Safe reinforcement learning-based V2X operation of EV fleets for demand-side flexibility | |
The global electric vehicle (EV) fleet is projected to reach 145 million units by 2030, posing new threats to the reliability of the power system. However, EVs can also play a key role as a source of demand-side flexibility to support the system in managing uncertainty resulting from the integration of renewable energy resources. The onsite coupling of photovoltaics (PVs), battery energy storage systems (BESS) and EV fleets with vehicle-to-grid (V2G) technology has shown promising performance in terms of demand-side flexibility provision. | |
Safe deep reinforcement learning for building control | |
Buildings are significant energy consumers, primarily due to the operation of heating, ventilation, and air conditioning (HVAC) systems. Effective control of such systems is crucial for enhancing overall energy efficiency. Typically, traditional rule-based controllers are used due to their affordability and interpretability. However, as complexity increases, these controllers suffer from non-optimal performance and limited scalability. Recent advancements in Deep Reinforcement Learning (DRL) provide a data-driven alternative, demonstrating promising control performance without the need for explicit system modeling. Despite these advantages, conventional DRL approaches often fail to account for specific operational constraints present in HVAC systems. One critical constraint is the requirement for smooth control actions with a limited number of on-off switches, as frequent switching can lead to faster deterioration of the controlled systems. Therefore, it is imperative to develop data-driven control strategies that not only optimize energy consumption but also adhere to these operational constraints. This study, part of the Euthermo Project, aims to develop safe reinforcement learning algorithms for building climate control. | |
Multi Agent Deep Reinforcement Learning for Building Control | |
Energy consumption in buildings is a critical concern, primarily driven by the operation of heating, ventilation, and air conditioning (HVAC) systems, lighting, and other appliances. Efficient control of these systems is paramount for achieving significant energy savings and reducing environmental impact. Traditional rulebased controllers, while cost-effective and easy to implement, often fail to provide optimal performance and lack scalability as system complexity grows. Recent advancements in Deep Reinforcement Learning (DRL) offer a powerful, data-driven alternative. DRL has shown promising results in optimizing control performance without the need for explicit system modeling. However, the complexity of managing multiple interdependent control variables within a building remains a challenge. For instance, the heating control of individual rooms can influence each other, and shading controls can affect both heating and cooling demands. | |
Control strategy development for energy exchange in a decentralized heating systems | |
The goal of this project is 1) to explore potential control strategies to manage the energy exchange for a decentralized, bidirectional district heating network and 2) to implement the most promising controller, in terms of complexity and effectiveness, in a simulation environment | |
Development of direct ink writing methods of liquid metal for soft electromagnet | |
Liquid metal is one of the conductor, maintaining liquid state. This unique characteristics is useful for soft wearable devices, especially soft actuators and sensors owing to their intrinsic softness and self-healing capability [1]. In the previous project, we successfully developed liquid metal pattern for soft electromagnet utilizing direct ink writing (DIW) method (Figure 1) [2]. It allows 300 µm width & coiling pitches. This fine width and pitches allows to generate strong magnetic field. However, current DIW is not reliable. This is because of the printing substrate, silicone elastomer. Its high hydrophobicity makes difficult to adhere the liquid metal to substrate. | |
Development of bistable soft actuator consisting of stretchable magnet and liquid metal based electromagnet | |
Liquid metal is one of the conductor, maintaining liquid state. This unique characteristics is useful for soft wearable devices, especially soft actuators and sensors owing to their intrinsic softness and self-healing capability. In the previous project, we successfully developed liquid metal based soft electromagnet for bistable soft actuators [1]. Soft electromagnet consists of EGaIn (eutectic gallium-indium) as for liquid metal in spiral pattern covered by silicone elastomer. What has been lacking towards the bistable soft actuators is that stretchable permanent magnet on the side of soft electromagnet. It composed of Neodymium powder and soft elastomer. It creates compression force. | |
Flexibility Potential Quantification of Prosumers: How to integrate Users’ Behavior? | |
In recent years, the penetration of renewable energy resources in distribution grids has been steadily increasing, raising new issues such as voltage violations or line congestions. Due to their large thermal inertia, individual buildings can regulate their heating system to support distribution system operation. In our previous work, we proposed a quantification of the flexibility potential of an electric heating system, using the concept of energy flexibility envelopes, and accounting for the impact of various uncertainties: the weather forecast, the building thermal model inaccuracy, and the uncertain inhabitants’ behavior. However, we considered that uncertainties are independent of the requested flexibility. Yet, in practice, the inhabitants’ behavior is correlated to flexibility requests as optimal control strategies. For example, a request to shift the consumption may increase the room temperature, which in turn impacts the inhabitant behaviors, possibly reducing energy efficiency. | |
How robust is data-driven controller in real implementation? (ETH/Empa) | |
Data-driven controllers' performance can be compromised for many reasons, such as corrupt data. This can have significant implications, especially in systems that rely on accurate data for decision-making. In this project, we will focus on the robustness of data-driven controllers for applications in buildings. | |
Do data-driven control decisions make sense? (ETH/Empa) | |
Data has been the centered of novel control and decision making process. System control interpretability is essential for user trust. This project will investigate the interpretability of in-house data-driven controller decisions. The objective is to improve transparency, thus facilitating their application in managing energy systems. By focusing on transparent control mechanisms, this research aims to bridge the gap between complex control strategies and user trust. | |
Fatigue and static strength of WAAM joints under tension and bending | |
Recent developments in Wire-Arc Additive Manufacturing (WAAM) technology have created new possibilities within the construction industry. These developments stem from the capacity of this technology to attain free-form and optimal design configurations while maintaining a relatively high manufacturing speed, resulting in cost and material savings. WAAM can be used either for fabrication of new components or application to existing structures for strengthening or repair purposes. In any case, ensuring the structural integrity of WAAM components is an essential parameter in their applications at an industrial scale. Joint areas are accounted as one of the critical details in WAAM components. The joints can exist between WAAM depositions or be-tween WAAM layers and metallic profiles, depending on the application, e.g. hollow-section joints. These areas can introduce complexities for printing strategies as well as predictions regarding static and fatigue behavior. This study aims at numerically and experimentally investigating the static and fatigue behavior of WAAM joints subjected to tensile and bending loads. | |
Post-doctoral Fellow or Research Scientist Positions: Earth Commission Novel Entities Group | |
In collaboration with the Earth Commission, the University of Toronto School of the Environment is excited to announce opportunities two Post-doctoral Fellows (PDFs) or Research Scientists to join the core research team focussed on assessing “Safe and Just Earth System Boundaries” for Novel Entities (ESBs). The following are the areas that will be addressed by the PDFs, with exact subject areas to be divided by the two PDFs based on expertise and interest of each candidate: • Scan of methods that could be used to determine ESBs for a subset of novel entities. The methods that could be used include material flow analysis, life cycle assessment and other types of impact assessment methods. • Undertake “horizon scanning” of the literature to document present and future novel entities that are or could transgress either a Safe or Just ESB. This scan will add the broad context to the case studies of specific novel entities. • Undertake case studies of several specific novel entities to develop Safe (biophysically-based) and Just (according to interspecies, intra-societal and intergenerational equity) ESBs. • Consider transformation pathways that moves society from our current situation to within safe and just ESBs. This analysis could consider novel entities in broad categories and could also focus in on the case studies undertaken for the ESB evaluation. | |
Optimal inspection and intervention planning for maximizing life and minimizing environmental impact of road infrastructure | |
Optimal inspection and maintenance planning of road infrastructure is a challenging process, since decision-making needs to be made on the basis of uncertain observations of various origin. These uncertainties are linked to some level of uncertainty relating to varying environmental and loading conditions, modelling errors, inefficiency of the measuring system and so on. Although this issue spans over several disciplines, including social sciences, it is also quite commonly met within an engineering framework. The goal of this project is to build an optimal inspection/maintenance framework for different bridge materials and typologies, maximizing their service life and minimizing environmental im-pact, and address the challenges of infrastructure management. Thus, the student will work on some of the following objectives depending on their interest: • Review established and state-of-the-art non-destructive-testing and rehabilitation methods and derive indications on which methods work better for which different bridge typologies/materials • Utilize existing finite element models (in SoFiSTiK) of various bridge types to simulate various short-and long-term effects. • Explore decision processes frameworks for modeling decision making in situations with partly random or decision-dependent outcomes. Suggested Courses: Depending on interest: Method of Finite Elements (or similar) and/or Non Destructive Evaluation & Rehabilitation of Existing Structures Suggested Competencies: Basic knowledge on MATLAB or Python Possible to select as: obligatorishce Projektarbeit, praktische Projektarbeit, forschungsbezogene Projektarbeit Limitation of offerings: 2 students (group work is possible) | |
Automatic Segmentation of Nanoparticles in Scanning Transmission Electron Microscopy Images by Deep Learning | |
The Electron Microscopy Center at Empa Dübendorf is offering an exciting project that will provide the candidate with hands-on experience synthesizing catalyst model systems, operating an aberration-corrected transmission electron microscope, generating synthetic data, and training supervised machine learning models. | |
Identification of optimal technological assets for prosumer communities in the Swiss scenario | |
Prosumer communities leverage and share locally generated energy. In response to the Swiss Energy Strategy 2050 and potential gas supply vulnerabilities, transitioning from singular, gas/oil-reliant energy systems to multi-energy networks is imperative. Unlike conventional individual systems where devices are often oversized and underutilized due to being designed for demand peaks, a prosumer community can optimize device use, reducing both energy costs, investment overheads and ecological impact. Therefore, when properly designed and operated, prosumer communities mitigate the inefficiencies typical of traditional setups. However, achieving optimal design is challenging due to complex interactions among stakeholders. The economic and ecological aspects of prosumer communities need to be thoroughly investigated taking into account different scenarios, energy demands and building types. This project aims to investigate various case studies for prosumer communities in the Swiss scenario and explore the technological solutions that mostly benefit from energy sharing. | |
Time Resolved Photoluminescence Spectroscopy and Simulations | |
We are looking for a motivated Master Thesis student to join the Nanomaterials Spectroscopy and Imaging Group at Empa, located in the vibrant city of Zurich, Switzerland. Investigate carrier dynamics of cutting-edge optoelectronic materials, including thin films and 2D materials. Simulate Time Resolved Photoluminescence (TRPL) response. Fit experimental results and extract crucial optoelectronic parameters like diffusion coefficient, carrier concentration, and carrier lifetime. Gain hands-on experience in a dynamic research environment. Collaborate with leading experts in the field. Work on cutting-edge projects with real-world applications. | |
Internship, Semester Thesis, Master Thesis | |
Factors driving uptake of renewable energy systems in Swiss households | |
Development of an ML-driven Mobile App for Quality and Species Analysis of Roundwood | |
Roundwood is sorted by quality and species in the forestry (and in sawmills). Based on the dataset with images of roundwood (from partner sawmill³) containing quality and species labels, a classification model has been developed for predicting the species and the quality of a stem based on the image of its cross-section. The detailed description of the model architecture and the results can be found in the paper: https://www.sciencedirect.com/science/article/pii/S0957417423036217. The task of the project is to develop an iOS or Android App that based on the model results. | |
Enhancing Models for Roundwood Classification via Label Noise Correction | |
The project's goal is to optimize the processes in the wood industry through automation of a roundwood sorting into different quality grades based on image classification. The dataset comprises cross-sectional images of roundwood labelled with quality grades: B, C and D (B standing for the highest, D - the lowest quality) as well as tree species (spruce and fir). The implemented DL models provide 91% accuracy for the species classification task and 80% accuracy for quality classification on the cured data (no label noise). The goal of the thesis is to use the noisy data to improve the quality of the models. | |
Non-Intrusive Load Monitoring and Customer Segmentation assisted demand flexibility provision in Swiss Households | |
Switzerland is committed to transitioning to a renewable energy system. The Swiss government has set a target of achieving net-zero carbon emissions by 2050. This will require a significant increase in the use of renewable energy sources. The Swiss power grid is also vulnerable to imbalances be-tween supply and demand. Demand flexibility can help to mitigate this risk and ensure the reliable operation of the power grid. Demand flexibility is the ability to shift or reduce energy use in response to changes in sup-ply or price. This is becoming increasingly important as the power grid transitions to renewable energy sources, such as solar and wind power, which are intermittent and less predictable. Demand flexibility can help to balance the grid and reduce the need for expensive and polluting backup power plants. Non-Intrusive Load Monitoring (NILM) and customer segmentation modeling are powerful tools that can be used to develop demand flexibility programs. NILM can be used to identify high-energy-consuming appliances and to track their energy usage over time. Customer segmentation modeling can be used to identify different groups of customers based on their energy consumption patterns. This information can then be used to develop targeted demand flexibility programs that are more likely to be effective for each group of customers. |
Powered by SiROP - the academic career network