Periodic Reporting for period 1 - ENCODING (ENabling sustainable COmbustion technologies using hybrid physics-based Data-driven modelING)
Reporting period: 2023-01-01 to 2024-12-31
The global transition towards carbon neutrality by 2050 necessitates innovative solutions for decarbonising energy-intensive industries (EIIs). The ENCODING project (Enabling Sustainable Combustion Technologies using Hybrid Physics-Based, Data-Driven Modelling) addresses this challenge by developing cutting-edge technologies to facilitate the adoption of Renewable Synthetic Fuels (RSFs) in industrial applications. The ENCODING project is founded on interdisciplinary research in combustion science, fluid dynamics, and machine learning, aiming to create a robust digital combustion infrastructure that ensures stable, efficient, and low-emission industrial energy conversion.
ENCODING addresses these challenges by training ten Doctoral Candidates (DCs) to become future leaders in sustainable combustion technologies. These researchers focus on experimental methodologies and computation essential for advancing RSF-based combustion systems. By integrating fundamental combustion science with digital modelling, ENCODING aims to:
1. Develop predictive tools capable of assessing the impact of RSFs on combustion processes in large-scale industrial furnaces.
2. Ensure stable combustion with near-zero emissions through advanced hybrid modelling techniques.
3. Enhance the overall efficiency of industrial heat generation while reducing environmental footprints.
The project's unique approach integrates physics-based and machine learning-driven models to optimise fuel-flexible combustion systems, providing industry stakeholders with reliable, computationally efficient solutions to navigate the transition to sustainable energy sources.
ENCODING addresses these challenges by training ten Doctoral Candidates (DCs) to become future leaders in sustainable combustion technologies. These researchers focus on experimental methodologies and computation essential for advancing RSF-based combustion systems. By integrating fundamental combustion science with digital modelling, ENCODING aims to:
1. Develop predictive tools capable of assessing the impact of RSFs on combustion processes in large-scale industrial furnaces.
2. Ensure stable combustion with near-zero emissions through advanced hybrid modelling techniques.
3. Enhance the overall efficiency of industrial heat generation while reducing environmental footprints.
The project's unique approach integrates physics-based and machine learning-driven models to optimise fuel-flexible combustion systems, providing industry stakeholders with reliable, computationally efficient solutions to navigate the transition to sustainable energy sources.
The project has made significant progress in advancing the understanding and implementation of Renewable and Sustainable Fuels (RSFs) for cleaner and more efficient combustion. Through a combination of experimental campaigns, high-fidelity simulations, data-driven approaches, and digital twin development, the project is generating valuable insights to support the transition towards low-carbon energy solutions.
Our experimental investigation has identified:
- The optimal conditions for ammonia-hydrogen combustion in industrial and laboratory-scale setups.
- The fuel flexibility, combustion stability, and emissions in MILD combustion conditions.
We have developed advanced computational techniques to complement experimental data, in particular:
- Hybrid simulation models combine physics-based models and machine learning to reduce computational costs while maintaining accuracy.
- Direct Numerical Simulations (DNS) to explore turbulence-flame interactions and the impact of hydrogen enrichment on pollutant formation.
Researchers have developed novel physics-based, unsupervised learning techniques, namely:
- The application of low-cost Singular Value Decomposition (lcSVD) and Higher-Order Dynamic Mode Decomposition (HODMD) for analysing massive combustion data.
- The development of Soft Local Reduced-Order Models (SL-ROMs) to optimise the development of ROMs.
In the field of Digital Twins, the project has focused on integrating heterogeneous data streams (experimental and numerical, with different fidelity) to update digital virtual replicas of complex combustion systems continuously.
Our experimental investigation has identified:
- The optimal conditions for ammonia-hydrogen combustion in industrial and laboratory-scale setups.
- The fuel flexibility, combustion stability, and emissions in MILD combustion conditions.
We have developed advanced computational techniques to complement experimental data, in particular:
- Hybrid simulation models combine physics-based models and machine learning to reduce computational costs while maintaining accuracy.
- Direct Numerical Simulations (DNS) to explore turbulence-flame interactions and the impact of hydrogen enrichment on pollutant formation.
Researchers have developed novel physics-based, unsupervised learning techniques, namely:
- The application of low-cost Singular Value Decomposition (lcSVD) and Higher-Order Dynamic Mode Decomposition (HODMD) for analysing massive combustion data.
- The development of Soft Local Reduced-Order Models (SL-ROMs) to optimise the development of ROMs.
In the field of Digital Twins, the project has focused on integrating heterogeneous data streams (experimental and numerical, with different fidelity) to update digital virtual replicas of complex combustion systems continuously.
Our project has achieved significant progress in the science of combustion, leveraging data analytics, machine learning, and digital twin technology to pave the way for cleaner, more efficient industrial combustion systems. These advancements are crucial for the widespread adoption of renewable synthetic fuels (RSFs) like ammonia and hydrogen.
WP1 - We've conducted detailed experiments to understand how ammonia and hydrogen burn under specific conditions known as Moderate or Intense Low-oxygen Dilution (MILD). By identifying the ideal fuel mixtures, we've found ways to balance stable combustion and minimal emissions. This research will directly inform the design of improved burners and combustion strategies for industrial use, as well as contribute to the development of regulations that support the adoption of ammonia as a clean fuel.
WP2 - We've employed machine learning techniques to make combustion simulations faster and more accurate. By training neural networks with high-quality data from Direct Numerical Simulations (DNS), we've significantly enhanced our ability to model combustion processes. We've also validated that MILD combustion effectively reduces harmful emissions and developed efficient models that predict emissions without requiring vast computational resources.
WP3 - We've developed a new method called low-cost Singular Value Decomposition (lcSVD) to analyze combustion data in real time. This allows us to model complex combustion processes using live data streams. Additionally, advanced feature extraction techniques help us identify critical combustion patterns, enabling faster data processing without sacrificing accuracy.
WP4 - Our project has focused on creating more accurate closure models using neural networks to improve predictions for renewable synthetic fuels. These virtual models enable the optimisation of combustion processes, leading to significant reductions in emissions. These models will be invaluable tools for industries seeking to enhance efficiency and minimise environmental impact.
WP5 - We've successfully developed digital twin technology to predict emissions, temperature, and combustion behaviour in real-time. We've also improved sensor strategies, allowing for more accurate combustion measurements with fewer sensors. These real-time digital twins will enhance efficiency, safety, and sustainability in industrial combustion systems while lowering costs and reducing pollution.
WP1 - We've conducted detailed experiments to understand how ammonia and hydrogen burn under specific conditions known as Moderate or Intense Low-oxygen Dilution (MILD). By identifying the ideal fuel mixtures, we've found ways to balance stable combustion and minimal emissions. This research will directly inform the design of improved burners and combustion strategies for industrial use, as well as contribute to the development of regulations that support the adoption of ammonia as a clean fuel.
WP2 - We've employed machine learning techniques to make combustion simulations faster and more accurate. By training neural networks with high-quality data from Direct Numerical Simulations (DNS), we've significantly enhanced our ability to model combustion processes. We've also validated that MILD combustion effectively reduces harmful emissions and developed efficient models that predict emissions without requiring vast computational resources.
WP3 - We've developed a new method called low-cost Singular Value Decomposition (lcSVD) to analyze combustion data in real time. This allows us to model complex combustion processes using live data streams. Additionally, advanced feature extraction techniques help us identify critical combustion patterns, enabling faster data processing without sacrificing accuracy.
WP4 - Our project has focused on creating more accurate closure models using neural networks to improve predictions for renewable synthetic fuels. These virtual models enable the optimisation of combustion processes, leading to significant reductions in emissions. These models will be invaluable tools for industries seeking to enhance efficiency and minimise environmental impact.
WP5 - We've successfully developed digital twin technology to predict emissions, temperature, and combustion behaviour in real-time. We've also improved sensor strategies, allowing for more accurate combustion measurements with fewer sensors. These real-time digital twins will enhance efficiency, safety, and sustainability in industrial combustion systems while lowering costs and reducing pollution.