Periodic Reporting for period 2 - TSCALE (MULTI-SCALE MODELLING FOR TURBOMACHINERY FLOWS USING HIGH-FIDELITY COMPUTATIONAL DATA)
Berichtszeitraum: 2024-04-01 bis 2025-03-31
The challenge of decarbonising power generation and aviation is vast. While the future energy systems are likely to be reliant on renewable energy sources, a suite of low carbon solutions are needed to absorb the intermittency and fluctuations in wind and solar power generation. Gas turbines are uniquely positioned to complement the energy mix due to their high-power density and flexibility in operation. The operation of gas turbines for power generation will also shift towards more fuel flexibility, on-demand operation and efficient part-load operation. These requirements impose additional constraints on gas turbine design and require better consideration of its operation at off-design conditions which historically were not as critical for power generation operation.
Current design methods rely on semi-analytical and experimental correlations to predict losses and deviation angles which are critical for gas turbine operation. However, these methods are not reliable for evaluating compressor designs outside the well explored design space. In such cases, more advanced modelling techniques such as (U)RANS can be used, but even they are known to have problems accurately predicting losses and deviation angles at high incidences. This is exacerbated in multi-stage environments where single blade-row inaccuracies compound. Further still, current design trends often render methods such as URANS inappropriate due to a lack of the spectral gap between deterministic and stochastic unsteadiness.
Experiments and high-fidelity simulations are commonly used to inform design tools at off-design conditions, but often assume simplified inflow conditions. In contrast, real multi-stage compressors exhibit complex unsteadiness due to interactions and accumulation of wakes and turbulence from upstream passages (see Figure 1). Capturing this requires high-fidelity data across a wide range of representative conditions. However, due to the computational cost, such simulations are impractical for routine design use. Therefore, it is essential to extract meaningful physical insights from the large datasets generated.
This project addresses that gap by applying a data-driven flow decomposition framework to turbomachinery flows. The methodology enables the study of inter-scale energy transfers and the role of various dynamics in turbulence production, dissipation, and energy redistribution within the cascade. By combining multiple high-fidelity datasets representative of engine-relevant conditions, the project aims to define a new modelling paradigm that captures the full 3D unsteady flowfield. The ultimate goal is to develop a set of low-order models, each targeting specific flow scales, to better represent compressor flow physics. These results are expected to offer critical insights into multi-scale interactions in complex flows such as those in aero-engines.
Current design methods rely on semi-analytical and experimental correlations to predict losses and deviation angles which are critical for gas turbine operation. However, these methods are not reliable for evaluating compressor designs outside the well explored design space. In such cases, more advanced modelling techniques such as (U)RANS can be used, but even they are known to have problems accurately predicting losses and deviation angles at high incidences. This is exacerbated in multi-stage environments where single blade-row inaccuracies compound. Further still, current design trends often render methods such as URANS inappropriate due to a lack of the spectral gap between deterministic and stochastic unsteadiness.
Experiments and high-fidelity simulations are commonly used to inform design tools at off-design conditions, but often assume simplified inflow conditions. In contrast, real multi-stage compressors exhibit complex unsteadiness due to interactions and accumulation of wakes and turbulence from upstream passages (see Figure 1). Capturing this requires high-fidelity data across a wide range of representative conditions. However, due to the computational cost, such simulations are impractical for routine design use. Therefore, it is essential to extract meaningful physical insights from the large datasets generated.
This project addresses that gap by applying a data-driven flow decomposition framework to turbomachinery flows. The methodology enables the study of inter-scale energy transfers and the role of various dynamics in turbulence production, dissipation, and energy redistribution within the cascade. By combining multiple high-fidelity datasets representative of engine-relevant conditions, the project aims to define a new modelling paradigm that captures the full 3D unsteady flowfield. The ultimate goal is to develop a set of low-order models, each targeting specific flow scales, to better represent compressor flow physics. These results are expected to offer critical insights into multi-scale interactions in complex flows such as those in aero-engines.
This report summarises the work performed throughout the TSCALE project and outlines the main results achieved over the full duration of the fellowship (April 2022 – March 2025). Despite initial delays related to the Australian visa, which led to the fellowship starting at the University of Genova, the project made substantial progress and largely met its objectives across all work packages.
At its core, TSCALE focused on generating high-fidelity datasets for turbomachinery flows and applying advanced data-driven techniques for flow decomposition and modelling. Over 6 million CPU-hours and 300,000 GPU-hours were secured across HPC platforms in Europe, Australia, and the US, resulting in 52 datasets (18 highly resolved) totalling over 70TB. These included simulations under steady inflow, off-design (Fig. 2a, b), and multi-stage compressor (Fig. 3) conditions.
In parallel, the project successfully developed and implemented multiple post-processing tools. These included analytical double/triple decomposition methods, Fourier and Proper Orthogonal Decomposition (POD) techniques, and a statistical framework for scale separation based on the computation of triadic interactions. This enabled, for the first time, large-scale quantification of triadic energy exchanges in turbomachinery, that may form the basis for two innovative modelling strategies: a multi-scale transport model and a Galerkin-projected low-order model based on data-derived triadic dynamics.
Dissemination was broad and effective: 5 major publications (4 journal, 1 proceedings), 2 best paper nominations (ASME TurboExpo), over 25 presentations, including 11 invited talks, 2 industry collaborations, and multiple outreach activities. A dedicated project website (turbscale.com) supported further engagement.
The TSCALE datasets are being curated for broader reuse, with efforts underway in collaboration with Sorbonne University and the COMPOSE MSCA (Horizon EU, Grant n. 101205669) to establish a standardized, non-proprietary high-fidelity database for the fluid dynamics community.
Overall, TSCALE has advanced the state of the art in turbomachinery modelling, built lasting academic and industrial collaborations, and had a strong impact on the ER’s scientific career.
At its core, TSCALE focused on generating high-fidelity datasets for turbomachinery flows and applying advanced data-driven techniques for flow decomposition and modelling. Over 6 million CPU-hours and 300,000 GPU-hours were secured across HPC platforms in Europe, Australia, and the US, resulting in 52 datasets (18 highly resolved) totalling over 70TB. These included simulations under steady inflow, off-design (Fig. 2a, b), and multi-stage compressor (Fig. 3) conditions.
In parallel, the project successfully developed and implemented multiple post-processing tools. These included analytical double/triple decomposition methods, Fourier and Proper Orthogonal Decomposition (POD) techniques, and a statistical framework for scale separation based on the computation of triadic interactions. This enabled, for the first time, large-scale quantification of triadic energy exchanges in turbomachinery, that may form the basis for two innovative modelling strategies: a multi-scale transport model and a Galerkin-projected low-order model based on data-derived triadic dynamics.
Dissemination was broad and effective: 5 major publications (4 journal, 1 proceedings), 2 best paper nominations (ASME TurboExpo), over 25 presentations, including 11 invited talks, 2 industry collaborations, and multiple outreach activities. A dedicated project website (turbscale.com) supported further engagement.
The TSCALE datasets are being curated for broader reuse, with efforts underway in collaboration with Sorbonne University and the COMPOSE MSCA (Horizon EU, Grant n. 101205669) to establish a standardized, non-proprietary high-fidelity database for the fluid dynamics community.
Overall, TSCALE has advanced the state of the art in turbomachinery modelling, built lasting academic and industrial collaborations, and had a strong impact on the ER’s scientific career.
The TSCALE project has achieved notable progress beyond the state of the art in high-fidelity turbomachinery simulations and data-driven modelling. A key outcome was the identification of limitations in the commonly used aerodynamic loss coefficient, which was found unreliable under high turbulence and Reynolds number conditions. This challenges established validation practices and has important implications for both experiments and simulations.
Another major advancement was the development of a novel data-driven framework to analyze inter-scale energy transfers in compressors. This led to the first large-scale quantification of triadic interactions in multi-stage compressors and laid the groundwork for two promising modelling approaches: a generalized turbulence transport model and a reduced-order ODE-based formulation.
Project potential impacts:
Scientific impact: The methodologies and datasets produced have resulted in multiple journal publications, with further work pending. The data decomposition and modelling tools developed will serve as a reference for future turbulence research.
Industrial relevance: The findings directly address the growing need for accurate loss prediction and robust analysis tools for turbomachinery components for more fuel-flexible and load-responsive gas turbines. Engagements with industries have also validated the project's industrial relevance.
Societal implications: The shift towards low-carbon and renewable energy sources requires thermal systems (such as gas turbines) to operate under more varied and less predictable conditions. The TSCALE project contributes to this transition by improving the tools available for compressor data analysis, thus enabling the possibility to design more efficient and adaptable energy systems.
Another major advancement was the development of a novel data-driven framework to analyze inter-scale energy transfers in compressors. This led to the first large-scale quantification of triadic interactions in multi-stage compressors and laid the groundwork for two promising modelling approaches: a generalized turbulence transport model and a reduced-order ODE-based formulation.
Project potential impacts:
Scientific impact: The methodologies and datasets produced have resulted in multiple journal publications, with further work pending. The data decomposition and modelling tools developed will serve as a reference for future turbulence research.
Industrial relevance: The findings directly address the growing need for accurate loss prediction and robust analysis tools for turbomachinery components for more fuel-flexible and load-responsive gas turbines. Engagements with industries have also validated the project's industrial relevance.
Societal implications: The shift towards low-carbon and renewable energy sources requires thermal systems (such as gas turbines) to operate under more varied and less predictable conditions. The TSCALE project contributes to this transition by improving the tools available for compressor data analysis, thus enabling the possibility to design more efficient and adaptable energy systems.