer advancements in the energy-efficient Generative AI models (GenAI), focusing on achieving faster inference times and reduced model sizes without compromising performance and increasing carbon emissions. As GenAI becomes increasingly central to a wide range of applications, from generating images to generating videos and music, their computational demand and the time required for training and inference have escalated. This research seeks to address these challenges by developing innovative techniques for efficiency, including architectural innovations, compression strategies, algorithmic improvements, and system level optimizations. The goal is to enable the deployment of state-of-the-art GenAI models across broader scenarios of computing environments, from high-end servers to consumer-level machines. This project will contribute to making GenAI more democratic, efficient, and scalable, paving the way for their application in real-time and resource-constrained scenarios.

Objectives

1. To develop cutting-edge techniques for model compression, such as pruning, quantization, and knowledge distillation, tailored for GenAI models.
2. To design and experiment with new GenAI architectures that are more efficient, requiring less computational power and memory.
3. To create new algorithms and system wide optimizations to accelerate both training and inference processes for GenAI, making them more suitable for deployment across a variety of computing environments.
4. To develop and utilize benchmarks and metrics specifically designed to evaluate the energy-aware efficiency and performance of GenAI under various computational constraints.

Methodology

Via a thorough literature review and benchmarking of existing models, we aim to identify key limitations and research gaps ripe for advancement. Possible solutions may relate to tailored pruning techniques, which selectively remove less critical parameters, alongside quantization methods that lower the precision of numerical values to significantly decrease model sizes. These optimization methods are still under-explored in the area of GenAI. Moreover, knowledge distillation will be utilized to investigate the potential for subgraph optimization. We also aim to investigate potential of distributed computing along with memory reuse and caching. These strategic directions are anticipated to reduce energy consumption, inference times and computational demands, without sacrificing the performance of the models.

You might be also asked to explore new model architectures designed to balance computational efficiency with robust performance, focusing on the potential of subgraph decomposition to break down complex models into simpler, more manageable components. This strategy aims to facilitate more efficient processing and parallelization. Enhancing the underlying algorithms of GenAI models, such as optimizing attention mechanisms and streamlining data processing, will be another crucial aspect for improving overall model responsiveness and efficiency. Additionally, the project will investigate adaptive deployment strategies, enabling models to dynamically adjust their computational complexity in real-time, ensuring optimal performance across diverse hardware environments. We will conduct comprehensive experimental validation and testing, this approach seeks to push the boundaries of what's possible with GenAI, setting new standards for efficiency and accessibility in AI technologies.

References
[1] R. Verdecchia, J. Sallou, L. Cruz, A systematic review of Green AI, Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 13 (2023).
[2] Li, Yanyu, et al. "Snapfusion: Text-to-image diffusion model on mobile devices within two seconds." Advances in Neural Information Processing Systems 36 (2024).
[3] Li, Yanjing, et al. "Q-dm: An efficient low-bit quantized diffusion model." Advances in Neural Information Processing Systems 36 (2024)
[4] Wang, Yunke, et al. "Learning to schedule in diffusion probabilistic models." Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2023.

Justification for Team

This project will require expertise from the domain of ML, Generative AI and Systems. The main supervisor Dr Zhiwu will provide his expertise on ML and GenAI. He has published numerous works in top conferences such as CVPR, NeurIPS and AAAI. The external supervisor Dr Jagmohan from UCL will provide his expertise on efficient ML systems, edge computing and on device learning crucial to implement developed techniques on devices and understand the tradeoffs between accuracy and efficiency. He has a track record in publishing papers in top conferences such as ICML, NeurIPS, SenSys and IPSN. Together, Dr Zhiwu and Dr Jagmohan will be able to provide in depth expertise to lead the student to successful completion of their PhD. More details about the supervisors are present on their Google Scholar pages. Link here: https://scholar.google.com/citations?user=f9gTyI4AAAAJ&hl=en and here
https://scholar.google.com/citations?user=yh6t92AAAAAJ&hl=en

PhD desirable skills
Strong ML background
Good Programming Skills
Background in Systems will be an additional advantage but is not necessary
Sustainable AIZhiwu Huang
Sustainability Integration in Energy Infrastructure Project Prediction: Extending greyfly.ai’s Intelligent Project Prediction Platform.The transition to sustainable energy infrastructure represents a critical contemporary challenge requiring sophisticated coordination between technological innovation, project governance, and environmental stewardship (Alka et al., 2025; Elhamahmy et al., 2025). Established project prediction methodologies exhibit fundamental limitations in their capacity to integrate sustainability considerations with traditional performance metrics (Banihashemi et al., 2017; Silvius & Schipper, 2014; Stanitsas et al., 2021), creating systematic inefficiencies in energy infrastructure delivery. This limitation becomes particularly acute when deploying artificial intelligence solutions (Dacre et al., 2025; Dacre & Kockum, 2022a; Ghafari & Samaei, 2025), which introduce additional processing complexity whilst offering transformative potential for optimising complex energy systems.

Contemporary project governance frameworks predominantly employ retrospective analysis of cost, schedule, and technical risks (Armenia et al., 2019), hitherto fail to incorporate forward-looking sustainability metrics that capture lifecycle environmental impacts, resource circularity, and systemic interdependencies (Kivilä et al., 2017; Rosengart et al., 2023). This methodological limitation stems from the inherent complexity of operationalising sustainability indicators within real-time predictive models (Klug & Kmoch, 2015; Olawumi & Oladapo, 2025), particularly when addressing the temporal and spatial variations characteristic of energy infrastructure projects. The International Energy Agency projects that global energy infrastructure investment must reach $4.5 trillion annually through 2030 to achieve net-zero targets (Abdulla et al., 2024), however established project prediction methodologies largely treat sustainability as an external constraint rather than an integral component of predictive modelling frameworks (Sabini & Alderman, 2021). This approach is compounded by the inherent complexity of communicating sustainability trade-offs to stakeholders, who require both technical accuracy and explanations of AI-driven recommendations (Kunkel et al., 2023).

Furthermore, project management approaches typically treat individual projects as isolated entities (Martinsuo et al., 2022), failing to capture complex interdependencies between project delivery outcomes, broader energy system resilience, and supply chain sustainability (da Silva et al., 2025). This reductionist perspective undermines the systemic benefits that AI could potentially deliver in coordinating multiple infrastructure projects across regional or national scales (Chen et al., 2025). The emergence of digital transformation approaches in energy infrastructure necessitates new theoretical frameworks that can accommodate the complexity of sustainability-aware project prediction (Gong et al., 2022; Zarifis, 2024; Zarifis & Cheng, 2023).

As such this PhD will extend greyfly.ai's Intelligent Project Prediction (IPP) platform to incorporate the sustainability implications of large-scale energy infrastructure projects by addressing three interconnected challenges:

• First, the development of methodologies for identifying, modelling, and integrating sustainability indicators alongside traditional project data. This involves creating frameworks that can process lifecycle carbon assessments, energy demand projections, and circularity measures within predictive models without compromising computational efficiency.
• Second, the application of explainable AI techniques to ensure that sustainability-related predictions maintain transparency and persuasive power for decision-makers across diverse stakeholder groups.
• Third, the implementation of systems thinking approaches to capture interdependencies between project delivery, energy systems, and supply chains, ensuring that AI predictions reflect real-world complexity rather than simplified project-level metrics (Dacre & Kockum, 2022b).

The research methodology will employ a combination of predictive model development, theoretical framework construction, and empirical validation through industry partnership with greyfly.ai (Brookes et al., 2020; Hsu et al., 2021). The investigation will begin with a systematic examination of sustainability metrics applicable to energy infrastructure projects, developing novel multi-dimensional scoring systems that incorporate environmental, economic, and social indicators. This foundation will support the creation of interpretable models specifically optimised for multi-criteria sustainability assessment, with particular attention to stakeholder-specific communication requirements (Dacre & Kockum, 2022a; Kunkel et al., 2023).

Expected academic contributions include novel theoretical frameworks for integrating sustainability science with project management AI systems, peer-reviewed outputs for explainable AI in multi-stakeholder sustainability contexts, and empirical evidence regarding the effectiveness of systems approaches to project prediction in energy infrastructure. Practical contributions will comprise validated prototype implementations demonstrating the operationalisation of sustainability metrics within commercial project prediction platforms, governance frameworks for sustainable project supply chain management, and policy recommendations for regulatory adoption of AI-powered sustainability assessment tools.

Whilst early outputs will align with greyfly.ai's ongoing development initiatives, the scope of the PhD extends significantly further. The doctoral research will address longer-term challenges that reach beyond a typical 9-month development cycle, including the integration of lifecycle sustainability metrics into predictive project models, the application of explainable AI to improve stakeholder trust, and the design of governance approaches for sustainable project supply chains.

References
Abdulla, S., Brown, J., Gratcheva, E., Greig, C., Iychettira, K., Spokas, K., Syed, M., & Waltzer, K. (2024). Clean Energy Finance Must Address Three Crucial Challenges to Achieve Ambitious Climate Goals. Available at SSRN 5016309.
Alka, T., Raman, R., & Suresh, M. (2025). Critical success factors for successful technology innovation development in sustainable energy enterprises. Scientific Reports, 15(1), 14138.
Armenia, S., Dangelico, R. M., Nonino, F., & Pompei, A. (2019). Sustainable project management: A conceptualization-oriented review and a framework proposal for future studies. Sustainability, 11(9), 2664.
Banihashemi, S., Hosseini, M. R., Golizadeh, H., & Sankaran, S. (2017). Critical success factors (CSFs) for integration of sustainability into construction project management practices in developing countries. International journal of project management, 35(6), 1103-1119.
Brookes, N., Lattuf Flores, L., Dyer, R., Stewart, I., Wang, K., & Dacre, N. (2020). Project Data Analytics: The State of the Art and Science. Association for Project Management.
Chen, K., Zhou, X., Bao, Z., Skibniewski, M. J., & Fang, W. (2025). Artificial intelligence in infrastructure construction: A critical review. Frontiers of Engineering Management, 12(1), 24-38.
da Silva, A. D. S., da Silva, W. V., da Silva, L. S. C. V., da Cruz, N. J. T., Su, Z., & da Veiga, C. P. (2025). Interdependence between supply chains and sustainable development: global insights from a systematic review. Review of Managerial Science, 19(3), 931-962.
Dacre, N., Baxter, D., Dong, H., Al-Mhdawi, M. K. S., Abeysooriya, R., & Shen, Y. (2025). Digital transformation and the AI imperative in public and private sector projects: Methods and skills for project management. Association for Project Management.
Dacre, N., & Kockum, F. (2022a). Artificial Intelligence in Project Management: A review of AI’s usefulness and future considerations for the project profession. Association for Project Management.
Dacre, N., & Kockum, F. (2022b). Strategic Project Leadership: The Case for Integrating Systems Thinking and Artificial Intelligence Operational Research Society, 20.
Elhamahmy, A., Gohar, H. T. H., & Galal, A. (2025). Sustainable project management in renewable energy and infrastructure projects: Challenges and opportunities. International Journal of Engineering Research & Technology, 14(05).
Ghafari, R., & Samaei, S. R. (2025). Integrated AI and digital twin technologies for green project management in resilient coastal and port infrastructure systems. Proceedings of the Third International Conference on Advanced Research in Civil Engineering, Architecture, and Urban Planning, Munich, Germany,
Gong, Z., Dacre, N., & Dong, H. (2022). Fostering digital transformation through project integration management. Project and Program Management, 19.
Hsu, M.-w., Dacre, N., & Senyo, P. K. (2021). Applied Algorithmic Machine Learning for Intelligent Project Prediction: Towards an AI Framework of Project Success. Advanced Project Management, 21, 6.
Kivilä, J., Martinsuo, M., & Vuorinen, L. (2017). Sustainable project management through project control in infrastructure projects. International journal of project management, 35(6), 1167-1183.
Klug, H., & Kmoch, A. (2015). Operationalizing environmental indicators for real time multi-purpose decision making and action support. Ecological Modelling, 295, 66-74.
Kunkel, S., Schmelzle, F., Niehoff, S., & Beier, G. (2023). More sustainable artificial intelligence systems through stakeholder involvement? GAIA-Ecological Perspectives for Science and Society, 32(1), 64-70.
Martinsuo, M., Teerikangas, S., Stensaker, I., & Meredith, J. (2022). Managing strategic projects and programs in and between organizations. International journal of project management, 40(5), 499-504.
Olawumi, M. A., & Oladapo, B. I. (2025). AI-driven predictive models for sustainability. Journal of Environmental Management, 373, 123472.
Rosengart, A., Granzotto, M., Wierer, R., Pazzaglia, G., Salvi, A., & Dotelli, G. (2023). The green value engineering methodology: A sustainability-driven project management tool for capital projects in process industry. Sustainability, 15(20), 14827.
Sabini, L., & Alderman, N. (2021). The paradoxical profession: Project management and the contradictory nature of sustainable project objectives. Project Management Journal, 52(4), 379-393.
Silvius, A., & Schipper, R. P. (2014). Sustainability in project management: A literature review and impact analysis. Social business, 4(1), 63-96.
Stanitsas, M., Kirytopoulos, K., & Leopoulos, V. (2021). Integrating sustainability indicators into project management: The case of construction industry. Journal of Cleaner Production, 279, 123774.
Zarifis, A. (2024). Leadership in Fintech Builds Trust and Reduces Vulnerability More When Combined with Leadership in Sustainability. Sustainability, 16(13), 5757.
Zarifis, A., & Cheng, X. (2023). The five emerging business models of Fintech for AI adoption, growth and building trust. In Business digital transformation: selected cases from industry leaders (pp. 73-97). Springer.
AI for Sustainable Energy and Buildings, AI for Sustainable Operations and Circular EconomyNicholas Dacre
Explainable artificial intelligence-based digital twins for optimal renewable energy harvesting under inherent uncertaintyThe renewable energy sector is experiencing increased demand due to recent energy crises and ever-increasing consumption scenarios around the world. Wind energy [1, 2], in particular, holds promise for reducing fossil fuel dependence and attracting global investments, with an annual growth rate of over 10%.

Wind turbines are energy generation systems that need to be optimized for different environments, which can be costly due to a lack of knowledge and uncertainties. Vibration and dynamic stability remain persistent issues for wind turbines, limiting the potential of this technology. This project aims to develop smart wind turbines for a sustainable energy supply system based on intelligent materials and active structures for optimal dynamic performance [3, 4] while maximizing the power output considering temporal fluctuations and uncertainty [5] in wind speed and energy demand. An explainable artificial intelligence-based digital twin will be developed for active vibration control and optimal power output through quantification and minimization of associated uncertainties with a component of intellectual oversight and human intelligence.

Modern wind turbines operate in highly dynamic and uncertain environments, where unmitigated vibration and fluctuating wind loads can reduce performance and shorten operational life. To address this, the project will create a physics-informed digital twin that learns from real-time sensor data and updates predictions on structural behaviour, power output, and health status, ensuring that model predictions and decisions are interpretable and trustworthy. These methods not only support greater annual energy production and longer service life but also reduce the embodied and operational carbon of wind energy systems.

The project will make the doctoral student an expert in multidisciplinary areas concerning renewable energy, sustainable system design, digital twins and artificial intelligence that can be explained with human insights, optimization under uncertainty and multi-scale design for optimal structural performance. Such unique and multi-disciplinary skills developed while contributing to an impactful and ambitious research project in collaboration with leading industry partners will prepare the doctoral candidate to be a global leader in tackling challenging research projects concerning energy sustainability through deep technological innovation.

This PhD project is suitable for ambitious PhD students who are willing to make impactful contributions towards sustainable energy generation exploiting the emerging concepts of artificial intelligence and advanced system designs. Students with background in any one of the following areas (and willing to learn new concepts) are suitable for this project: engineering, physics, materials and manufacturing, and artificial intelligence. If you have further queries, do not hesitate to contact: T.Mukhopadhyay@soton.ac.uk

References:
(1) Herbert, G. J., Iniyan, S., Sreevalsan, E., & Rajapandian, S. (2007). A review of wind energy technologies. Renewable and sustainable energy Reviews, 11(6), 1117-1145.
(2) Garcia Marquez, F. P., Peinado Gonzalo, A. (2022). A comprehensive review of artificial intelligence and wind energy. Archives of Computational Methods in Engineering, 29(5), 2935-2958
(3) Sinha, P., & Mukhopadhyay, T. (2023). Programmable multi-physical mechanics of mechanical metamaterials. Materials Science and Engineering: R: Reports, 155, 100745.
(4) Xie, F., Aly, A. M. (2020). Structural control and vibration issues in wind turbines: A review. Engineering Structures, 210, 110087.
(5) Dey, S., Mukhopadhyay, T., Adhikari, S. (2018). Uncertainty quantification in laminated composites: a meta-model based approach. CRC Press.		AI for Sustainable Energy and BuildingsTanmoy Mukhopadhyay
AI-powered and Swarm intellegence inspired point-of-care biosensor for environmental monitoring Background:
1. Problem we’re solving
Unsafe water and surfaces are under-monitored because microbial contamination testing is slow, lab-dependent, and costly. Culture and PCR are accurate but require skilled staff and instruments, with hours–days turnaround, so frequent, on-site, quantitative checks rarely happen. Existing smartphone biosensors improve portability but often struggle with sensitivity, robustness, and repeatable quantification outside the lab.
2. Why swarm intelligence?
Swarm intelligence describes collective behaviours that emerge from simple local interactions among individuals and their environment, without central control [1–4]. In nature, many organisms exhibit such behaviours (e.g., fish schools, bird flocks), which enhance survivability and enable complex, adaptive decision-making under changing conditions [5–7]. These systems rely on self-organisation and decentralised rules to generate global patterns from local cues [8,9]. Inspired by this, swarm principles have driven breakthroughs in AI [10,11], robotics [12–15], optimisation [16–20], and network systems [21–24], yet remain underexplored in healthcare diagnostics. Here we mimic swarm principles by using magnetic microbeads (MBs) as “agents.” Antibody-coated MBs bind target proteins and, under a tuned magnetic field, self-assemble into distinct, naked-eye-visible patterns. These macroscale patterns emerge from simple local rules (binding + dipolar interactions + field guidance) and encode analyte identity and concentration.
3. Why AI?
While these bead patterns are visually distinctive, real-world imaging introduces variability in lighting, focus, and background that makes manual interpretation unreliable. AI, and particularly YOLO-based object detection, enables rapid, robust, and on-device analysis of these complex visual signals. It can classify patterns, estimate contaminant concentration, and improve continuously as more labelled data are added. Thus, AI could bridge the gap between emergent swarm patterns and quantitative, reproducible diagnostics suitable for field deployment. With on-phone AI analysis, the readout is rapid and user-friendly, removing bulky equipment and specialised training.

Project summary:
This PhD project will develop a swarm-inspired, AI-enabled biosensing platform for rapid on-site detection of microbial contamination. Building on preliminary work where magnetic microbeads self-assemble into visible patterns in response to proteins, the project will extend the approach to detect both proteins and nucleic acids, the two key biomarkers of microbial presence. The research will investigate the fundamental mechanisms behind pattern formation, examine how environmental conditions affect reproducibility, and establish design rules for dual-mode sensing. Captured images will be analysed using YOLO for fast and robust classification and quantification, benchmarked against conventional machine learning methods. Validated against culture, PCR, and ELISA/isothermal assays on spiked and real environmental samples, the outcome will be a portable, intelligent, and scalable diagnostic tool that advances first-principles understanding of swarm-based biosensing.

Research questions:
1. What physicochemical interactions and environmental factors govern the self-assembly of magnetic microbeads into stable, analyte-specific patterns?
2. How do protein- and nucleic-acid-driven binding events produce distinct, reproducible patterns, and what design rules link biomolecular recognition to macroscopic readouts?
3. Can YOLO-based image analysis reliably classify bead patterns and quantify microbial contamination under variable real-world imaging conditions, and how does it compare with CNN/transfer learning approaches?
4. How does the dual-mode platform perform in terms of sensitivity, specificity, time-to-result, and robustness when benchmarked against culture, PCR/RT-qPCR, and ELISA/isothermal methods?
5. Can modelling of pH, ionic strength, temperature, and microbial load enable calibration strategies that maintain accuracy for on-site microbial monitoring?

Expected outcome
The expected outcome of this PhD project is a portable, intelligent, and dual-mode biosensing platform that can rapidly and reliably detect microbial contamination on site by recognising both protein and nucleic acid biomarkers. Scientifically, the work will generate first-principles design rules linking molecular recognition events to emergent, analyte-specific bead patterns, deepening understanding of swarm-like self-assembly in biosensing. Technologically, it will deliver a validated YOLO-based AI pipeline capable of robust pattern classification and quantification across variable field conditions. Practically, the platform will be benchmarked against culture, PCR, and ELISA/isothermal standards, demonstrating comparable sensitivity and specificity within minutes rather than hours or days. Overall, the project is expected to produce both a fundamental advance in swarm-inspired biosensor design and a field-ready prototype that enables frequent, on-site microbial monitoring.

Reference:
1 Hinchey, M. G., Sterritt, R. & Rouff, C. Swarms and swarm intelligence. Computer 40, 111-113 (2007).
2 Kaspar, C., Ravoo, B. J., van der Wiel, W. G., Wegner, S. V. & Pernice, W. H. The rise of intelligent matter. Nature 594, 345-355 (2021).
3 Krause, J., Ruxton, G. D. & Krause, S. Swarm intelligence in animals and humans. Trends in ecology & evolution 25, 28-34 (2010).
4 Buhl, C. et al. From disorder to order in marching locusts. Science 312, 1402-1406 (2006).
5 Sasaki, T., Granovskiy, B., Mann, R. P., Sumpter, D. J. & Pratt, S. C. Ant colonies outperform individuals when a sensory discrimination task is difficult but not when it is easy. Proceedings of the National Academy of Sciences 110, 13769-13773 (2013).
6 Cavagna, A. et al. Scale-free correlations in starling flocks. Proceedings of the National Academy of Sciences 107, 11865-11870 (2010).
7 Berdahl, A., Torney, C. J., Ioannou, C. C., Faria, J. J. & Couzin, I. D. Emergent sensing of complex environments by mobile animal groups. Science 339, 574-576 (2013).
8 Garnier, S., Gautrais, J. & Theraulaz, G. The biological principles of swarm intelligence. Swarm intelligence 1, 3-31 (2007).
9 Ballerini, M. et al. Interaction ruling animal collective behavior depends on topological rather than metric distance: Evidence from a field study. Proceedings of the national academy of sciences 105, 1232-1237 (2008).
10 Warnat-Herresthal, S. et al. Swarm learning for decentralized and confidential clinical machine learning. Nature 594, 265-270 (2021).
11 Yang, L. et al. Autonomous environment-adaptive microrobot swarm navigation enabled by deep learning-based real-time distribution planning. Nature Machine Intelligence 4, 480-493 (2022).
12 Li, S. et al. Particle robotics based on statistical mechanics of loosely coupled components. Nature 567, 361-365 (2019).
13 Krieger, M. J., Billeter, J.-B. & Keller, L. Ant-like task allocation and recruitment in cooperative robots. Nature 406, 992-995 (2000).
14 Halloy, J. et al. Social integration of robots into groups of cockroaches to control self-organized choices. Science 318, 1155-1158 (2007).
15 Yang, K., Won, S., Park, J. E., Jeon, J. & Wie, J. J. Magnetic swarm intelligence of mass-produced, programmable microrobot assemblies for versatile task execution. Device 3, 100626 (2025).
16 Bonabeau, E., Dorigo, M. & Theraulaz, G. Inspiration for optimization from social insect behaviour. Nature 406, 39-42 (2000).
17 Heuthe, V.-L., Panizon, E., Gu, H. & Bechinger, C. Counterfactual rewards promote collective transport using individually controlled swarm microrobots. Science Robotics 9, eado5888 (2024).
18 Heins, C. et al. Collective behavior from surprise minimization. Proceedings of the National Academy of Sciences 121, e2320239121 (2024).
19 Garg, S., Shiragur, K., Gordon, D. M. & Charikar, M. Distributed algorithms from arboreal ants for the shortest path problem. Proceedings of the National Academy of Sciences 120, e2207959120 (2023).
20 Li, B. et al. Machine learning-enabled globally guaranteed evolutionary computation. Nature Machine Intelligence 5, 457-467 (2023).
21 Mateo, D., Horsevad, N., Hassani, V., Chamanbaz, M. & Bouffanais, R. Optimal network topology for responsive collective behavior. Science advances 5, eaau0999 (2019).
22 Tiwari, A., Devasia, S. & Riley, J. J. Low-distortion information propagation with noise suppression in swarm networks. Proceedings of the National Academy of Sciences 120, e2219948120 (2023).
23 Horsevad, N., Mateo, D., Kooij, R. E., Barrat, A. & Bouffanais, R. Transition from simple to complex contagion in collective decision-making. Nature communications 13, 1442 (2022).
24 Rosenthal, S. B., Twomey, C. R., Hartnett, A. T., Wu, H. S. & Couzin, I. D. Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion. Proceedings of the National Academy of Sciences 112, 4690-4695 (2015). 		AI for the Natural EnvironmentChengchen Zhang
Submodularity-based Computational Sustainability for Remotely Sensed DataOverview and General Goal:
Computational Sustainability (CS) seeks to balance environmental, economic, and social needs by developing optimization and machine learning methods that support sustainable decision-making. This project focuses on leveraging remote sensing data (satellite imagery and related datasets) to produce high-resolution, low-cost socioeconomic and environmental maps, with a particular emphasis on scalable algorithms rooted in submodularity and combinatorial optimization.

The broader goal is to transform the growing availability of satellite imagery (e.g., from Planet Labs, SkyBox, SpaceX Starlink) into actionable insights for sustainability domains such as poverty mapping, pollution monitoring, and transportation planning, where ground-truth data is scarce or expensive to collect.


Research Gap and Novelty
Prior work has shown that neural networks on satellite imagery can predict poverty indicators (e.g., Jean, Science, 2016; Yeh., Nature, 2020), and that remote sensing supports pollution mapping (e.g., Duncan et al., Environmental Science & Technology, 2014), the key challenges remain:

Scalability: Current methods rely heavily on deep learning with extensive ground-truth data. These models are computationally expensive and struggle to adapt across regions with limited labeled data.

Interpretability: Deep learning approaches often provide little transparency in feature selection, which is critical for policy adoption in sustainability contexts.

Efficient Use of Limited Ground Data: Ground-truth surveys (e.g., World Bank LSMS) are sparse, costly, and unevenly distributed. Methods to optimally select where and what data to collect remain underdeveloped.

Novelty of this Project: Introduce submodularity-based optimization into the remote sensing for sustainability pipeline. Submodular functions, with their property of diminishing returns, are well-suited for:

- Data subset selection (choosing the most informative satellite images or features to annotate/ground-truth).
- Sensor placement & survey design (optimizing where to deploy costly ground measurements).
- Interpretability (since submodular optimization often selects representative and diverse features, offering insights into spatial drivers of poverty or pollution).
- Integrate nonnegative matrix factorization (NMF) with submodular priors to generate interpretable latent factors from high-dimensional satellite data.
- Develop a low-cost pipeline that requires minimal ground data but achieves comparable or better predictive performance than purely deep learning-based approaches.
This combination of remote sensing + submodular optimization + interpretable factorization has not yet been systematically applied to computational sustainability, filling a research gap between “black-box” predictive models and actionable, resource-aware methods.

Project Objectives: Design scalable submodularity-based algorithms for
- Selecting informative subsets of satellite images for poverty/pollution estimation.
- Guiding sparse survey collection to complement imagery.
- Integrate remote sensing with interpretable optimization methods (NMF + graph-theoretic models) for socioeconomic mapping.
- Benchmark against state-of-the-art methods on poverty prediction (DHS, LSMS datasets) and air quality mapping.

Demonstrate applications in Smart Cities:
- Poverty and inequality mapping at neighborhood-level resolution.
- Traffic congestion monitoring using road network submodularity.
- Urban air pollution estimation.

Input and Output
Input: Multi-modal satellite imagery (optical, SAR, nightlights) + sparse socioeconomic surveys (LSMS, DHS) or pollution sensors.
Output: Scalable, high-resolution poverty/pollution/transportation maps that are interpretable, resource-efficient, and cost-effective.

Methods and Tools
Submodular Optimization: For data subset selection, sensor placement, and interpretable feature discovery (Bilmes, 2022).
Graph Theory: To model spatial and network interactions (e.g., traffic networks, spatial poverty clusters).
Nonnegative Matrix Factorization (NMF): To extract interpretable latent factors from high-dimensional imagery.
Remote Sensing Analysis: Handling multi-spectral and temporal satellite data streams.

References
Bilmes, (2022). Submodularity in Machine Learning and Artificial Intelligence. arXiv:2202.00132.
Jean,, et al. (2016). Combining satellite imagery and machine learning to predict poverty. Science, 353(6301), 790–794.
Yeh, et al. (2020). Using publicly available satellite imagery and deep learning to understand economic well-being in Africa. Nature Communications, 11(1).
Duncan, et al. (2014). Satellite data of atmospheric pollution for U.S. air quality applications. Environmental Science & Technology.
World Bank LSMS Initiative. https://www.worldbank.org/en/programs/lsms		AI for Sustainable Operations and Circular Economy, AI for the Natural Environment, AI for Transportation and LogisticsAndersen Ang
Revising the processing–microstructure–properties–performance (PMPP) paradigm to include sustainabilityMuch of the recent technology we enjoy has been made possible through materials design. Recent examples of it include ultra-light case aluminium alloys for the Apple Watch and re-entry rockets skin materials. Such ‘Materials by Design’ concept, is based on a systems-engineering analysis where the desired performance of components, such as a smart watch or a reusable rocket, are related to the processing schedules needed to produce and shape the material into the final component. Such processing is related to the material microstructure, ultimately influencing properties. Only by relating the product engineering performance with its underlying microstructure and properties, can we confidently ensure the component safety and reliability.

The processing – microstructure – properties – performance (PMPP) linkage necessitates computational models; but the full complexity of such links escapes the power of physical modelling. Additionally, the need to reduce emissions upon materials production, increase their circularity, and in general maximise their sustainability, imposes even more complex constraints to materials design. AI provides a suite of models that can accelerate handling large datasets where the complex PMPP interactions can be captured more reliably, expediting the development of novel sustainable materials.

This project will combine AI and physical models for sustainable materials design. Our focus will be powder-based alloys for additive manufacturing (3D printing). Five groups of alloys will be considered: steel, aluminium, titanium, nickel, and high-entropy alloys. Although each of those alloy families have their own metallurgy, a more or less general approach to tackle them is shown in Fig. 9 in https://www.mdpi.com/2071-1050/15/20/15081. Towards the end of the project we will have developed methods to maximise the sustainability through AI-powered alloy design for 3D printed aerospace components.


Research questions
• Link an alloy producer (Globus Metal Powders) and aerospace components manufacturer (GKN Aerospace) to develop a methodology to evolve metallic powder additive manufacturing technology towards greater sustainability.
• Demonstrate the methodology by applying it first to the development of additively manufactured sustainable nickel-based superalloys.

AI for Sustainable Operations and Circular EconomyPedro Rivera
AI-driven Design and Optimisation of Functional Metamaterials for Sustainable Energy SystemsMetamaterials—artificially structured media with tailored subwavelength features—are redefining the limits of electromagnetic and thermal control. Their ability to precisely tailor how light and heat move, interact, and are directed enables unprecedented performance in solar energy harvesting, thermophotovoltaics, passive radiative cooling, and thermal management in smart buildings. Yet, despite their transformative potential, the design of high-performance metamaterials remains an intractable challenge: the vast, high-dimensional parameter space and strong multi-physics coupling make traditional optimisation prohibitively slow and limited in scope.
In reality, modern lithography can produce essentially arbitrary 2D or 3D shapes, aperiodic arrangements, and multiscale combinations without added fabrication cost — yet this flexibility is rarely exploited because human-guided design cannot feasibly search such an immense, non-convex parameter space. The result is a performance ceiling defined not by physics or fabrication limits, but by the constraints of our design tools and workflows. This project is not just an acceleration of current methods: by combining high-fidelity experimental data with AI-enabled inverse design, it will unlock solutions beyond human intuition, expanding exploration from the narrow confines of conventional resonator arrays to the full diversity of patterns that lithography can offer.
The research will develop next-generation AI-driven inverse design frameworks for functional metamaterials in sustainable energy systems. Datasets will be generated using finite-difference time-domain (FDTD) full-wave electromagnetic simulations, calibrated against experimental measurements to ensure high fidelity and physical realism. These datasets will then be used to train advanced AI models, including deep generative architectures (e.g. conditional GANs), reinforcement learning agents, and physics-informed neural networks. The resulting framework will deliver orders-of-magnitude acceleration in the exploration and optimisation of complex metamaterial design spaces, uncovering device architectures and performance regimes that lie beyond the reach of traditional human-guided design.
The PhD project will aim to address three critical technical challenges:
1. Multi-physics, multi-objective optimisation — simultaneously maximising optical/thermal performance, structural robustness, and manufacturability, incorporating fabrication constraints directly into the optimisation loop.
2. Generalisation and robustness — developing AI models capable of extrapolating to unseen parameter regimes, incorporating uncertainty quantification to mitigate performance degradation under fabrication imperfections.
3. Design–fabrication–characterisation integration — establishing a closed-loop pipeline in which AI-generated designs are fabricated via scalable nanomanufacturing techniques (e.g., nanoimprint lithography, roll-to-roll processing), optically and thermally characterised, and iteratively refined via active learning.
The project sits at the convergence of computational photonics, multi-physics simulation, and machine learning, leveraging recent advances in AI for inverse design (Liu et al., Nature Photonics, 2021; Malkiel et al., Light: Science & Applications, 2019; So et al., Nature Computational Science, 2022) and physics-informed learning (Raissi et al., Journal of Computational Physics, 2019). It will generate both fundamental advances in AI-enhanced materials design methodology and high-impact prototypes for sustainable energy applications. To accelerate translation and maximise societal benefit, we will engage with industrial partners across different sectors, including Jaguar Land Rover , Malvern Optical and Pilkington, to co-create application contexts such as smart windows and energy-efficient building envelopes.		AI for Sustainable Energy and BuildingsRuomeng Huang
Designing Low-Power Digital Integrated Circuits for Machine Learning-Based Signal ProcessingIn various fields including healthcare, robotics, audio, and Internet of Things (IoT) devices, there is a significant surge in the need for high-throughput and ultra-low-power signal processing. Traditional DSP engines find it challenging to keep up, whereas compact AI models are ready to provide enhanced accuracy with a reduced code footprint, provided we can effectively implement them on silicon for remote locations, edge devices, and sustainable applications.

Thus, we have the opportunity to re-envision how small silicon components can execute robust machine-learning algorithms using only hundreds of microwatts to milliwatts. Become a part of the Lab of Efficient Machine Intelligence to develop cutting-edge digital integrated circuits that provide sustainable and on-chip intelligence.

The project has the following three objectives: (i) To architect and prototype network topologies tailored to large-scale deep learning driven signal processing algorithms. Novel topologies are supposed to provide load-aware distributed processing on the network, while enabling sustainable energy management. (ii) To implement and validate the operation of the introduced networks in (i) using FPGA. This step further helps translation of a FPGA design on silicon in a specific complementary metal-oxide-semiconductor (CMOS) technology (e.g. 180 nm). (iii) To benchmark energy, performance and accuracy against the state-of-the-arts. Our interest is to develop designs utilizing the steps (i) and (ii) for real-time and on-chip classification of biomedical signals, however, other applications can be identified. For example, we have already developed complex algorithms for brain-computer interfaces that can be translated into hardware using the proposed flow for low-power and sustainable uses.

This is an ideal project for a PhD student who enjoys problem solving and interdisciplinary research, involving network topologies, signal processing, machine learning, low-power hardware architectures and microelectronics design. The candidate will have the opportunity to learn many desirable knowledge and scientific skills during this PhD.

This is a 4-year integrated PhD (iPhD) programme and is part of the UKRI AI Centre for Doctoral Training in AI for Sustainability (SustAI). For more information about SustAI, please see: https://sustai.info/.

The candidate will have a chance to the training and networking opportunities offered through the CDT and to collaborate with a strong multidisciplinary team of academics at University of Southampton (UoS). There are also to state-of-the-art laboratory facilities from ECS (DHBE and SET) and Zepler institute. Applicant will be provided with access to MATLAB, FPGA board, Cadence software and fabrication facilities.
Sustainable AIMajid Zamani
Event-based computing for real-time computer vision systemsComputer vision systems (CVS) help machines see and understand the world around them, e.g. self-driving cars, robots, and security cameras that need to work instantly and efficiently. Today’s CVSs have trouble keeping up in fast-changing situations, which capture full pictures or videos at fixed times, like taking snapshots or video frames; meanwhile, most of the information in each picture doesn't change much [1]. Thus, these systems are inefficient, slow and use a lot of power. Event-based computing is a new way of doing this. Instead of capturing everything all the time, event-based systems only notice and respond to things that change—like movement or flashes of light—right when they happen. This means AI hardware can process much faster and use far less energy to recognise patterns (moving objects) [2]. Nevertheless, the CVS systems in the market today still rely on the expensive dynamic vision sensor (DVS) cameras. This research project will develop novel event-based computing chips and their AI-hardware implementation, enabling ultra-low powered and low-cost DVS-free CVS.

In the first year, you will build event-based computing chips based on emerging nanodevices, utilising our state-of-the-art cleanroom facility. You will learn nanofabrication (thin film engineering and lithography technique) to fabricate a wafer-scale massive array of memdiodes and surface/interface analysis (advanced microscopy and spectroscopy tools) to evaluate the quality of the chips.

In the second year, you will develop a protocol to encode and decode input signals generated by moving images. The student will learn electrical characterisation techniques to evaluate the response of the chips and simulate how these chips compute moving images.

Finally, in the third year, you will test the full system in real-world conditions—making sure it works reliably and more efficiently than today’s CVS. You will design and build the systems on FPGAs to process image data in real time (captured by a conventional camera) via an AI algorithm.

Ref:
1. Ji, M. et al. SCTN: Event-based object tracking with energy-efficient deep convolutional spiking neural networks. Front. Neurosci. 17, (2023) doi:10.3389/fnins.2023.1123698.
2. Xu, Y. et al. Optimizing event-based neural networks on digital neuromorphic architecture: a comprehensive design space exploration. Front. Neurosci. 18, (2024) doi:10.3389/fnins.2024.1335422.

Sustainable AIFirman Simanjuntak
Predicting the Skies for Perpetual FlightWe aim to work towards the dream of perpetual flight, UAVs that can stay aloft for weeks or even months without fuel as part of the fight against climate change. UAVs capable of harvesting atmospheric energy could transform environmental monitoring, disaster response, and global connectivity, while leaving a zero-carbon footprint. Why is it important? We can capture high resolution data that satellites can't. This data is essential for many applications, from searching for environmental data related to global warming to search and rescue during disasters.

This PhD will focus on developing predictive soaring technologies that allow UAVs to anticipate and exploit soaring conditions in a timely and efficient manner. A key part of the project will be to explore the most effective low-power sensing strategies for detecting these conditions, ranging from acoustic and pressure sensors to alternative novel methods. These sensors will then be combined with spatio-temporal machine learning and environmental propagation models to forecast temperature fields and turbulence at a distance.

The challenge is to design algorithms that are efficient enough to run onboard UAVs yet powerful enough to predict soaring opportunities in real time. Beyond UAVs, the AI techniques you will develop, such as spatio-temporal modelling, uncertainty-aware prediction, and sensor–model fusion, have broader applications in climate modelling, renewable energy forecasting, environmental monitoring, and autonomous robotics operating in complex dynamic environments. If you are excited by pushing the boundaries of AI for sustainability, this PhD offers the chance to apply your skills to one of aviation’s most ambitious frontiers.

We hypothesise that an actively emitted, low-power acoustic pulse (when processed with advanced signal conditioning and ML-based classification) can infer vertical wind patterns and energy gradients in the lower atmosphere. Such echolocation-inspired sensing, tuned for specific atmospheric frequency bands, can be embedded into small UAVs without the energy and weight penalties of conventional systems.

Similar to established Sonic Detection and Ranging (SODAR) systems [1], transmitted acoustic pulses are scattered by temperature and density fluctuations associated with air and turbulent eddies, with a fraction of this energy backscattered and received as echoes. The time delay of these echoes yields scatter height, while the Doppler shift provides radial wind velocity information [1]. Understanding the physics of acoustic scattering in high-altitude turbulence is central to the success of this project. If these processes are mischaracterised, the system could be optimised for the wrong frequency ranges, transmit powers, or signal structures, severely limiting its effectiveness. To mitigate this risk, we include a dedicated physics work package to quantify scattering coefficients, capture the effects of intermittency, and model stratospheric absorption under realistic conditions. This approach not only safeguards the performance of the proposed miniaturised SODAR on UAVs but also establishes a validated knowledge base that can benefit the broader atmospheric sensing community.

Recent studies on wave propagation in random, time-varying media show that nonlinear multiple-scattering dynamics can emerge, reinforcing the need for optimised excitation strategies that account for temporal disorder [2]. By carefully selecting frequency bands with minimal absorption and employing low-duty-cycle modulation to capture nonlinear multi-scattered responses, a low-power SODAR can maintain sensitivity comparable to conventional systems while using significantly less transmitted energy, making it lightweight, UAV compatible, and suitable for high-altitude operation. Operating from an airborne platform introduces an additional challenge, as propeller wash, structural vibrations, and unsteady aerodynamic loading raise the noise floor compared to stationary ground-based systems. To mitigate these effects, the system must be strategically positioned on the UAV to minimise exposure to self-generated noise, while the signal chain will incorporate advanced processing techniques such as adaptive noise suppression, pulse compression with modulated waveforms, and coherent averaging across multiple pulses. This represents one of the open research questions to be addressed in the study.
The supervisory team have used ML techniques [3], learning from highly nonlinear systems, we can design a system that works effectively with a minimal set of actuators and sensors. However their suitability for a highly nonlinear and unpredictable condition like Turbulence need to be checked and new models need to be built. To minimise energy consumption, the acoustic excitation mode and pulse modulation will be carefully selected and emitted at low duty cycles, using burst-modulation schemes that balance signal strength with power constraints. Pulse compression and digital signal processing techniques such as matched filtering and time-frequency transforms will be employed to enhance the signal-to-noise ratio of received echoes. The system may require the use of adaptive scheduling and beamforming methods to improve directional sensitivity and spatial resolution.

Specific Research Objectives/Outcomes

- Design a reliable, compact, energy-efficient acoustic actuator–sensor pair capable of atmospheric echolocation.
- Develop ML-based algorithms to classify return signals and infer vertical flow conditions.
- Validate echolocation sensing against reference probes in a controlled lab and outdoor environments.

The proposed system will consist of compact, multi-directional acoustic emitters paired with sensitive receivers capable of capturing atmospheric reflections from actively generated pulses. These units will be mounted on UAV platforms or test rigs and designed to operate in frequency bands optimal for atmospheric return detection, typically in the infrasonic or low-frequency audible range (20 Hz–2 kHz). The achievable sensing range will depend on both transmit power and excitation frequency, with the system designed to reach distances of over 500 m.

References
[1] M. A. Kallistratova, “Acoustic waves in the turbulent atmosphere: A review,” J. Atmos. Oceanic Technol., vol. 19, no. 8, pp. 1139–1150, 2002.
[2] R. Pierrat, J. Rocha, and R. Carminati, “Causality and instability in wave propagation in random time-varying media,” Phys. Rev. Lett., vol. 134, no. 23, p. 233801, 2025.
[3] L. Yule, N. Harris, M. Hill, and B. Zaghari, “An experimental study of machine-learning-driven temperature monitoring for printed circuit boards (PCBs) using ultrasonic guided waves,” NDT, vol. 3, no. 1, p. 1, 2025, doi: 10.3390/ndt3010001.
AI for Transportation and Logistics, Sustainable AIBahareh Zaghari
Balancing the benefits and risks of beaver rewilding in UK riversBeaver rewilding is expanding rapidly in the UK, yet there is limited predictive capacity to assess how their dam-building alters hydrology, biodiversity and ecosystem services across contrasting river types. This project will use field data, ecological modelling and AI methods to quantify and forecast the ecological and hydrological impacts of beaver reintroductions in different UK catchments (e.g. Bircham Valley, surface-fed urban beaver enclosure in Plymouth, and the River Meon, chalk stream rural beaver enclosure in Southampton).

Research challenge: Beavers provide biodiversity and hydrological benefits (e.g. flood attenuation, increased habitat complexity; Kemp et al. 2012, Brazier et al. 2020, Grudzinski et al. 2022), but may also create trade-offs (e.g. local flooding of farmland, sediment accumulation, reduced connectivity for some taxa; Auster et al. 2020, Dgebuadze et al. 2021, Wilson et al. 2025). Understanding when these changes tip from beneficial to detrimental is central to sustainable management.

The PhD will address three interlinked components:
(i) Data collection, with field surveys of biodiversity (macroinvertebrates, riparian vegetation, fish via eDNA), hydrological conditions (pond depth, water quality, sediment, temperature) across beaver-modified and control reaches (Antognazza et al. 2021). These will give the student a good understanding of what is living where and the characteristics of the environments associated with that.
(ii) Individual-based modelling to simulate ecological responses to habitat changes induced by beaver activity, focusing on key fish species (e.g. salmonids). These models uses species traits (e.g. growth, fecundity) and movement traits (e.g. probability of dispersal, behavioural reasoning for choosing a particular habitat) to map the distribution of species in dynamics landscapes (in this case, landscapes that are exposed to changes due to beaver activity through damming, channel changes, etc.; Dominguez Almela et al. 2020, 2022).
(iii) AI integration: Apply machine learning to detect thresholds, forecast ecological responses and develop a decision-support prototype for managers to evaluate rewilding trade-offs under different scenarios. Methods like Random Forest or supervised machine learning algorithms could be used to achieve this component (e.g. Cordier et al. 2017, Zong et al. 2024).

Team combines senior and early career expertise across two academic institutions and one practitioner: Vicky Dominguez and Paul Kemp (University of Southampton) will bring modelling and fish expertise respectively, Demetra Andreou (Bournemouth University) will add eDNA skills and Jane Ashford (Plymouth Council) will share her knowledge on managing beaver enclosures. The student will spend time embedded across all three institutions, which will strength their networks and expose them to different working environments.

References on beavers:
Kemp, P.S., Worthington, T.A., Langford, T.E.L., Tree, A.R.J. and Gaywood, M.J. (2012), Qualitative and quantitative effects of reintroduced beavers on stream fish. Fish and Fisheries, 13: 158-181.
Brazier RE, Puttock A, Graham HA, Auster RE, Davies KH, Brown CML. Beaver: Nature's ecosystem engineers. WIREs Water. (2021); 8:e1494.
Grudzinski, B.P., Fritz, K., Golden, H.E., Newcomer-Johnson, T.A., Rech, J.A., Levy, J., Fain, J., McCarty, J.L., Johnson, B., Vang, T.K. and Maurer, K., (2022). A global review of beaver dam impacts: Stream conservation implications across biomes. Global ecology and conservation, 37, p.e02163.
Auster, R. E., Barr, S., & Brazier, R. (2020). Alternative perspectives of the angling community on Eurasian beaver (Castor fiber) reintroduction in the River Otter Beaver Trial. Journal of Environmental Planning and Management, 64(7), 1252–1270.
Dgebuadze, Y.Y., Bashinskiy, I.V. & Osipov, V.V. (2021). The influence of Eurasian beaver Castor fiber activity on fish assemblages in small steppe rivers in Russia. Environ Biol Fish 104, 689–700.
Wilson, K., Bryce, J., Skilbeck, A., Needham, R., McAllister, S., and Campbell-Palmer, R. (2025). How to create woodlands that are resilient in the presence of beaver (Castor fiber): a review of current evidence. NatureScot Research Report 1368.

References on methods:
Antognazza CM, Britton RJ, Read DS, et al. Application of eDNA metabarcoding in a fragmented lowland river: Spatial and methodological comparison of fish species composition. Environmental DNA. 2021; 3: 458–471.
Dominguez Almela, V., Palmer, S.C.F., Gillingham, P.K. et al. (2020). Integrating an individual-based model with approximate Bayesian computation to predict the invasion of a freshwater fish provides insights into dispersal and range expansion dynamics. Biol Invasions 22, 1461–1480.
Dominguez Almela, V., Palmer, S. C. F., Andreou, D., Gillingham, P. K., Travis, J. M. J., & Britton, J. R. (2022). Predicting the influence of river network configuration, biological traits and habitat quality interactions on riverine fish invasions. Diversity and Distributions, 28, 257–270.
Cordier, T., Esling, P., Lejzerowicz, F., Visco, J., Ouadahi, A., Martins, C., Cedhagen, T. and Pawlowski, J., (2017). Predicting the ecological quality status of marine environments from eDNA metabarcoding data using supervised machine learning. Environmental science & technology, 51(16), pp.9118-9126.
Zong, S., Brantschen, J., Zhang, X., Albouy, C., Valentini, A., Zhang, H., Altermatt, F. and Pellissier, L. (2024), Combining environmental DNA with remote sensing variables to map fish species distributions along a large river. Remote Sens Ecol Conserv, 10: 220-235. 		AI for the Natural Environment, Sustainable AIVicky Dominguez Almela
Techno-economic modelling of sustainable data centre integration in the European energy systemData centres are rapidly becoming one of the most energy-intensive infrastructures in Europe. With growing demand for cloud services, AI, and digitalisation, their electricity consumption and cooling needs pose significant challenges to the energy system. At the same time, data centres can also provide flexibility to the grid through demand response, waste heat utilisation, and co-location with renewable energy sources.

This PhD project will further develop an optimisation model to analyse the sustainable integration of data centres into the European energy system. The project will investigate trade-offs between energy demand, system flexibility, emissions, and cost-effectiveness under different policy, market, and technology scenarios.

The successful candidate will:
• Further develop a long-term energy system model to capture the role of large-scale data centres in Europe’s net-zero transition.
• Analyse the interaction between data centres and renewable integration, including grid balancing, flexibility services, and storage.
• Explore sustainable cooling solutions and waste heat recovery for district heating and industrial use.
• Conduct scenario-based techno-economic assessments, evaluating the impact of data centre growth on investment and operational decisions in generation, transmission, and storage.
• Provide policy-relevant insights for European decarbonisation strategies.

The student will join the University of Southampton’s School of Mathematical Sciences, working at the intersection of optimisation, energy modelling, and sustainability analysis. There will be opportunities to collaborate internationally with leading research groups in energy systems.

We are seeking highly motivated candidates with a strong background in one or more of the following:
• Energy systems modelling and optimisation
• Techno-economic analysis or operations research
• Data science or applied mathematics
• Strong programming skills (Python, Julia, or similar)

The PhD will provide:
• Specialist training in energy systems modelling, techno-economic analysis, and sustainability assessment
• Opportunities to attend international conferences and workshops
• A structured four-year research programme, including the preparation of high-impact publications		AI for Sustainable Operations and Circular EconomyHongyu Zhang
Sustainable production line using Digital Twin The manufacturing sector is facing rising energy costs, creating a pressing need to reduce energy consumption while maintaining throughput. At the same time, technological solutions are required to support an aging workforce. One promising approach is to augment workers with digital twin (DT) technology. This project therefore proposes a human-centric digital twin framework to address these challenges.

The research focuses on two questions: (1) How can a digital twin be developed as an augmentation tool for manufacturing workers? (2) How can the effectiveness of the digital twin be evaluated? Onggo et al. (2021) outline key components of simulation-based digital twins relevant to this project. A simulation model of a production line will be developed using Python libraries such as SimPy or Salabim and integrated with an optimisation model to suggest optimal production sequences.

A major challenge of hybrid simulation–optimisation for real-time decision-making is computation time. Two approaches may address this: (i) simheuristics, combining simulation with fast heuristics (see Onggo et al., 2025), or (ii) multi-fidelity modelling (e.g., Cao et al., 2021; Rhodes-Leader et al., 2022), where a low-fidelity metamodel (such as a feedforward neural network) approximates a high-fidelity simulation.

The expected output is a digital twin capable of generating optimal production schedules and supporting workers through interactive what-if experimentation. To evaluate effectiveness, performance indicators will combine production metrics (e.g., throughput, rework rates) with human-centric metrics (e.g., situational awareness, digital stress). Few studies evaluate digital twins from this perspective; this research aims to fill that gap.

Note: Subject to NDA approval, the case study can be seen in the file attachment. The industry partner will provide the data needed for developing and validating the models.

References
Onggo BS, Martin XA, Corlu CG, Panadero J, Juan AA (2025). Stochastic Capacitated Dispersion Problems in Disaster Preparedness for Mass Casualty Incident. Journal of Heuristics 31:31. https://doi.org/10.1007/s10732-025-09566-1
Rhodes-Leader L, Nelson BL, Onggo BS, Worthington DJ (2022). A multi-fidelity modelling approach for airline disruption management using simulation. Journal of the Operational Research Society, 73(10):2228-2241. https://doi.org/10.1080/01605682.2021.1971574
Onggo BS, Corlu CG, Juan AA, Monks T, and de la Torre R (2021). Combining Symbiotic Simulation Systems with Enterprise Data Storage Systems for Real-Time Decision Making. Enterprise Information Systems, 15(2):230-247. https://doi.org/10.1080/17517575.2020.1777587
Cao Y-y, Currie CSM, Onggo BS, Higgins M (2021). Simulation Optimization for a Digital Twin using a Multi-Fidelity Framework. In Proceedings of the 2021 Winter Simulation Conference, pp. 1-12. 10.1109/WSC52266.2021.9715498
AI for Sustainable Operations and Circular EconomyStephan Onggo
Towards a Sustainable Global Food Supply through Geospatial Intelligence and Deep ReasoningGlobal food supply is increasingly unstable under rapid climate and socio-economic change. Even small rises in global temperature have been linked to major disruptions, including prolonged droughts, severe flooding, heatwaves, and soil degradation, all of which affect agricultural productivity (IPCC, 2023). Food security is further threatened by heavy reliance on imports. over 50% of staple crops in some countries, such as the UK, leaving them more vulnerable to climate extremes (FAO, 2023). Monitoring and forecasting agricultural health is therefore critical. This includes major crops (e.g., cereals, vegetables, fruits) and livestock systems, which also contribute substantially to greenhouse gas emissions. The growing availability of satellite Earth observation data – several petabytes generated annually – offers an unprecedented opportunity to observe these systems at scale. However, these multimodal datasets (imagery, hyperspectral data, radar, time series) are complex, making rapid and consistent insight extraction challenging.

This PhD project will apply & develop a physics-informed but low-cost AI framework to monitor and forecast global agricultural health. You will combine satellite remote sensing, field measurements, and process-based ecosystem models to investigate how crop and livestock management influence carbon storage, greenhouse gas fluxes, and food supply. Fieldwork will be based at Rothamsted Research’s North Wyke farm platform, with opportunities to collaborate internationally – including sites in Germany, the ICOS data centre in Sweden, United Unions, and agricultural monitoring sites in China. You will gain hands-on experience with eddy covariance flux towers, GPS surveys, and land surface monitoring. On the computational side, you will apply advanced AI methods, including computer vision and large language models, to extract information from hyperspectral and high-resolution satellite imagery. You will combine AI with process-based ecosystem modelling and data assimilation to improve carbon and nitrogen cycle simulations, scaling findings from local sites to regional and global levels. Beyond ecosystem processes, you will integrate climate projections to assess how agricultural production affects sustainability and food security under future scenarios.

This interdisciplinary project will give you cutting-edge skills in Earth observation, AI, and ecosystem science, preparing you for careers in research, policy, and industry. You will work with leading AI and environmental science groups at the University of Southampton and global academic & industry partners.
AI for the Natural Environment, Sustainable AISongyan Zhu
Statistically-guided Deep Active Learning for Sustainable Large Language Model TrainingLarge Language Models (LLMs) have shown to deliver ground breaking performance but at the expense of huge power costs, both during LLM pre-training (e.g. GPT-5 is rumoured to be trained on 25,000 GPU's for many days) and LLM fine-tuning (e.g. the modest LLaMa3 90B model fine tuning needs a minimum of 2 x A100 GPU cards and several days; these days 500B+ open source LLMs being released which are very GPU hungry). The upward curve of LLM GPU training resource requirements is not sustainable environmentally.

This inter-disciplinary PhD will explore combining statistical approaches from the discipline of Optimal Experimental Design with LLM deep active learning algorithms from the discipline of Computer Science. Deep active learning aims to use LLM patterns within data encodings and model output distributions to identify the best unlabelled training data instance which, if labelled and trained with, would train the LLM much more efficiently than a randomly chosen training data instance. OED uses statistical information stored in unlabelled data to choose the subsample to train the model on. OED is well studied for parametric models, while its use in LLMs has not been explored before. This is due to the need for the derivatives of the model with respect to model parameters. Our hypothesis is that even though LLMs are nonparametric, the necessary gradient information can be approximated using automated differentiation. This will identify the best subsamples to be used for non-linear deep active learning algorithms for LLM training cycles and ultimately train LLMs, given a target accuracy, with less training data & epochs compared to traditional methods. If successful this project has the potential to discover new ways to substantially reduce the GPU footprint of companies fine-tuning LLMs for downstream applications, and if adopted by a major LLM vendor reduce the massive LLM pre-training footprint.
Sustainable AIStuart Middleton
Machine learning for energy demand optimization in sub-Saharan Africa mini-grid systemsGlobally there are around 800 million people without access to electricity with around 600 million living in Sub Saharan Africa. The Energy for Development (e4D) programme at the University of Southampton was created in 2010 to address this challenge and initiated seminal studies in electricity access for hard-to-reach poor areas in Sub-Saharan Africa and beyond [1]. This included the design and construction of five solar PV based mini-grids with partners in Kenya and Uganda. The supervisory team continues to actively monitor and support the operation of these mini-grids. One key observation over the greater than ten years’ operation has been very different rates of growth in energy demand in different locations, despite broadly similar characteristics [2]. This clearly has implications for cost benefit analysis and system component lifetimes.
Within this programme, this project aims to further the understanding of electricity demand, its growth and optimisation within mini-grids in Kenya and Uganda. Specific challenges include:
- Estimation of electricity consumption profiles for households and businesses of different types based on measured datasets: these can be used to train models and construct synthetic datasets.
- Prediction of electricity demand growth over mini-grid lifetime: this is a key aspect of the cost-benefit analysis for mini-grid projects.
- Risk analysis of combined loads exceeding design thresholds: given the contained nature of mini-grids, this is critical for ensuring systems do not fail earlier than predicted.
- Optimal demand management in mini-grids under conditions of constrained generation and storage capacity: how to manage and control heavier loads to maximize availability and overall benefit to mini-grid users.

References
1. https://doi.org/10.3390/en12050778
2. https://doi.org/10.1109/jproc.2019.2924594		AI for Sustainable Energy and BuildingsAbuBakr Bahaj
Explainable and predicable building operation and maintenance optimisationThe urgent need to mitigate climate change has placed a spotlight on reducing energy demand within the building sector, which accounts for a significant portion of global energy consumption and carbon emissions. Traditional retrofit approaches often focus on individual technologies, lacking the integration necessary for optimal efficiency and unable to exploit the economies of scale that could improve the cost-effectiveness of retrofit interventions. This project addresses this gap by developing a multifunctional toolkit designed to facilitate large-scale building retrofit interventions that are cost-optimal with respect to investment, carbon reduction, and primary energy savings.

The proposed research aims to create a comprehensive toolkit that seamlessly integrates various energy efficiency measures and technologies. These include enhancements to building envelopes, installation of heat pumps, incorporation of photovoltaic systems, deployment of advanced energy management and control systems, and utilization of thermal and electric storage solutions. By combining these elements, the toolkit will enable a holistic approach to retrofitting that maximizes energy savings and minimizes carbon footprints across a wide array of buildings.

The research methodology entails the creation of hybrid models that incorporate physics, machine learning, and statistics. These models must be capable of managing the complexities associated with the integration of multiple technologies across a variety of building types. As part of the research activity extensive datasets will be analysed on building performance, occupant behaviour, and energy consumption patterns by leveraging the models that have been developed. The insights identified in the modelling and analysis process will inform the optimization process, allowing the toolkit to recommend tailored retrofit strategies that align with specific building characteristics and usage patterns.

To ensure practical applicability, the toolkit will be built and tested using large-scale project dataset from Provelio. This real-world validation will provide critical feedback for refining the toolkit and demonstrate its effectiveness in achieving energy and carbon reduction goals at scale, verified by means of building monitoring. This collaboration will facilitate the deployment of the toolkit and support the transition toward more sustainable building practices.		AI for Sustainable Energy and BuildingsStephanie Gauthier
Machine learning for optimization of energy supply in UK commercial buildingsThe UK’s commitment to net-zero emissions by 2050 has made renewable energy supply essential. Commercial property investors, tasked with retrofitting existing stock, are at the forefront of this change. Increasingly, Landlords view solar installations as a strategic investment with strong returns, driven by rising EPC requirements, green lease options, and the demand for sustainable energy. Unfortunately, rooftops represent to often a missed opportunity to harness solar radiation already present and generate profits.
This project aims to optimise the potential for commercial solar with financial returns. Landlords can install solar on their rooftops and sell power back to the occupants, either through a newly formed Power Purchase Agreement(PPA) or through existing service charge or energy recharge mechanisms under the lease. Excess energy generated that the occupant does not use can be sold back to the grid under a variety of different mechanisms.
The research methodology entails the creation of hybrid models that incorporate physics, machine learning, and statistics. These models must be capable of managing the complexities associated with the integration of solar system across a variety of building energy demand and climate across the UK. These models will be built and tested using large-scale project dataset from Absolar. 		AI for Sustainable Energy and BuildingsStephanie Gauthier
Resilient Resource Allocation in Dynamic Settings under Uncertainty (RRADSU)Allocation of scare resources is prevalent in our lives. From allocating drivers and vans for delivery of goods, to the allocation of beds and doctors in hospitals, to the allocation of paramedics and ambulances in disaster response. Take Emergency Services as an example. They play a critical role by dispatching vehicles and qualified professionals to emergency situations. While providing skilled and well-equipped staff in a short time is key in most scenarios, they operate with limited resources and hence it is crucial to ensure that right level of resources is used to deal with a situation, while maintaining the resilience of the system for upcoming emergencies.

An efficient allocation of resources makes the best use of the limited resources available and consequently enhances the system’s sustainability. In many applications, it is important for the users to perceive the allocation as ‘’fair’’ for the allocation mechanism to be deployed and sustained successfully. Efficiency and fairness have been studied extensively. However, most studies make assumptions that do not hold in many real-life scenarios. An allocation is resilient if, should a problem occur (e.g. an ambulance breaks down and becomes unavailable), it can be amended with minimal loss to efficiency and fairness. Very few studies address resilience in limited settings.

We aim to address the existing limitations of literature by considering settings that are dynamic and incorporate uncertainty.

(1) Dynamic: The set of resources and tasks/agents (e.g. ambulances and patients) change over time. Allocation of resources are temporal (e.g. an ambulance is allocated to a patient for a length of time and will be available again after the patient reaches the hospital). Resources may have spatial attributes and their location plays a role in how an allocation should be made and how to evaluate the quality of an allocation.

(2) Partial and dynamic preferences: Agents’ preferences over available resources, and resources’ priorities over agents, may not be known to the system designer or even to the agents and resources themselves. There could be privacy or security concerns which raise the need to design systems that can produce good results without access to full preferences. Preferences can change over time when more information becomes available or the status of the agent changes. For example, a patient in an emergency may be happy to go to any hospital with an available bed, but when the situation stabilises, they prefer to stay in a hospital closer to home.

The main goal of the project is to use multi-agent systems and machine learning techniques to design fair, efficient and resilient allocation of scarce, and possibly heavily constrained, resources in dynamic settings with partial preferences. Our research objective could include (1) formalising what constitutes a resilient mobility/logistics system, (2) analysing the trade-off between resilience and efficiency and fairness, (3) investigating the resilient, efficient and fair allocation of scare resources, (4) utilising agent-based simulations to evaluate various coordination and resource allocation mechanisms or policies. The specifics of the project can be adapted to the PhD candidate's skills and interests.

Relevant Publications

Akbarpour, Li and Oveis Gharan, "Thickness and information in dynamic matching markets", J. Political Econ, 2019.

Garg and Murhekar: "On fair and efficient allocations of indivisible goods", AAAI, 2021

Zeng and Psomas, "Fairness-Efficiency Tradeoffs in Dynamic Fair Division", EC, 2020

Lodi, Olivier, et al. "Fairness over time in dynamic resource allocation with an application in healthcare". arXiv, 2022

Yang, R., Ford, B. J., Tambe, M., & Lemieux, A. (2014, May). Adaptive resource allocation for wildlife protection against illegal poachers. In AAAMAS (pp. 453-460).
AI for Transportation and LogisticsBahar Rastegari
Neurosymbolic Machine Learning for Distributed Fibre Optic Sensing (DFOS)This PhD project focuses on leveraging Neurosymbolic Machine Learning (ML) to drive advancements in Sustainable AI, particularly by optimising the efficiency of AI systems to reduce energy consumption. Neurosymbolic ML, which combines the interpretability of symbolic AI with the flexibility of subsymbolic techniques like deep learning, is a promising approach to achieving sustainability in AI. Symbolic reasoning, with lower computational demands, can significantly reduce the resource consumption of AI models, making them more eco-friendly without sacrificing accuracy or adaptability.

The focus of this PhD project will be on using neurosymbolic ML methods to analyse Distributed Fibre Optic Sensing (DFOS) data. DFOS technology is capable of providing highly granular data on environmental vibrations, structural integrity, and other signals critical for monitoring urban infrastructure and natural environments. One of the key challenges of this technology is the vast density of data, which needs massive storage solutions, as well as computational power to analyse it. The student will explore whether neurosymbolic methods can be leveraged to reduce the data storage and computational resources necessary to work with this type of data.

This research will be tightly aligned with several research projects across the University and the National Oceanography Centre (NOC), using DFOS collected in different environments. In particular, the student will benefit from involvement in a new grant which started in January 2025 which will collect DFOS data in the cities of Southampton and London, getting prime access to a novel and one-of-a-kind dataset. This grant will also count with collaborators from different fields, from social sciences to humanities, to collectively analyse the impacts of DFOS data in urban settings. Students with interest in impact and interdisciplinary research will find opportunities for career progression in these areas. Moreover, the student will also have access to data held at NOC, collected through different campaigns in marine environments, which could be applied for research on AI and environment. The availability of these diverse data sources will enable the investigation of noise-tolerant sustainable ML models capable of operating in multiparameter environments, such as those influenced by urban infrastructure or marine ecosystems.

The main research challenge lies in developing event detection systems that can operate effectively in complex, noisy environments while maintaining low energy consumption. Traditional deep learning methods, though powerful, require significant computational resources, especially when applied to large datasets like those generated by DFOS platforms. Neurosymbolic ML, by contrast, offers a more computationally efficient approach, enabling the development of models that can detect and classify events with reduced energy demands.

You will be supervised by a team of interdisciplinary researchers in machine learning, signal processing and distributed systems, and will have the opportunity to collaborate with industry partners to further your research. You will join the School of Electronics and Computer Science in collaboration with the National Oceanography Centre (NOC), and will have professional development opportunities through the Alan Turing Institute, the UKRI TAS Hub, Responsible AI UK (Rai UK) and access to Future Worlds to explore commercialisation for your research.

While the primary theme of this project is Sustainable AI, it also has broader implications for other SustAI CDT themes. DFOS data can be applied to sectors such as Transportation and Logistics, where real-time monitoring and optimisation can reduce resource consumption and enhance sustainability. Moreover, it could be applied for environmental research, to assess the effects of climate change in coastal cities, detection of extreme weather events, and so on.		Sustainable AI, AI for the Natural Environment, AI for Transportation and LogisticsRafael Mestre
Optoelectronic AI processors for Cryogenic and Space ElectronicsCryogenic processors are gaining interest due to the rising demand for cryogenic and space electronics. However, today’s processors are based on CMOS technology that is prone to freeze attack and magnetic interference at low temperatures. In this project, you will invent a novel building block of processors based on memristor technology to achieve error-tolerant AI chips. Memristor is an emerging memory technology that can mimic a biological neural network, rendering low-powered neuromorphic computing. Memristor can also be programmed with optical stimulation, thus called optomemristor, demonstrating its unique feasibility to be integrated with photonics for ultra-fast operation. Von Ardenne GmbH, Germany (VA) will be the industrial partner of this project, and you will have the opportunity to visit VA facilities and collaborate with VA engineers who can provide support on materials processing and characterisations.

In the 1st year, you will build your first optomemristor chip prototype utilising Southampton’s and Von Ardenne’s state-of-the-art cleanroom facility. You will learn nanofabrication (thin film engineering and lithography technique) to fabricate a wafer-scale massive array of optomemristor and surface/interface analysis (advanced microscopy and spectroscopy tools) to evaluate the quality of your prototype.

In the 2nd year, you will develop a protocol to encode data via laser pulses under a magnetic field at cryogenic temperatures. You will learn electrical characterisation techniques to evaluate the synaptic capability and performance of your prototype.

In the 3rd year, you will design a neural network based on the experimental results and evaluate the computational accuracy of your chip. You will learn how to inject noises into the algorithm to emulate the interference induced by magnetic field and cryogenic temperatures, and develop error correction mitigation strategies.
Sustainable AIFirman Simanjuntak
Sustainable architectures and fast training approaches for large language modelsRecently, large language models (LLMs), such as OpenAI’s ChatGPT, have caught enormous attention of the public. They can generate remarkably realistic, coherent text based on a user's input and have the potential to be general-purpose tools used throughout society, e.g. for customer service, summarizing texts, answering questions, code generation and solving math problems. These LLMs are typically built on Transformer architectures with tens of billions of parameters and trained on trillions of tokens.

However, the large model size and high computational complexity of LLMs results in significant computational power and storage requirements far surpassing what is currently available with standard consumer hardware. For instance, even the relatively modest-sized model LLaMA-65B needs 130GB of GPU RAM for inference and more than 780 GB for fine-tuning with new data. Therefore, the aim of this PhD project is to develop sustainable architectures and fast training approaches for LLMs such that they can be accessible and deployed, given limited computational resources. This will promote sustainable language model usage in more scenarios in terms of power and hardware consumption.

In this project, the student will explore various pathways to achieve sustainable LLM architecture and training, including
1) Developing alternative architectures to Transformers to reduce the size of language models, through model compression or fundamentally new networks;
2) Data selection during fine-tuning, where selection mechanisms of data will be developed instead of using entire training dataset.

Students will work in a collaborative team with both experts in machine learning and natural language processing involved. We strongly suggest that students with strong motivation dive into the research of sustainable large language models to make GPT-type models accessible even with limited computational resources.
Sustainable AIZhanxing Zhu
AI-Enabled Sustainability Reporting and Assurance for the Circular Economy: Aligning Real-Time Supply Chain Data with Global Sustainability Reporting Frameworks.The recent global pressure on organisations to accelerate the transition towards a circular economy reflects an urgent need to decouple economic growth from resource depletion and environmental harm (Kirchherr, Reike & Hekkert, 2018; Geissdoerfer et al., 2017). This shift places intense demand for timely, accurate, and credible information on material and product flows, as well as their associated environmental impacts, across global supply chains (Opferkuch et al., 2022).

However, existing sustainability reporting, whether aligned with the GRI, the IFRS (S1/S2) or the CSRD/ESRS, remains essentially periodic and backwards-looking (Adams & Abhayawansa, 2022; Bebbington & Unerman, 2020). As such, most firms continue to compile and release annual or semi-annual disclosures based on manually aggregated data, creating a significant time lag between operational reality and reporting. This time lag limits the decision-usefulness of sustainability reports and may create space for greenwashing and selective disclosure (Cho et al., 2015; Pizzi et al., 2024).

Meanwhile, advances in artificial intelligence (AI), and the Internet of Things (IoT) technologies create new opportunities to close this gap. While many studies focus on creating models, frameworks, approaches, and solutions (Lee et al., 2012; Ma et al., 2018), few test their usability, application, or generalisability. This gap could be addressed by using real-world data to evaluate the proposed items. For instance, AI techniques such as artificial neural network (ANN), product -life cycle management (PLM), with multi-agent systems (MASs) and machine learning algorithms can capture, analyse and forecast real-time supply-chain data on emissions, energy use, material loops and product life cycles. ANNs models can detect anomalies, forecast resource needs, and map circular material flows, while blockchain and other distributed ledger technologies can help provide time-stamped evidence trails (Li et al., 2024).

Yet, the accounting and assurance implications of embedding such high-frequency machine generated data within recognised sustainability reporting frameworks remain underexplored (Quattrone, 2022). Traditional audit and assurance methods, designed for low-volume, human-recorded transactions, are ill-equipped to cope with the variety and veracity of AI-driven data streams (Li et al., 2024). This creates a critical gap at the intersection of sustainability reporting, accounting theory and digital assurance which this project aims to address.

This PhD project will thus address that gap by designing and testing an AI-enabled sustainability reporting and assurance model that aligns real-time supply chain data with global reporting frameworks (GRI, IFRS S1/S2, CSRD/ESRS) using innovative methodologies and interdisciplinary approaches. The student will work in an interdisciplinary environment, supervised jointly by AI specialists from the computing department (Dr Vahid Yazdanpanah) and sustainability accounting and reporting experts from the Centre of Excellence in Sustainability Accounting and Reporting (CESAR) at the University of Southampton Business School (Dr Ishmael Tingbani and Dr Renata Konadu). It will aim to recruit an exceptional candidate with the technical ability and skills needed to deliver the project as scheduled.
Research objectives:
1. To design and develop an AI-enabled framework for sustainability reporting and assurance that can recognise and determine reporting and compliance anomalies, process, track and disclose real-time supply-chain data in compliance with global standards such as GRI, IFRS S2, and CSRD/ESRS.

2. Develop and evaluate AI-driven assurance protocols for sustainability reporting – (Create and trial new audit procedures, including automated exception testing and blockchain-secured evidence trails to verify high-frequency AI data and strengthen stakeholder trust in sustainability disclosures).

3. To empirically investigate, via multi-case empirical research with SMEs across sectors and supply-chain positions, how AI-enabled data capture (IoT), analytics (ML/ANN) and tamper-evident evidence (e.g., blockchain) change SMEs’ sustainability reporting and assurance capabilities.


Key References
Adams, C. A., & Abhayawansa, S. (2022). Connecting the COVID-19 pandemic, environmental, social and governance (ESG) investing and calls for ‘harmonisation’of sustainability reporting. Critical perspectives on accounting, 82, 102309.
Bebbington, J., & Unerman, J. (2020). Advancing research into accounting and the UN sustainable development goals. Accounting, Auditing & Accountability Journal, 33(7), 1657-1670.
Cho, C. H., Laine, M., Roberts, R. W., & Rodrigue, M. (2015). Organized hypocrisy, organizational façades, and sustainability reporting. Accounting, organizations and society, 40, 78-94.
Geissdoerfer, M., Savaget, P., Bocken, N. M., & Hultink, E. J. (2017). The Circular Economy–A new sustainability paradigm?. Journal of cleaner production, 143, 757-768.
Kirchherr, J., Reike, D., & Hekkert, M. (2017). Conceptualizing the circular economy: An analysis of 114 definitions. Resources, conservation and recycling, 127, 221-232.
Lee, W.-I., Shih, B.-Y., & Chen, C.-Y. (2012). Retracted: A hybrid artificial intelligence sales-forecasting system in the convenience store industry. Human Factors and Ergonomics in Manufacturing & Service Industries, 22, 188–196. https://doi.org/ 10.1002/hfm.20272
Li, N., Kim, M., Dai, J., & Vasarhelyi, M. A. (2024). Using artificial intelligence in ESG assurance. Journal of Emerging Technologies in Accounting, 21(2), 83-99.
Ma, Z., Leung, J. Y., & Zanon, S. (2018). Integration of artificial intelligence and
production data analysis for shale heterogeneity characterization in steam-assisted gravity-drainage reservoirs. Journal of Petroleum Science and Engineering, 163, 139–155. https://doi.org/10.1016/j.petrol.2017.12.046
Opferkuch, K., Caeiro, S., Salomone, R., & Ramos, T. B. (2022). Circular economy disclosure in corporate sustainability reports: The case of European companies in sustainability rankings. Sustainable Production and Consumption, 32, 436-456.
Pizzi, S., Venturelli, A., & Caputo, F. (2024). Restoring trust in sustainability reporting: the enabling role of the external assurance. Current Opinion in Environmental Sustainability, 68, 101437.
Quattrone, P. (2022). Seeking transparency makes one blind: how to rethink disclosure, account for nature and make corporations sustainable. Accounting, Auditing & Accountability Journal, 35(2), 547-566.		AI for Sustainable Operations and Circular EconomyIshmael Tingbani
Predictive Thermal Management for Electric Vehicle Battery PacksThe performance, longevity, and safety of lithium-ion batteries, the cornerstone of electric vehicles, are critically dependent on maintaining a narrow optimal operating temperature range. Conventional Battery Thermal Management Systems (BTMS), typically based on complex liquid cooling, are reactive and consume significant parasitic energy, reducing overall system efficiency. This research project shifts thermal management from reactive to predictive to provide a more efficient and reliable solution.

This project will investigate a novel BTMS architecture that pairs solid-state thermoelectric Peltier elements with an advanced artificial intelligence (AI) control system. Peltier elements offer significant advantages, including high reliability, silent operation, and precise, bidirectional (heating and cooling) temperature control. Critically, they also offer rapid response rates compared to traditional fluid-based cooling systems.

The core of this project is the development of a hybrid AI controller designed to minimise the parasitic energy consumption of the Peltier-based BTMS, while also giving more accurate, real-time, thermal control of the battery cells. This controller will consist of two synergistic components:
(1) Predictive, to forecast the battery's future thermal load based on real-time operational (e.g. throttle position and recent usage) and environmental data.
(2) Decision, to proactively control the Peltier element. By using the predictive algorithms’ anticipated thermal needs, the decision algorithm will operate the Peltier element in its most efficient profile to pre-emptively manage temperature fluctuations of the battery cells.

The methodology involves developing a high-fidelity, multi-physics simulation environment in MATLAB/Simulink to model the coupled electrochemical-thermal dynamics of the battery cells and the thermoelectric properties of the Peltier element. This virtual testbed will be used to train and validate the AI controller across a range of dynamic vehicle drive cycles and environmental conditions. The performance of the final AI-driven system will be experimentally verified using test packs, battery analysers and environmental chambers within the Energy Technology Research group.

The anticipated contribution of this research is a validated framework for an intelligent, energy-efficient, and highly reliable solid-state BTMS. By demonstrating the viability of AI-controlled thermoelectric devices, this work aims to advance battery management technology, leading to safer, longer-lasting, and more efficient electrical energy storage systems.		AI for Transportation and Logistics, AI for Sustainable Operations and Circular EconomyRichard Wills
Truth or Greenwash? Strategic Disclosure and AI Auditing in Sustainable OperationsSustainability has become a central concern across industries, but greenwashing—the misrepresentation or exaggeration of sustainability achievements—remains widespread. Organizations greenwash because genuine improvements are costly, sustainability outcomes are difficult to measure, and audiences reward “green” claims that are hard to independently verify. Under such information asymmetry, overstating achievements can temporarily boost reputation and market share, making greenwashing a rational but socially harmful strategy.

Attitudes toward greenwashing are also heterogeneous. Some actors benefit privately from overstated claims, while others bear reputational or financial risks if misconduct is exposed. This divergence creates persistent tension: without credible auditing, sustainability disclosure often reflects strategic communication rather than verifiable performance.

The emergence of AI-powered auditing and verification technologies offers new opportunities to address this challenge. Advances in machine learning, natural language processing, and remote sensing enable the analysis of large, diverse datasets—from corporate reports and media content to satellite imagery and sensor data—to detect inconsistencies between claims and observable reality. AI systems can improve the credibility of sustainability information, but adoption is not straightforward. Organizations must weigh the benefits of credibility and trust against costs, implementation challenges, and the possibility of exposing their own shortcomings.

These dynamics highlight the need for research that integrates computational approaches to detection with models of strategic incentives. Building on the greenwashing literature (e.g., Zhang, 2024) and work on cooperative responsibility under externalities (Fang & Cho, 2020), this project will investigate how AI can reshape the landscape of sustainability disclosure. Specifically, we will ask: Under what conditions will organizations voluntarily adopt AI-based auditing and verification tools? To what extent can AI create credible separation between genuine and misleading claims, or do strategic behaviors persist that undermine its effectiveness? How do shared platforms, certification consortia, or cooperative mechanisms affect incentives to adopt AI-based auditing? Which forms of policy intervention best align AI adoption with broader social welfare objectives?

Methodologically, we will combine game-theoretic and analytical modeling of strategic behavior with computational experiments using AI tools for sustainability auditing. This dual approach will allow us to examine both the incentives surrounding adoption and the technical feasibility of detecting greenwashing across different contexts.

Fang, X., & Cho, S. H. (2020). Cooperative approaches to managing social responsibility in a market with externalities. Manufacturing & Service Operations Management, 22(6), 1215-1233.

Zhang, D. (2024). The pathway to curb greenwashing in sustainable growth: The role of artificial intelligence. Energy Economics, 133, 107562.

AI for Sustainable Operations and Circular EconomyFangsheng Ge
Fairness in AI for TransportAI is now used for forecasting, pricing, routing, and operations across transport and logistics. It supports tasks such as demand prediction, fleet allocation, dynamic pricing, and real-time control. While these applications can improve efficiency, they can also reproduce or amplify existing inequalities, especially when algorithms learn from data sets that embed underlying biases [1,2]. Data bias feeding into algorithmic bias can, for instance, manifest in unequal bike-rebalancing service provision when models over-serve central or affluent areas while neglecting peripheral or low-income districts [3]. Other examples include hidden distributional effects in simulations and digital twins that rely on synthetic populations lacking key socio-demographic detail (for example, gender, ethnicity, disability), thereby masking inequities [4]. Transport is a domain already fraught with bias along gender, income, ethnicity, education, ability, and age dimensions, among others (see, for example, [5-7]). It is critical that any newly introduced technologies, including AI, tackle rather than perpetuate or worsen these inequities.

In transport research, algorithms—including optimisation and machine learning—are often treated as neutral. A key question, therefore, is whether fairness criteria are considered and, if not, how to incorporate them alongside predictive accuracy, efficiency, and reliability. There is strong policy demand [8] for cities, regulators, and operators to demonstrate how inclusive AI-enabled mobility is and whether, and how, AI can be trusted. Yet practical guidance on designing and evaluating fair transport algorithms remains scarce. This PhD will address that gap by integrating fairness directly into the design and operation of algorithms, aiming to ensure that AI systems are not biased or discriminatory towards particular groups.

The aim is to develop and validate methods to diagnose and reduce unfairness in AI-driven transport while maintaining acceptable operational performance. Specifically, it will explore:
• To what extent existing allocation, pricing, or other AI-enabled methods reproduce or amplify spatial and socio-demographic inequities.
• Which fairness definitions (equality, equity, need-based) are most appropriate for transport decision-making, and under what conditions.
• How fairness-constrained optimisation, machine learning, or other methods can be designed and tested to preserve acceptable efficiency.
• How different sociodemographic groups perceive fairness, and how those perceptions influence trust and adoption.

One possible direction is to create and test a fairness-aware framework for allocating shared mobility services. The framework would measure disparities, select a context-appropriate fairness definition, and embed it in optimisation as explicit constraints, demonstrating the resulting efficiency trade-offs. Other directions include a focus on fairness in surge pricing in ride-hailing (for example, Uber), routing in journey planners or demand-responsive buses, sensors in autonomous vehicles and traffic systems, or the design of simulation models and digital twins.

The project takes an interdisciplinary approach, combining operations research, computer science, and transport psychology. It links algorithmic fairness (detecting and mitigating bias in models) and fairness of outcomes (setting equity goals and assessing efficiency–equity trade-offs) with perceived fairness and acceptance (gathering users’ views).

References:
[1] Fermanian, J-D., Xidonas, P. & Corrente, S. (2025) Machine learning and fairness: an integrated multicriteria approach for the evaluation of supervised classifiers. J. Oper. Res. Soc. 0:1–13. https://doi.org/10.1080/01605682.2025.2554749
[2] Zhang, X., Ke, Q. & Zhao, X. (2024) Travel demand forecasting: a fair AI approach. IEEE Trans. Intell. Transp. Syst. 25:14611–14627. https://doi.org/10.1109/TITS.2024.3395061
[3] Villa-Zapata, L.M., Rodriguez-Roman, D., Flórez-Coronel, J.E., González-López, J.M. & Figueroa-Medina, A.M. (2024) Incorporating equity in the vehicle rebalancing operations of dockless micromobility services. Lat. Am. Transp. Stud. 2:100009. https://doi.org/10.1016/j.latran.2024.100009
[4] Nag, D., Brandel-Tanis, F., Pramestri, Z.A., Pitera, K. & Frøyen, Y.K. (2025) Exploring digital twins for transport planning: a review. Eur. Transp. Res. Rev. 17:15. https://doi.org/10.1186/s12544-025-00713-0
[5] Suel, E., Lynch, C., Wood, M., Murat, T., Casey, G. & Dennett, A. (2024) Measuring transport-associated urban inequalities: where are we and where do we go from here? Transp. Rev. 44(6):1235–1257. https://doi.org/10.1080/01441647.2024.2389800
[6] Parnell, K.J., Pope, K.A., Hart, S., Sturgess, E., Hayward, R., Leonard, P. & Madeira-Revell, K. (2022) ‘It’s a man’s world’: a gender-equitable scoping review of gender, transportation, and work. Ergonomics 65(11):1537–1553. https://doi.org/10.1080/00140139.2022.2070662
[7] Transport Committee (2025) Access denied: rights versus reality in disabled people’s access to transport. UK Parliament. Available at: https://committees.parliament.uk/publications/48336/documents/252972/default/ (accessed 25.09.2025)
[8] Department for Transport (2025) Transport Artificial Intelligence Action Plan: Transforming Ambitions. London: Department for Transport. Available at: https://assets.publishing.service.gov.uk/media/68519a48510376e43ffdb344/transport-ai-action-plan.pdf (accessed 23.09.2025)		AI for Transportation and LogisticsAgnieszka Stefaniec
Using AI to Improve Sustainability and Efficiency in Maritime Integration TestingThe maritime defence sector faces growing challenges in achieving sustainability and operational efficiency, particularly in integration testing for new technologies onboard future ships. Maritime integration testing verifies that various ship or offshore platform components function together as intended, ensuring functionality, performance, reliability, compatibility, and safety under different scenarios. Energy security, climate resilience, and emissions reduction are increasingly critical concerns (A. Gonzalez. et al. 2008). Artificial Intelligence (AI) presents a promising avenue for addressing these challenges by improving energy management, optimizing test simulations, and automating analysis to enhance sustainability and efficiency (Munim, Z. H. et al. 2020).

Research Problem and Questions

This project investigates how AI can enhance maritime integration testing by addressing key sustainability and efficiency concerns. The core research problem is: How can AI-driven approaches improve energy security, climate resilience, and emissions reduction while optimizing the efficiency of maritime integration testing?

Methodology

The research will employ a combination of AI methodologies, data analytics, and simulation-based approaches. The methodology includes:
Machine Learning & Predictive Analytics: AI algorithms will be used to model energy consumption and predict system performance, helping to improve energy efficiency and security (Gavin Yeo, et al. 2019).
Reinforcement Learning: Optimization techniques will be applied to enhance the efficiency of test procedures, reducing the resource intensity of simulations.
AI-driven Climate Risk Modeling: Machine learning models will assess climate resilience by analyzing historical and real-time environmental data.
Data Mining & Automated Analysis: AI will be leveraged to enhance emissions monitoring, optimize sustainability strategies, and support decision-making processes in maritime testing (Maria Fleischer Fauske, et al. 2019).
Data-driven Simulation: Virtual models of integration testing environments will be developed to evaluate AI-driven improvements before real-world implementation (Christoph H. Loch, et al. 2001).

Reference
A. Gonzalez, E. Piel, H. -G. Gross and M. Glandrup, "Testing Challenges of Maritime Safety and Security Systems-of-Systems," Testing: Academic & Industrial Conference - Practice and Research Techniques (taic part 2008), Windsor, UK, 2008, pp. 35-39, https://doi.org/ 10.1109/TAIC-PART.2008.14
Munim, Z. H., Dushenko, M., Jimenez, V . J., Shakil, M. H., & Imset, M. (2020). Big data and artificial intelligence in the maritime industry: a bibliometric review and future research directions. Maritime Policy & Management, 47(5), 577–597. https://doi.org/10.1080/03088839.2020.1788731
Gavin Yeo, Shiau Hong Lim, Laura Wynter, Hifaz Hassan (2019) MPA-IBM Project SAFER: Sense-Making Analytics for Maritime Event Recognition. INFORMS Journal on Applied Analytics 49(4):269-280. https://doi.org/10.1287/inte.2019.0997
Maria Fleischer Fauske, Carlo Mannino, Paolo Ventura (2019) Generalized Periodic Vehicle Routing and Maritime Surveillance. Transportation Science 54(1):164-183. https://doi.org/10.1287/trsc.2019.0899
Christoph H. Loch, Christian Terwiesch, Stefan Thomke, (2001) Parallel and Sequential Testing of Design Alternatives. Management Science 47(5):663-678. https://doi.org/10.1287/mnsc.47.5.663.10480		AI for Transportation and Logistics, AI for Sustainable Energy and BuildingsHuan Yu
Improving Instruction-Level Parallelism in ManycoresGeneral-purpose processors (CPUs) with hundreds of cores are expected to eventually take over the computing industry from embedded to server processors. This is due to their increased energy efficiency for naturally parallel problems such as AI, as well as their simplicity with which this is achieved. As opposed to specialised accelerators that can achieve better efficiency for more ephemeral workloads, manycore CPUs excel in critical aspects including generality and easy programmability.

Modern CPUs include various prediction mechanisms in hardware to increase their performance, while remaining abstracted from software. These include the cache replacement policies, data prefetchers and other predictors. The goal of such mechanisms is to minimise inefficiencies such the observed memory access latency to execute as many instructions as possible in a fixed time frame. Since multiple instructions can be executed per CPU cycle, improving these mechanisms contribute to instruction-level parallelism (ILP).

While these are thoroughly studied areas in computer architecture research, most advancements have been conducted in the context of single-thread performance. While they have proven effective for CPUs with multiple cores, there are multiple research gaps when it comes to approaching optimal performance. Furthermore, the increase in computing capacity has been growing faster than what the main memory technology can support. Thus, additional effort is needed to make the best of the available memory bandwidth that feeds these cores with data.

In this project, novel ILP prediction mechanisms will be developed and tailored for manycore processors. A design space exploration will be conducted to identify the best behaving design attributes of the corresponding algorithms. An analytical approach will be of particular importance in this project, in order to establish the theoretical foundations for impactful findings on future manycores. The resulting hardware algorithms will improve ILP for highly-demanding workloads of today and tomorrow including AI.		Sustainable AIPhilippos Papaphilippou
Improving the Sustainability of Non-Emergency Hospital TransportWe consider the problem of transport to and from hospitals and other major healthcare facilities for staff, visitors and non-emergency patients. Healthcare facilities employ and serve a huge number of people every day. For example, UHS employs over 13,000 staff members, offers over 1,500 outpatient appointments per day, has over 1,300 beds and sees approximately 400 patients per day in its emergency department. This adds up to a huge number of people travelling to and from the hospital each day.

Increasing the proportion of patients, visitors and staff who arrive at a hospital by shared transport would reduce transport emissions significantly and could improve the hospital environment by reducing the traffic in and around the hospital grounds. Currently patients with mobility needs are brought to the hospital using non-emergency hospital transport. This incurs an expense to the healthcare authorities and can be slow and inefficient, with patients sometimes having to wait a long while at the hospital before and/or after their appointments.
In this project, we will investigate the use of shared transport open to all users, where staff, visitors and patients without mobility needs pay to use the transport and those with mobility needs still get the service for free. The research will investigate several key questions.

RQ1: How should the journeys on the service be optimised to take account of passenger constraints and appointment times?
RQ2: How should the service be priced to make it attractive to hospital users?
RQ3: At what scale of usage does the shared transport become cost neutral for the healthcare authority?
Other research questions may become apparent during the first year of the research and in discussions with stakeholders.
The project will involve designing and coding algorithms for optimal routing and scheduling and some stochastic optimisation for optimal pricing. Evaluation of methods will be done via simulation and this will also need to be coded by the student. We anticipate working with one or more hospital authorities and the student will also need to have excellent communication skills to be able to discuss their ideas and results with non-experts and glean information about existing systems from hospital representatives.
AI for Transportation and LogisticsChristine Currie
AI-Enabled Predictive Analytics for Sustainable Maritime Decarbonization: Trajectory Forecasting and Alternative Fuel Optimization Using Uncertain AIS DataThe global shipping industry faces urgent pressure to reduce pollution to the environment and decarbonize while maintaining economic viability. With the IMO targeting net-zero emissions by 2050, transitions to alternative fuels like green ammonia, liquid hydrogen, and dual-fuel systems promise reductions of up to 90% in CO2, but introduce challenges: limited refuelling infrastructure, varying energy densities, and route dependencies that could inadvertently increase emissions if not optimized. This PhD project, funded through Sust.AI, harnesses cutting-edge AI and data analytics to address these issues, developing tools that predict vessel behaviours from imperfect real-world data and simulate sustainable pathways forward.

Hosted at the Department of Decision Analytics and Risk under the supervision of Prof. Patrick Beullens, a leading expert in logistics analytics and alternative shipping fuels (see references), and Dr James Stallwood, an expert in computer science and AI, the project leverages exclusive access to high-fidelity Automatic Identification System (AIS) data. AIS provides near-real-time vessel positions, speeds, and headings, but suffers from gaps, delays, and inaccuracies reported by captains—issues that undermine traditional forecasting. This research will advance probabilistic AI models to deliver reliable, uncertainty-aware predictions.

Core Innovations:
• Trajectory Prediction Under Uncertainty: Employ Transformer-based deep learning (e.g., attention mechanisms for long-range dependencies) fused with Bayesian neural networks to forecast destinations and routes for tankers like VLCCs. This enables "what-if" scenarios: How might economic shifts (e.g., volatile oil markets) alter shipping patterns, helping operators select profit-maximizing ports while avoiding congestion?
• Decision Optimization: Integrate predictions into Partially Observable Markov Decision Processes (POMDP) frameworks enhanced by reinforcement learning, allowing vessels to dynamically choose routes that balance long-term profits with sustainability constraints—e.g., detouring to hydrogen-compatible ports despite added distance.
• Environmental Impact Forecasting: Simulate fleet-wide transitions using agent-based models, quantifying trade-offs for fuels like ammonia (high energy density but NOx risks) versus hydrogen (zero-emission but short range). Outputs will include emission heatmaps, infrastructure gap analyses, and policy recommendations, revealing how better movement understanding can amplify decarbonization benefits (e.g., 20-30% GHG cuts via optimized bunkering).

The successful candidate—a motivated PhD with a background in AI, data science, or environmental engineering—will work in a collaborative environment with industry partners, gaining hands-on experience with proprietary datasets and publishing in top relevant journals and presentation of their work at leading academic conferences. This project not only advances AI's role in handling "noisy" real-world data but also contributes to a greener maritime future, fostering innovations that could save billions in fuel costs and avert millions of tons of emissions annually. Applications are welcome from diverse candidates passionate about AI for planetary good.

References

Beullens, P. 2025. Liquid Hydrogen Mobile Storage and Dispensing for Aircraft in the UK. University of Southampton, Deliverable to Innovate UK project 10076104, 204 pages.
Xu, K., Brito, M.P. and Beullens, P., 2025. Multi-criteria feature selection on maritime emission abatement alternatives. Research in Transportation Business & Management, 59, p.101288.
Svirschi, O., Beullens, P. and Arruda, E., Optimal Speed and Hedging Strategies for Tramp Shipping Operators in Volatile Freight Markets. Available at SSRN 5114763.
Song, Y., Beullens, P. and Hudson, D., Job Acceptance and Economic Travel Time of a Tramp Ship Under Risk. Available at SSRN 4821126.
Beullens, P., Ge, F. and Hudson, D., 2023. The economic ship speed under time charter contract—A cash flow approach. Transportation Research Part E: Logistics and Transportation Review, 170, p.102996.
Ge, F., Beullens, P. and Hudson, D., 2021. Optimal economic ship speeds, the chain effect, and future profit potential. Transportation Research Part B: Methodological, 147, pp.168-196.
AI for Transportation and LogisticsPatrick Beullens
Sustainable Interactive Information Access for Recycling (STAIR)Household waste recycling is a cornerstone of the UK’s environmental strategy, yet major challenges remain. In 2021, the UK achieved a 44.6% recycling rate for household waste, with significant regional variation. Despite progress, surveys indicate that UK households discard billions of plastic items annually, with only a fraction being recycled domestically. Improving recycling efficiency and accuracy is therefore critical to reducing landfill use and promoting a circular economy.

This project explores how interactive information access (IIA) and AI-powered systems can enhance recycling by enabling accurate, efficient, and sustainable identification and sorting of waste materials. Traditional recycling methods often rely on manual sorting, which is time-consuming and error-prone. Computer vision, multimodal information processing, and interactive retrieval technologies could be used to design systems that distinguish materials, reduce contamination, and improve recycling throughput.

The PhD will focus on developing advanced controllers and learning mechanisms inspired by human motor control. These will combine a priori internal models with adaptive learning from multimodal data, including image recognition and sensor inputs, to improve robotic grasping and sorting of recyclables. Generative AI may also be explored to enrich multimodal data, increasing the robustness of control strategies.

Key research questions include:

How can interactive information access and multimodal learning reduce the error rate in material identification and sorting?

What control strategies allow robots to handle complex, irregular waste objects with poorly defined material properties?

How can sustainability be embedded into system design, ensuring reduced carbon intensity and improved efficiency?

The project will employ experimental robotics, multimodal data modelling, generative AI, and evaluation in collaboration with industrial recycling partners. Outcomes will support more efficient, sustainable, and scalable recycling infrastructures.		AI for Sustainable Energy and Buildings, Sustainable AI, AI for Sustainable Operations and Circular EconomyIan Williams
Energy-Efficient Verification of Privacy-Preserving AI with Hardware AccelerationPrivacy-preserving technologies are increasingly embedded into AI systems to ensure legal compliance and build public trust. Two key mechanisms are:
- Differential Privacy (DP), which ensures individual training data cannot be inferred from model outputs;
- Machine Unlearning (MU), which enables AI models to remove the influence of specific data points after training, as required under regulations like the GDPR.

While both methods protect user privacy, they pose a significant challenge: there is no standard, efficient way to verify that these protections have actually been applied. Service providers may need to prove to regulators, auditors, or even end users that Differential Privacy was correctly enforced during training, or that Machine Unlearning was properly executed in response to a data deletion request. Without such verification, compliance claims remain untrustworthy and unverifiable.

Cryptographic techniques such as Zero-Knowledge Proofs (ZKPs) offer a powerful way to enable this verifiability without revealing sensitive model details or data. However, they are computationally intensive and energy-hungry, especially when applied to modern AI models at scale or in real-time applications — making their practical deployment a significant sustainability concern.

This project aims to explore how hardware-software co-design can enable energy-efficient verification of privacy-preserving AI mechanisms. Specifically, the student will:
- Develop verifiable protocols for both DP (during training) and MU (during model updates and deletion requests);
- Design and evaluate Zero-Knowledge Proof systems that certify the correct application of DP noise or MU adjustments;
- Implement computationally intensive components (e.g., matrix operations, commitments, modular arithmetic) on Field-Programmable Gate Arrays (FPGAs);
- Analyse the trade-offs between privacy guarantees, computational cost, accuracy, and energy consumption.

A key innovation is treating verifiability as a first-class sustainability concern. As privacy regulations become stricter, large-scale systems will need to provide auditable proof of compliance. Doing so at low energy cost is critical for the long-term viability of responsible AI.		Sustainable AIBasel Halak
AI-Driven design and circularity optimisation of novel hybrid composite materialsComposite materials hold tremendous potential in enabling a low-carbon and sustainable future, particularly in renewable energy production, sustainable energy storage, and lightweight solutions for transportation. Yet, conventional approaches to designing and deploying composites are not inherently sustainable and often lack the efficiency needed to meet rising performance demands under extreme conditions.

Hybrid composites—combinations of different fibres and matrix systems—offer unique opportunities to achieve superior balances of mechanical performance, cost-effectiveness, and sustainability. For instance, recent studies have shown that glass–carbon hybrid composites can improve energy absorption by up to 78% compared with single-fibre carbon systems under impact loading for an Aero-engine application, therefore improve the design efficiency and reduce fuel consumption [1] . Despite this promise, the design space for hybrid composites is huge. Variations in fibre type, stacking sequence and volume fraction etc produce complex constitutional structure–property relationships and pose significant challenges for recyclability and circularity at end-of-life for these multi-material systems. Traditional trial-and-error methods and purely physics-based simulations are both time-intensive and costly, limiting the exploration of this high-dimensional space.

Advances in artificial intelligence (AI), particularly machine learning (ML) and generative design, now offer powerful tools to address these challenges in conventional composites [2]. By integrating experimental datasets, computational simulations, and AI-driven optimisation methods, it becomes possible to rapidly identify and design novel hybrid composite configurations with tailored multifunctional properties, while also considering recyclability and circularity.

This project will develop an AI-driven framework to optimise the design and circularity of hybrid composite materials. The framework will integrate physics-based modelling, data-driven approaches, and experimental validation to accelerate the discovery of sustainable, high-performance hybrid composites with enhanced lifecycle performance. The student will:
1) collect data from experiment, published database, and also data that generated from pre-established 2D axisymmetric and multi-scale model,
2) apply ML methods to predict mechanical properties from compositional and other parameters,
3) employ multi-objective optimisation techniques (e.g., Bayesian optimisation) to identify trade-offs between mechanical performance, cost, and recyclability, incorporating lifecycle assessment (LCA) tools and circularity indicators,
4) perform experimental validation for these AI-optimised design.

[1] Wu X, Finlayson J, Wisnom MR, Hallett SR, Improved energy absorption of novel hybrid configurations under static indentation, proceedings of the 20th European Conference on Composite Materials ECCM20. 26-30 June, 2022, Lausanne, Switzerland.
[2] Wang YF, Wang K, Zhang C, Applications of artificial intelligence/machine learning to high-performance composites, Compos B Eng, 285 (2024) 111740.
AI for Sustainable Operations and Circular Economy, AI for Transportation and LogisticsXun Wu
Safeguarded AI for a Sustainable Power GridThe transition towards Net Zero requires power systems that can accommodate vast amounts of distributed renewable energy, while maintaining security, reliability, and cost-effectiveness. In this respect, Artificial Intelligence (AI) offers significant potential to optimise power system operations. However, unlike traditional power system operation which rely on centralized control, future energy systems will increasingly be operated by crowds of decentralised AI agents, each controlled by different stakeholders, that can be potentially vulnerable to cyber-attacks.

The core research question is: how can we guarantee safety in power systems with decentralised AI agents operating across market and physical layers, ensuring resilience against malicious interference? While individually “safe” AI agents may comply with safeguards, their collective behaviour can still lead to unsafe system-wide outcomes, as highlighted in control theory and multi-agent systems research. This problem is largely unaddressed in the current literature, where most safety frameworks assume centralised oversight.

This PhD project proposes to develop and validate an integrated workflow for detecting and mitigating unsafe emergent behaviours within crowds of AI agents in power system operations. The research will draw on methods from optimisation, control theory, game theory, and cybersecurity to ensure resilience not only under normal operating conditions but also in the presence of adversarial threats. These approaches will be tested in scenarios including electricity markets, distributed resource coordination, and renewable energy integration. By systematically benchmarking these methods, the project aims to deliver the first scalable approach to guaranteed safe and resilient decentralised AI for power systems.

Join us to make AI-driven power systems safer by reducing the multi-billion-pound costs of balancing the UK grid, enabling higher renewable generation, and lowering the risk of large-scale blackouts. With this PhD project you will work on advancing the field of safeguarded AI and tackle one of its most critical application areas: the energy transition.
AI for Sustainable Energy and BuildingsErisa Karafili
Sensors that see and think: Sustainable AI hardware with ultrathin semiconductorsThe project aims to design the next generation of sustainable AI hardware! The PhD student will explore how atomically thin semiconductors can be engineered into devices that combine sensing, memory, and computing in one platform. You’ll investigate cutting-edge approaches to in-memory computing and retinomorphic vision sensors, inspired by the human brain and eye, with the goal of reducing the energy footprint of AI.

Project Overview
Conventional computing architectures face severe energy bottlenecks due to the separation of memory and processing. This so-called von Neumann bottleneck limits the efficiency of artificial intelligence (AI) systems, particularly in data-intensive tasks such as image recognition and real-time sensing. In this project, you will explore how atomically thin (2D) materials – such as layered ferroelectrics and semiconductors – can be engineered into novel devices that combine sensing, memory, and processing within the same physical platform.(1)

Research Challenges

In-memory computation: How can ferroelectric resistive switching in ultrathin materials be harnessed for energy-efficient logic and memory that can mimic the human brain?

Retinomorphic sensing: Can we design photonic/electronic devices that directly process visual information (edge detection, contrast enhancement) at the sensor level, mimicking the human retina?

Scalability: What device engineering strategies will improve device reliability (2) and enable integration into larger arrays while maintaining low-power, high-speed operation?

Sustainability: How can these emerging device technologies contribute to Sustainable AI by reducing the carbon footprint of training and deploying AI hardware?

Methods and Approach
As part of the MINDS lab (2dminds.co.uk) and Sustainable Electronic Technologies group, the student will build on the recently pioneered van der Waals contacts technology to fabricate and characterise reliable nano devices based on 2D ferroelectrics, semiconductors, and heterostructures (3,4). Electrical and optical testing will probe their memory, switching, and sensory capabilities. The project combines nanofabrication at Southampton's world-class cleanrooms, with opportunities for collaboration in system-level modelling and AI applications. Collaborations with the University of Cambridge and National Physical Laboratory on materials science and characterisation will allow benchmarking against industry standards.

(1) Feng Guangdi, et al. "Retinomorphic hardware for in‐sensor computing." InfoMat 5.9 (2023): e12473.
(2) Wang Yan, Sarkar Soumya et al. "Critical challenges in the development of electronics based on two-dimensional transition metal dichalcogenides." Nature Electronics 7.8 (2024): 638-645.
(3) Sarkar, Soumya, et al. "Spin injection in graphene using ferromagnetic van der Waals contacts of indium and cobalt." Nature Electronics 8.3 (2025): 215-221.
(4) Sarkar, Soumya, et al. "Multistate ferroelectric diodes with high electroresistance based on van der Waals heterostructures." Nano letters 24.42 (2024): 13232-13237.		Sustainable AISoumya Sarkar
Deep Learning for Predicting Defects in Semiconductors ManufacturingManufacturing semiconductor chips, such as processors, graphic or memory units, is one of the most complicated and costly industrial manufacturing processes. Each one of the advanced processor chips that we carry in our phones undergoes thousands of manufacturing steps at fabrication plants (Fabs), where the recipe in each step must be finely tuned to lead to the desired results. This makes the research, development, and testing cycles of semiconductors extremely costly, time-consuming and labour-intensive.

This PhD project aims to develop models and tools that predict nanoscale defects in semiconductor manufacturing, improving yield, reducing development cycles and making the industry more economical and sustainable. As a PhD student, you will be utilising the exceptional fabrication facilities we have at the University of Southampton, the large supercomputer cluster (IRIDIS) and collaborations with experts in AI. You will also work in collaboration with a team in Silvaco, a global industrial leader in fabrication modelling and work as part of the spinout company, Deep Fabrication Ltd., toward commercialising your models to the semiconductor industry.

To give you a flavour of the workflow within this project:
• The first stage will require the experimental fabrication of nanoscale structures using e-beam lithography, plasma etching, and material deposition tools.
• The second stage will require collecting data from characterising these nanostructures using microscopic and spectroscopic tools and feeding the data to AI/Machine learning algorithms.
• The algorithms will then be tested in the third stage in real scenarios to assess their capability to predict and mitigate defects in fabrication and optimise the fabrication process.

You will work in one of the world’s most advanced university cleanrooms and a comprehensive suite of characterisation labs. You’ll also use the IRIDIS supercomputing cluster and collaborate with AI experts across the university. Industrial links are central: you will interact with Silvaco (a global leader in fabrication modelling) and contribute to a commercial pathway within our new spin-out company Deep Fabrication Ltd.

The supervisory team, Dr Yasir Noori (ECS), Dr Ben Mills (ORC) and Dr Firman Simanjuntak (ECS), have combined an extensive set of expertise in semiconductor fabrication, artificial intelligence for manufacturing, and characterisation of devices and systems.

After completing your PhD you will have a unique skill set that will put you at the forefront of future semiconductor fabrication, making you well positioned to have a fruitful career in academia or industry.

References:
O. Buchnev, et al. “Deep-Learning-Assisted Focused Ion Beam Nanofabrication” Nano Letters, 22, 2734-2739, 2022.
M. Nandipati, et al. “Bridging Nanomanufacturing and Artificial Intelligence – A Comprehensive Review” Materials, 17, 1621, 2024.
G. Tello, et al. “Deep Structures Machine Learning Model for the Recognition of Mixed-Defect Patterns in Semiconductor Fabrication Processes”, IEEE Transactions on Semiconductor Manufacturing, 31, 2, 2018.
M. Maggipinto, et al. “DeepVM: A Deep Learning-based Approach with Automatic Feature Extraction for 2D Input Data Virtual Metrology” Journal of Process Control, 84, 24-34, 2019.
Yijie Liu, et al. “Towards Smart Scanning Probe Lithography: A Framework Acceleration Nanofabrication Process with in-situ Characterisation via Machine Learning” Microsystems & Nanoengineering, 9, 128, 2023.

AI for Sustainable Operations and Circular EconomyYasir Noori
Physical AI for sustainable robotsThis project addresses critical sustainability challenges in robotics by developing physical computing systems that eliminate dependence on semiconductor electronics. Current robots rely heavily on microprocessors for information processing and control, creating substantial supply chain vulnerabilities and electronic waste at end-of-life. Additionally, semiconductor electronics cannot function in hazardous environments such as gamma radiation zones, limiting robotic deployment.

This project pioneers the development of physical computing blocks for robot control and information processing made entirely from soft, biodegradable materials. This involves creating central pattern generators, sensory receptors (for sensing contact, light, etc.), logic gates, higher-order decision circuits utilising interactions between soft structures eliminating the need for microcontrollers or semiconductors. Building on recent unpublished results from Dr. Godaba’s group on electronics-free electrofluidic nervous systems, the work draws inspiration from information processing in simple invertebrates. The goal is to design minimal, efficient systems capable of processing sensory inputs and generating control signals to operate multi-actuator soft robots.

This project will pave the foundation for fully soft and biodegradable robots that can sense and compute without the need for semiconductor electronics, relying on computation by mechanical structures. In the long term, these foundational blocks can help create robots that can harvest energy from nature, conduct functional tasks and biodegrade completely. The approach addresses multiple sustainability challenges: eliminating electronic waste, reducing supply chain dependencies, enabling operation in hazardous environments, and creating truly sustainable robotic systems.

This project is suitable for students interested in hands-on fabrication of novel soft robots. The project offers significant creative opportunities in developing novel computational architectures and robot design.

References
• Sitti M. Physical intelligence as a new paradigm. Extreme Mech Lett. 2021 Apr 26;46:101340.
• M. Garrad et al. A soft matter computer for soft robots. Sci. Robot.4,eaaw6060(2019). DOI:10.1126/scirobotics.aaw6060
• Contact Dr Hareesh Godaba for further information.
Sustainable AI, AI for Sustainable Operations and Circular EconomyHareesh Godaba
AI for Ship Decarbonisation: Unravelling Heterogeneity, Causality, and Complexity in Fuel Consumption PredictionThe international shipping industry serves as the backbone of global trade, carrying over 80% of international cargo and playing a critical role in global economic integration (Fan et al., 2025; World Economic Forum, 2024; Yan et al., 2024). However, this vital role comes at a significant environmental cost. The sector is one of the largest consumers of fossil fuels, contributing approximately 3% of total global greenhouse gas (GHG) emissions (Transport & Environment, 2024). In response to the escalating climate crisis, the International Maritime Organization (IMO) has implemented stringent regulatory measures, such as the Energy Efficiency Index for Existing Ships (EEXI) and the Carbon Intensity Indicator (CII), to steer the industry toward a net-zero future (International Maritime Organization, 2023; Tadros et al., 2023).
Achieving these ambitious decarbonisation targets requires a profound transformation in how vessel energy efficiency is managed. While technological innovations are crucial, operational strategies, such as speed regulation, route optimisation, and trim adjustment, are recognised as the most cost-effective and immediately applicable solutions (Godet et al., 2024; Yan et al., 2024). The effectiveness of these strategies fundamentally depends on the accurate prediction and deep understanding of Ship Fuel Consumption (SFC). An accurate SFC model not only provides a quantitative foundation for evaluating energy-saving interventions but also forms the theoretical basis for intelligent energy management systems that will underpin a sustainable maritime future (Wang et al., 2023).

Despite recent advances in applying machine learning (ML) to SFC prediction, existing approaches face several fundamental limitations that constrain their practical applicability and scientific impact. This research identifies three core and interconnected challenges, which form the foundation of the proposed study.

Challenge 1: The “Homogeneity Assumption” and Operational Heterogeneity
Current modelling paradigms often treat the complex and dynamic process of ship navigation as a statistically homogeneous system, assuming that a single global model can represent SFC mechanisms across all operational scenarios (Feng et al., 2024). This “homogeneity assumption” neglects the intrinsic diversity of ship operations, where different conditions, such as cruising in rough seas, navigating calm waters, or manoeuvring in port, are governed by distinct physical dynamics and energy consumption behaviours (Fan et al., 2025). Such oversimplification obscures essential variations, limits model generalisability, and results in unreliable predictions under non-standard conditions, thereby impeding the development of targeted, condition-specific energy optimisation strategies (Iqbal et al., 2025).

Challenge 2: The “Black Box” Problem and the Correlation-Causation Fallacy
Although advanced ML models achieve high predictive accuracy, their “black box” nature hinders transparency, credibility, and adoption in safety-critical maritime operations (Wang et al., 2023). Most interpretability studies provide only a ranking of feature importance, failing to uncover the complex non-linear relationships, thresholds, and inflection points that truly describe how factors influence SFC (Xie et al., 2024). More importantly, these models reveal statistical correlations rather than causal mechanisms (Huang et al., 2024). This distinction is crucial: effective energy efficiency policies must be grounded in an understanding of causal drivers, not merely correlated variables.

Challenge 3: Methodological Fragmentation and High-Dimensional Complexity
Current research remains fragmented, with limited integration between predictive modelling, explainable AI (XAI), and causal inference methods (Li et al., 2024). Additionally, the fusion of multi-source data—such as noon reports, AIS, and environmental datasets—introduces challenges including feature redundancy, multicollinearity, and the curse of dimensionality, which can increase computational burden and lead to overfitting (Afshar & Usefi, 2020). Existing feature selection and hyperparameter optimisation techniques are often single-model-based, prone to bias, or computationally intensive, making them unsuitable for complex, high-dimensional maritime datasets (Muñoz et al., 2025).
These challenges lead to three central research questions that will guide this doctoral study:

RQ1: How can the intrinsic heterogeneity of ship operational states be systematically identified, modelled, and interpreted to develop more accurate, robust, and physically meaningful SFC prediction frameworks?

RQ2: How can we move beyond correlation to establish and quantify the true causal relationships between key operational, navigational, and environmental factors and SFC to inform effective and reliable intervention strategies?

RQ3: How can a unified analytical framework be developed to integrate advanced data fusion, robust feature selection, efficient model optimisation, and deep interpretability—ultimately creating transparent, trustworthy, and actionable decision-support tools for ship energy management?

This research aims to develop an AI-aided, interpretable, and causally informed framework for ship fuel consumption prediction, bridging the gap between predictive accuracy, scientific understanding, and practical applicability. By addressing the challenges of heterogeneity, causality, and complexity, this study will advance both the methodological frontier of maritime AI research and the practical goal of achieving low-carbon, energy-efficient shipping operations.

REFERENCES
Afshar, M., Usefi, H., 2020. High-dimensional feature selection for genomic datasets. Knowledge-Based Systems 206, 106370. https://doi.org/10.1016/j.knosys.2020.106370
Birchler, J.A., Veitia, R.A., 2007. The Gene Balance Hypothesis: From Classical Genetics to Modern Genomics. The Plant Cell 19 (2), 395-402. https://doi.org/10.1105/tpc.106.049338
Fan, A., Wang, Y., Hu, Z., Yang, L., Fan, X., Yang, Z., 2025. Multi-dimensional performance verification of ship fuel consumption prediction model under dynamic operating conditions. Energy 332, 137120. https://doi.org/10.1016/j.energy.2025.137120
Feng, Y., Wang, X., Chen, Q., Yang, Z., Wang, J., Li, H., Xia, G., Liu, Z., 2024. Prediction of the severity of marine accidents using improved machine learning. Transportation Research Part E: Logistics and Transportation Review 188, 103647. https://doi.org/10.1016/j.tre.2024.103647
Godet, A., Panagakos, G., Barfod, M.B., Lindstad, E., 2024. Operational cycles for maritime transportation: Consolidated methodology and assessments. Transportation Research Part D: Transport and Environment 132, 104238. https://doi.org/10.1016/j.trd.2024.104238
Huang, H., Li, B., Wang, Y., Zhang, Z., He, H., 2024. Analysis of factors influencing energy consumption of electric vehicles: Statistical, predictive, and causal perspectives. Applied Energy 375, 124110. https://doi.org/10.1016/j.apenergy.2024.124110
International Maritime Organization, 2023. EEXI and CII - ship carbon intensity and rating system. https://www.imo.org/en/mediacentre/hottopics/pages/eexi-cii-faq.aspx
Iqbal, S., Qureshi, A.N., Alhussein, M., Aurangzeb, K., Mahmood, A., Razalli Bin Azzuhri, S., 2025. Dynamic selectout and voting-based federated learning for enhanced medical image analysis. Machine Learning: Science and Technology 6 (1), 015010. https://doi.org/10.1088/2632-2153/ada0a6
Jia, Y., Hu, X., Kang, W., Dong, X., 2024. Unveiling Microbial Nitrogen Metabolism in Rivers using a Machine Learning Approach. Environmental Science & Technology 58 (15), 6605-6615. https://doi.org/10.1021/acs.est.3c09653
Li, S., Pu, Z., Cui, Z., Lee, S., Guo, X., Ngoduy, D., 2024. Inferring heterogeneous treatment effects of crashes on highway traffic: A doubly robust causal machine learning approach. Transportation Research Part C: Emerging Technologies 160, 104537. https://doi.org/10.1016/j.trc.2024.104537
Muñoz, V., Ballester, C., Copaci, D., Moreno, L., Blanco, D., 2025. Accelerating hyperparameter optimization with a secretary. Neurocomputing 625, 129455. https://doi.org/10.1016/j.neucom.2025.129455
Snyder, M., Gerstein, M., 2003. Defining Genes in the Genomics Era. Science 300 (5617), 258-260. https://doi.org/10.1126/science.1084354
Tadros, M., Ventura, M., Guedes Soares, C., 2023. Review of the IMO Initiatives for Ship Energy Efficiency and Their Implications. Journal of Marine Science and Application 22 (4), 662-680. https://doi.org/10.1007/s11804-023-00374-2
Transport & Environment, 2024. Climate impact of shipping. https://www.transportenvironment.org/topics/ships/climate-impact-shipping
Wang, H., Yan, R., Wang, S., Zhen, L., 2023. Innovative approaches to addressing the tradeoff between interpretability and accuracy in ship fuel consumption prediction. Transportation Research Part C: Emerging Technologies 157, 104361. https://doi.org/10.1016/j.trc.2023.104361
World Economic Forum, 2024. Reducing barriers to maritime fuel projects is key to decarbonizing shipping. https://www.weforum.org/stories/2024/04/why-reducing-barriers-for-maritime-fuel-projects-is-key-to-progressing-on-decarbonization
Xie, C., Hu, J., Vasdravellis, G., Wang, X., Cheng, S., 2024. Explainable AI model for predicting equivalent viscous damping in dual frame–wall resilient system. Journal of Building Engineering 96, 110564. https://doi.org/10.1016/j.jobe.2024.110564
Yan, R., Yang, D., Wang, T., Mo, H., Wang, S., 2024. Improving ship energy efficiency: Models, methods, and applications. Applied Energy 368, 123132. https://doi.org/10.1016/j.apenergy.2024.123132
AI for Sustainable Energy and BuildingsHuanhuan Li
AI Control in Chaotic Skies for Perpetual FlightPerpetual flight has long represented a pinnacle challenge in aerospace systems design, offering the potential for persistent aerial presence in applications ranging from atmospheric science to surveillance, climate monitoring, and communications. Achieving this requires real-time location and exploitation of atmospheric energy sources (such as thermals, convective currents, and updrafts) that can sustain flight.
The vision of perpetual flight, UAVs soaring indefinitely by harvesting energy from the atmosphere, offers an exciting path towards sustainable aviation and climate-friendly aerial platforms. But to make this dream a reality, we need control systems that can remain stable in the face of turbulence, gusts, and chaotic atmospheric behaviour.

This PhD will develop robust, adaptive control strategies that enable UAVs to capitalise on soaring opportunities while operating with partial and uncertain information. You will explore nonlinear control, reinforcement learning, and adaptive strategies that can balance energy harvesting with stability and safety in a constantly changing environment. The aim is to design controllers that are resilient to uncertainty yet efficient enough to deliver long-duration, energy-neutral flight.

The broader significance extends well beyond UAVs. The AI-driven control methods you develop for managing uncertainty, adapting to sparse sensing, and making real-time decisions in chaotic systems are directly applicable to autonomous vehicles, renewable energy systems, environmental robotics, and resilient infrastructure management. This project is ideal for candidates who want to tackle some of the hardest problems in control and autonomy while contributing to the development of climate-resilient technologies.

This research aim to answer:

1. How can reinforcement learning be integrated with nonlinear flight dynamics to enable UAVs to autonomously identify and exploit atmospheric energy sources in real time?

2. What adaptive control architectures can ensure stability and safety in UAVs operating under uncertain, partially observable, and turbulent atmospheric conditions?

3. How can uncertainty-aware AI models (e.g., Bayesian or physics-informed neural networks) be used to predict and respond to spatiotemporal variations in convective and thermal fields?

4. Can hybrid AI–control frameworks achieve sustained, energy-neutral flight by continuously balancing exploration (energy harvesting) and exploitation (stability and efficiency)?

References:

[1] Ericson, W., Achermann, F., Lawrance, N., Villinger, J., Andersson, O. and Siegwart, R., 2025. Evaluating machine‐learning models for wind‐speed downscaling from ECMWF‐IFS data. Quarterly Journal of the Royal Meteorological Society, p.e5063.

[2] Lim, J., Achermann, F., Lawrance, N. and Siegwart, R., 2024. Autonomous Active Mapping in Steep Alpine Environments with Fixed-wing Aerial Vehicles. arXiv preprint arXiv:2405.02011.

[3] McGee, T.G., Adams, D., Hibbard, K., Turtle, E., Lorenz, R., Amzajerdian, F. and Langelaan, J., 2018. Guidance, navigation, and control for exploration of titan with the dragonfly rotorcraft lander. In 2018 AIAA Guidance, Navigation, and Control Conference (p. 1330).

[4] Bird, J.J. and Langelaan, J., 2017. Design Space Exploration for Hybrid Solar/Soaring Aircraft. In 17th AIAA Aviation Technology, Integration, and Operations Conference (p. 4092). 		Sustainable AI, AI for Transportation and Logistics, AI for the Natural EnvironmentBahareh Zaghari
Realizing Green Electrification in Transportation with AI SupportRealizing green electrification in road freight is pivotal for net-zero, yet most evidence and tools were built for passenger EVs rather than heavy-duty commercial fleets whose choices hinge on route structure, depot topology, payload–range trade-offs, driver hours, and strict service windows. This project uses an AI-enabled, Smart Predict-then-Optimize (SPO) pipeline to model and ultimately understand how electrification alters effective (revealed) logistics demand: origin–destination flows, shipment size and timing, reliability requirements, and carrier/mode choice, under technology and policy scenarios, then differentiates through fleet sizing and electric vehicle routing with charging to produce prescriptions that minimise cost, meet service levels, and reduce emissions. AI improves in three ways that matter for electrification: it fuses high-dimensional, nonstationary data (ELD/telematics, TMS orders, energy prices, congestion, weather, charger availability) to learn nonlinear thresholds and interactions that drive shipper behaviour; it addresses endogeneity between price, service quality, and demand with causal ML rather than assuming fixed coefficients; and, via SPO, it trains forecasts for the downstream objective (cost, reliability, emissions) instead of generic statistical fit, delivering decisions that remain robust under uncertainty and distribution shift (Elmachtoub & Grigas, 2022; Bertsimas & Kallus, 2020).

The work concentrates on three questions. First, the decision-focused value: by how much does SPO-based, AI demand modelling improve total logistics cost, service reliability, and emissions versus parametric baselines (Elmachtoub & Grigas, 2022)? Second, the optimal fleet–charging mix under uncertainty: what combination of battery-electric trucks and charger types (depot, opportunity, megawatt), with what siting and schedules, minimises system cost while meeting service levels when demand is stochastic and temperature/grade affect usable range, benchmarked to current total-cost-of-ownership and adoption evidence for zero-emission HDVs (Basma et al., 2021–2023)? Third, policy efficiency: which instruments, capex subsidies, LCFS-style credits, carbon fees, access/toll exemptions, deliver the largest electrification gains per public pound once operational responses are optimised with SPO and demand reallocation is taken into account, and how sensitive are results to regional energy prices and corridor densities (International Energy Agency, 2023 & 2024)?

Methodologically, we will learn probabilistic, lane-level demand and service-elasticity models from telematics and order data; generate counterfactual demand under technology and policy scenarios; and embed these forecasts in multi-period fleet composition and electric VRP with time windows and charging, capturing dwell, degradation, payload penalties, and ambient conditions. Strategic analysis will evaluate competition and cooperation among carriers and infrastructure providers—shared depots, megawatt-charging hubs, and data pooling—linking micro-level routing costs to macro TCO and adoption timing. Validation will combine held-out corridors, event studies (e.g., depot openings, low-emission zone expansions), and reconciliation to ICCT benchmarks and IEA outlooks for external validity. The results will be converted into better routing and infrastructure choices, and identify the most cost-effective policies to accelerate heavy-duty fleet decarbonisation.

Basma, H., Saboori, A., & Rodríguez, F. (2021, November). Total cost of ownership for tractor-trailers in Europe: Battery electric versus diesel. International Council on Clean Transportation.
Basma, H., Buysse, C., Zhou, Y., & Rodríguez, F. (2023, April). Total cost of ownership of alternative powertrain technologies for Class 8 long-haul trucks in the United States. International Council on Clean Transportation.
Bertsimas, D., & Kallus, N. (2020). From predictive to prescriptive analytics. Management Science, 66(3), 1025–1044.
Elmachtoub, A. N., & Grigas, P. (2022). Smart “Predict, then Optimize”. Management Science, 68(1), 9–26.
International Energy Agency. (2023, April). Global EV Outlook 2023: Catching up with climate ambitions. IEA.
International Energy Agency. (2024, April). Global EV Outlook 2024: Moving towards increased affordability. IEA.		AI for Transportation and LogisticsFangsheng Ge
Statistical machine learning and uncertainty quantification to develop of a decision support tool for offshore wind foundation installationThe goal of this project is to develop a mechanism-based machine learning algorithm, trained using a large database of field data, to optimise the installation of offshore wind turbine foundations and shape the decision-making process during installation.

In the UK alone, thousands of anchors and foundations must be installed every year to support the (floating) offshore wind turbines necessary to achieve the net zero objectives. Piles are currently the most commonly used foundation type and pile driving is the main installation method. However, this pile hammering creates loud underwater noise which is particularly detrimental to marine mammals. On the contrary, suction piles are installed by pumping water on the inside of the pile, which creates a suction effect that pushes the pile into the ground without any impact noise.

The installation of suction piles is challenging. Premature refusal, which is the termination of the installation before reaching the targeted depth, may require the full removal of the pile before its relocation, which has a significant (carbon) cost. Refusal can be due to mechanisms such as piping (breaking the internal suction) or plug heave (internal soil upwards movement), but they are not yet well predicted, especially in challenging geologies such as hard or layered soils. Innovative techniques, such as suction cycling, have been recently introduced to overcome those issues, but their effect is not yet predictable.

The goal of this project is to develop a mechanism-based machine learning (ML) algorithm, with associated uncertainty quantification (UQ), to optimise the installation of suction pilesin terms of site selection or installation parameters (pumping sequence) to reduce installation time and risk. Standard ML may produce accurate predictions but will not typically account for engineering and physical knowledge of the suction process. Greater efficiency, in terms of required data and improved UQ, may be obtained by constraining the ML model via the known physical mechanisms. Such an ML model may also allow for limited extrapolation away from the training data. An effective model of the process can then be interrogated to propose new installation strategies and locations.

This goal will be achieved by meeting the following research objectives:

1. Identify the different failure mechanisms encountered during suction pile installation and determine their geotechnical features.
2. Develop a mechanism-based predictive model for suction pile installation that accounts for uncertainties in soil properties and provides model predictions that enable live decision making in the field. A first attempt might be using a physically constrained Gaussian process.
3. Propose new installation strategies, e.g, using techniques based on Bayesian optimisation, and quantify their effect on suction pile installation.

AI for Sustainable Operations and Circular Economy, AI for Transportation and LogisticsBenjamin Cerfontaine
Adiabatic logic for AI at the edgeFlexible electronics fabricated using, for example, Thin-Film Transistors (TFTs) are a promising technology for applications such as AI/ML at the edge. Such technologies often only implement N-type transistors, however, which is very inefficient in terms of energy. This technology can, however, be used to design complex systems [1].

Adiabatic logic was proposed as a low power technology several decades ago. The underlying principle is that logic gates should only be turned on as they are needed and not turned off when they are active. This is achieved by using the clock, or multiple clock phases, as the power supply for logic gates. Adiabatic logic designs have been proposed that use both NMOS and PMOS transistors, but designs that use only NMOS transistors are possible. In order to achieve the promised energy efficiency, however, such circuits are very slow compared with modern CMOS technologies. This means that adiabatic logic has never achieved widespread acceptance. On the other hand, modern flexible electronics operate at relatively low frequencies. Thus, adiabatic logic may prove to be an appropriate technique for low power design in these technologies.

This project will look at how adiabatic logic can be used for AI/ML. We have already shown that the energy-efficient clock speeds of adiabatic logic are compatible with the clock speeds achievable in flexible electronics [2]. Nevertheless, formidable design challenges exist. Existing design tools are not compatible with adiabatic design styles. Signal paths need to be carefully balanced in order that the correct clock phases activate logic in sequence. In principle, some energy recovery is possible in adiabatic circuits, but we need to consider on-chip energy storage.

Therefore, this project will focus on how basic Processing Elements (PEs) can be joined with memory elements to build Convolutional Neural Networks (CNNs) in adiabatic logic. The PEs will be Multiply-Accumulate units (MACs) with limited memory that use reduced precision fixed-point arithmetic. If time permits, we would hope to fabricate and test a basic system.

[1] Ozer, E., Kufel, J., Prakash, S. et al. Bendable non-silicon RISC-V microprocessor. Nature 634, 341–346 (2024). https://doi.org/10.1038/s41586-024-07976-y

[2] K. Chen, G. V. Merrett and M. Zwolinski, "Enabling Efficient Pure-NMOS Circuits Through Adiabatic Complementary Pass-Transistor Logic," 2025 23rd IEEE Interregional NEWCAS Conference (NEWCAS), Paris, France, 2025, pp. 158-162, doi: 10.1109/NewCAS64648.2025.11107155.
Sustainable AIMark Zwolinski
Smart AI "Predict-then-Optimise" in Maritime LogisticsThis project develops a new decision-focused learning framework by combining recent advances in the Smart Predict-then-Optimize (SPO) loss with deep neural networks. While the SPO loss provides a natural way to evaluate predictive models in optimization settings, its non-convexity and non-Lipschitz nature present challenges for both theory and practice. Building on recent progress in establishing generalisation bounds through margin-based relaxations, we extend the framework to deep learning models, offering both improved predictive accuracy and stronger decision guarantees.

Specifically, we would like to first use a deep neural network to predict unknown input parameters of an optimisation problem, and then make decisions by solving the optimisation problem using the predicted parameters. A natural loss function, referred as Smart Predict-then-Optimize (SPO) loss, is to measure the suboptimality of the decisions induced by the predicted input parameters, as opposed to measuring loss using input parameter prediction error. We will develop a tractable methodology for training neural networks under this SPO framework and derive dedicated theoretical garantuees such as consistency results and risk bounds.


The proposed method will be applied to maritime logistics, with a focus on emissions compliance under evolving international regulations. By predicting uncertain factors such as fuel prices, emissions rates, and operational conditions directly through neural networks and integrating them with optimization-based decision models, the project provides a robust tool for analyzing industry responses to policy interventions. The expected outcome is both theoretical—new generalization results for deep learning under SPO loss—and practical, offering evidence-based insights to guide regulators and shipping companies in balancing compliance costs with environmental objectives.

Refernece:

[1] Elmachtoub, A. N., & Grigas, P. (2022). Smart “predict, then optimize”. Management Science.

[2] Liu, H., & Grigas, P. (2021). Risk bounds and calibration for a smart predict-then-optimize method. Advances in Neural Information Processing Systems.

[3] El Balghiti, O., Elmachtoub, A. N., Grigas, P., & Tewari, A. (2019). Generalization bounds in the predict-then-optimize framework. Advances in neural information processing systems.

[4] Wen, J., Abeel, T., & de Weerdt, M. (2025). Performance and interaction assessment of neural network architectures and bivariate smart predict-then-optimize. Machine Learning.		AI for Transportation and LogisticsChao Zheng
AI modelling to prolong the life of civil engineering infrastructureCivil engineers must maintain existing, ageing transportation infrastructure to ensure that it is safe and resilient, while minimising resource use. This can be achieved by accounting for uncertainty and variability in ground and structural behaviour, for which traditional analysis techniques in civil engineering are poorly suited. This leads to over-conservatism in design and assessment that is unsustainable in terms of time, energy and cost. This project seeks to answer the central question: Can modern AI tools be used to design stable structures with more sustainable levels of conservatism?

Statistical methods have been proposed and developed for civil engineering analyses for many decades, but they have not been widely adopted, largely due to a lack of training and/or validation data. More recently, it has become much easier and cheaper to gather measurements from laboratory experiments and field monitoring that are suitable for statistical analyses. The Infrastructure Group at Southampton has gathered and published such data (Trinidad González et al. 2023; Huang et al. 2024; Briggs et al. 2024). Simultaneously, the increasing prominence of Statistical Machine Learning and AI has resulted in an influx of novel approaches to uncertainty quantification that could have significant impact in civil engineering. Members of the Statistics Research Group at Southampton have a track record in such methods (e.g. Cockayne et al. 2019; Oates et al. 2019).

This interdisciplinary PhD will explore whether the training of statistical models with appropriate input and validation data can be used to improve ground characterisation and the forecasting of structural performance. A stochastic finite element model will be developed to interpret uncertain ground conditions beneath civil engineering structures and forecast their long-term behaviour. These will include a trial embankment and a deep cutting in weathered clay. In each case, there are long-term ground monitoring datasets and associated ground investigation data available within the Infrastructure Group at Southampton to train and validate the model.

Stochastic back-analyses will enable engineers to determine the most likely composition of soil and rock strata providing the foundation of civil engineering structures. Statistical emulation will enable extrapolation of this approach for loading regimes (e.g. traffic, weather), ageing and ground materials (e.g. geology) applicable to a transportation network.

Initially we will focus on established approaches based on Gaussian processes, such as the Bayesian finite element method (Poot et al. 2024; Girolami et al. 2021) or emulation techniques (Kennedy & O’Hagan 2021). These are designed to accelerate components of the inference scheme described above, allowing for the computationally challenging exploration of design space to be accelerated, while still providing rigorous uncertainty quantification. The research will be used to quantify uncertainties when assessing and managing ageing infrastructure, to enable sustainable decision making and reduce over-conservatism.

References:
Briggs, K.M., Trinidad González, Y., Powrie, W., Butler, S. and Sartain, N., 2024. Quantifying CPT cone factors in clays derived from weathered mudstone. Quarterly Journal of Engineering Geology and Hydrogeology, 57(1), pp.qjegh2023-014.
Cockayne, J., Oates, C. J., Sullivan, T. J., & Girolami, M. (2019). Bayesian Probabilistic Numerical Methods. SIAM Review, 61(3), 756–789.
Huang, W., Loveridge, F.A., Briggs, K.M., Smethurst, J.A., Saffari, N. and Thomson, F., 2024. Forecast climate change impact on porewater pressure regimes for the design and assessment of clay earthworks. Quarterly Journal of Engineering Geology and Hydrogeology, 57(1), pp.qjegh2023-015.
Kennedy, M. C., & O’Hagan, A. (2001). Bayesian Calibration of Computer Models. Journal of the Royal Statistical Society Series B: Statistical Methodology, 63(3), 425–464.
Girolami, M., Febrianto, E., Yin, G., & Cirak, F. (2021). The statistical finite element method (statFEM) for coherent synthesis of observation data and model predictions. Computer Methods in Applied Mechanics and Engineering, 375, 113533.
Oates, C. J., Cockayne, J., Aykroyd, R. G., & Girolami, M. (2019). Bayesian Probabilistic Numerical Methods in Time-Dependent State Estimation for Industrial Hydrocyclone Equipment. Journal of the American Statistical Association, 114(528), 1518–1531.
Poot, A., Kerfriden, P., Rocha, I., & van der Meer, F. (2024). A Bayesian approach to modeling finite element discretization error. Statistics and Computing, 34(5).
Trinidad González, Y., Briggs, K.M., Svalova, A. and Glendinning, S., 2023. Evaluating the likelihood of slope failure in ageing earthworks using Bayesian updating. Infrastructure Asset Management, 10(4), pp.207-222.

AI for Sustainable Operations and Circular Economy, AI for Transportation and LogisticsKevin Briggs
AI-Enhanced Operation and Maintenance for Offshore Wind Farms: From Automated Risk Analysis to Resilient Energy SystemsOffshore wind energy plays a pivotal role in global decarbonisation strategies, yet its large-scale and deep-sea deployment faces major challenges, including high operation and maintenance (O&M) costs, limited offshore accessibility, unexpected equipment failures, and weather-related risks (Gao & Odgaard, 2023; H. Li & Guedes Soares, 2022; C. Zhang et al., 2025).
This project aims to develop a holistic AI-driven framework to optimise offshore wind farm O&M strategies by shifting from reactive to predictive maintenance. The proposed framework seeks to enhance system reliability, reduce lifecycle costs, and strengthen overall energy resilience (Alves Ribeiro et al., 2025; Díaz & Guedes Soares, 2020; M. Li et al., 2022; Zhong et al., 2019).

A key component of this project is leveraging sensor and Supervisory Control and Data Acquisition (SCADA) data to transform vast amounts of operational information into actionable insights. SCADA data refer to real-time operational data collected through Supervisory Control and Data Acquisition systems, which are widely used for monitoring and controlling industrial processes (Morrison et al., 2022). These data typically include measurements such as temperature, pressure, voltage, current, and equipment status, transmitted from sensors and controllers to a central control platform (Cheng et al., 2024; Jin et al., 2021). In the maritime and port context, SCADA data capture critical information on energy systems, cranes, pumps, and shore power operations, yet pose challenges due to their massive volume and complexity (Xiang et al., 2022; W. Zhang et al., 2022). Nevertheless, such data provide a valuable foundation for developing AI-based predictive maintenance, fault diagnosis, and energy optimisation models, enabling data-driven decision-making to support sustainable offshore energy operations (Badihi et al., 2022; Stetco et al., 2019).

The proposed research integrates advanced machine learning, natural language processing (NLP), uncertainty quantification, and decision optimisation to address four interconnected research challenges:
(1) Automated risk modelling: Use NLP to automatically extract failure knowledge from maintenance logs and reports, enabling systematic Failure Mode and Effects Analysis (FMEA), and apply uncertainty quantification to optimise sensor placement for better fault detection(Khurana et al., 2023; Sun et al., 2023).
(2) Intelligent fault prognosis: Develop multimodal fusion and multi-criteria decision models for early fault warning and prioritisation, and explainable AI algorithms to predict failures and estimate remaining useful life under uncertainty, enhancing model interpretability and trustworthiness (Y.-F. Li et al., 2024).
(3) Predictive maintenance and resource allocation: Build adaptive predictive maintenance schedules that account for weather-related risks to reduce downtime, minimise overall costs, and optimise resource use, including spare parts, vessel dispatch, and crew allocation (de Matos Sá et al., 2024; Si et al., 2025).
(4) Availability and resilience assessment: Create availability-based performance metrics and a framework to assess wind farm resilience, linking component reliability with overall energy output (Wilkie & Galasso, 2020).

By integrating these contributions, the project will deliver a transferable, data-driven methodology for intelligent O&M decision-making, with potential to significantly reduce operational costs and enhance the contribution of offshore wind to net-zero energy targets.

References
Alves Ribeiro, J., Alves Ribeiro, B., Pimenta, F., M.O. Tavares, S., Zhang, J., & Ahmed, F. (2025). Offshore wind turbine tower design and optimization: A review and AI-driven future directions. Applied Energy, 397, 126294. https://doi.org/10.1016/j.apenergy.2025.126294
Badihi, H., Zhang, Y., Jiang, B., Pillay, P., & Rakheja, S. (2022). A Comprehensive Review on Signal-Based and Model-Based Condition Monitoring of Wind Turbines: Fault Diagnosis and Lifetime Prognosis. Proceedings of the IEEE, 110(6), 754–806. https://doi.org/10.1109/JPROC.2022.3171691
Cheng, G., Lin, Y., Abur, A., Gómez-Expósito, A., & Wu, W. (2024). A Survey of Power System State Estimation Using Multiple Data Sources: PMUs, SCADA, AMI, and Beyond. IEEE Transactions on Smart Grid, 15(1), 1129–1151. https://doi.org/10.1109/TSG.2023.3286401
de Matos Sá, M., Correia da Fonseca, F. X., Amaral, L., & Castro, R. (2024). Optimising O&M scheduling in offshore wind farms considering weather forecast uncertainty and wake losses. Ocean Engineering, 301, 117518. https://doi.org/10.1016/j.oceaneng.2024.117518
Díaz, H., & Guedes Soares, C. (2020). Review of the current status, technology and future trends of offshore wind farms. Ocean Engineering, 209, 107381. https://doi.org/10.1016/j.oceaneng.2020.107381
Gao, Z., & Odgaard, P. (2023). Real-time monitoring, fault prediction and health management for offshore wind turbine systems. Renewable Energy, 218, 119258. https://doi.org/10.1016/j.renene.2023.119258
Jin, X., Xu, Z., & Qiao, W. (2021). Condition Monitoring of Wind Turbine Generators Using SCADA Data Analysis. IEEE Transactions on Sustainable Energy, 12(1), 202–210. https://doi.org/10.1109/TSTE.2020.2989220
Khurana, D., Koli, A., Khatter, K., & Singh, S. (2023). Natural language processing: State of the art, current trends and challenges. Multimedia Tools and Applications, 82(3), 3713–3744. https://doi.org/10.1007/s11042-022-13428-4
Li, H., & Guedes Soares, C. (2022). Assessment of failure rates and reliability of floating offshore wind turbines. Reliability Engineering & System Safety, 228, 108777. https://doi.org/10.1016/j.ress.2022.108777
Li, M., Jiang, X., Carroll, J., & Negenborn, R. R. (2022). A multi-objective maintenance strategy optimization framework for offshore wind farms considering uncertainty. Applied Energy, 321, 119284. https://doi.org/10.1016/j.apenergy.2022.119284
Li, Y.-F., Wang, H., & Sun, M. (2024). ChatGPT-like large-scale foundation models for prognostics and health management: A survey and roadmaps. Reliability Engineering & System Safety, 243, 109850. https://doi.org/10.1016/j.ress.2023.109850
Morrison, R., Liu, X., & Lin, Z. (2022). Anomaly detection in wind turbine SCADA data for power curve cleaning. Renewable Energy, 184, 473–486. https://doi.org/10.1016/j.renene.2021.11.118
Si, G., Xia, T., Gebraeel, N., Wang, D., Pan, E., & Xi, L. (2025). Holistic opportunistic maintenance scheduling and routing for offshore wind farms. Renewable and Sustainable Energy Reviews, 207, 114991. https://doi.org/10.1016/j.rser.2024.114991
Stetco, A., Dinmohammadi, F., Zhao, X., Robu, V., Flynn, D., Barnes, M., Keane, J., & Nenadic, G. (2019). Machine learning methods for wind turbine condition monitoring: A review. Renewable Energy, 133, 620–635. https://doi.org/10.1016/j.renene.2018.10.047
Sun, Y., Li, H., Sun, L., & Guedes Soares, C. (2023). Failure Analysis of Floating Offshore Wind Turbines with Correlated Failures. Reliability Engineering & System Safety, 238, 109485. https://doi.org/10.1016/j.ress.2023.109485
Wilkie, D., & Galasso, C. (2020). A probabilistic framework for offshore wind turbine loss assessment. Renewable Energy, 147, 1772–1783. https://doi.org/10.1016/j.renene.2019.09.043
Xiang, L., Yang, X., Hu, A., Su, H., & Wang, P. (2022). Condition monitoring and anomaly detection of wind turbine based on cascaded and bidirectional deep learning networks. Applied Energy, 305, 117925. https://doi.org/10.1016/j.apenergy.2021.117925
Zhang, C., Zeng, Q., Dui, H., Chen, R., & Wang, S. (2025). Reliability model and maintenance cost optimization of wind-photovoltaic hybrid power systems. Reliability Engineering & System Safety, 255, 110673. https://doi.org/10.1016/j.ress.2024.110673
Zhang, W., Lin, Z., & Liu, X. (2022). Short-term offshore wind power forecasting—A hybrid model based on Discrete Wavelet Transform (DWT), Seasonal Autoregressive Integrated Moving Average (SARIMA), and deep-learning-based Long Short-Term Memory (LSTM). Renewable Energy, 185, 611–628. https://doi.org/10.1016/j.renene.2021.12.100
Zhong, S., Pantelous, A. A., Goh, M., & Zhou, J. (2019). A reliability-and-cost-based fuzzy approach to optimize preventive maintenance scheduling for offshore wind farms. Mechanical Systems and Signal Processing, 124, 643–663. https://doi.org/10.1016/j.ymssp.2019.02.012		AI for Sustainable Energy and BuildingsHuanhuan Li
Wildfire Detection using Aerial Swarms Wildfires destroy over 12 million hectares of forest annually, nearly half the size of the UK, causing severe ecological and economic damage. Current monitoring methods lack the spatial and temporal resolution required for early detection and rapid situational awareness.

This PhD project focuses on developing advanced algorithms for collective perception and prediction in a distributed UAV swarm to enable real-time wildfire monitoring and forecasting. You will explore how a swarm of UAVs can autonomously sense, communicate, and reason about wildfire dynamics to produce actionable information. The project will specifically focus on fusion of distributed observations and interaction with a UAV swarm. The goal is to enable a swarm to jointly detect, localise, and map fire fronts under uncertainty and partial observability. Key algorithmic approaches include Probabilistic State Estimation: Implement distributed filtering methods for fusing noisy observations from multiple UAVs; Collective Perception Algorithms: Design decentralized swarm algorithms that allow the robots to share local detections and collectively estimate fire boundaries in real-time; Multi-Robot Coordination: Apply consensus algorithms and belief propagation to ensure the swarm maintains a coherent, up-to-date situational map despite intermittent communication; Adaptive Sensing Strategies: Use reinforcement learning or adaptive control to allocate the robots' attention to areas of high uncertainty or predicted fire growth.

While the focus of this research will be on developing algorithms using existing data in simulation, the student will have access to multiple small and large aerial platforms to conduct experiments, if interested. There will be the opportunity to participate in networking events and collaborate with a large consortium of world-leading researchers across the UK and US, and if interested, gain hands-on experience with UAVs being developed for wildfire detection and suppression.		AI for the Natural EnvironmentMohammad Soorati

Contact Us

Email: sustai@soton.ac.uk

GET IN TOUCH
MORE INFORMATION

Scroll to Top