Modal Decomposition-Based Post-Processing Techniques of High-Fidelity Data for Detailed Loss Characterization in Gas Turbines for Marine Applications

The continuous demand for higher efficiency and reduced emissions in marine propulsion systems is driving renewed interest in gas turbine–based solutions, particularly in comparison with conventional marine internal combustion engines. Achieving competitive performance levels requires a detailed understanding of the complex loss mechanisms that limit the efficiency of turbomachinery components and a clear identification of the flow phenomena on which design improvements should be focused. In this context, high-fidelity numerical simulations represent a key tool, as they provide access to the unsteady, multiscale dynamics governing loss generation, which cannot be fully captured by lower-order modeling approaches. In this work, the low-pressure turbine (LPT) is adopted as a representative case study, due to its central role in determining the overall efficiency and operability of gas turbine systems, particularly in marine propulsion applications. Unsteady Large Eddy Simulations (LES) performed on a dataset of six different LPT stage geometries are exploited to construct a comprehensive and generalizable framework for the characterization, localization, and prediction of total pressure losses. Specifically, an advanced post-processing procedure is developed to analyze loss-related quantities through modal decomposition techniques, including Proper Orthogonal Decomposition (POD), Spectral Proper Orthogonal Decomposition (SPOD), and Phase-Locked Proper Orthogonal Decomposition (PPOD). These methods enable the spatial, spectral, and phase-resolved characterization of loss sources, highlighting the role of unsteady interactions between upstream wakes, large-scale coherent structures, and boundary-layer dynamics. The analysis reveals that differences in aerodynamic performance across the investigated geometries can be primarily attributed to variations in blade loading distribution, which governs the location and intensity of wake–boundary layer interactions. Based on these findings, a reduced-order predictive model is constructed by parameterizing blade loading and loss fields through POD-based representations and linking them via simple regression models. Despite the limited size of the high-fidelity dataset, the model demonstrates good predictive capability and allows the identification of an optimal blade loading distribution associated with a reduction in global losses. Overall, the results demonstrate the potential of combining high-fidelity simulations, modal analysis, and data-driven modeling to support the design of more efficient turbomachinery components. The proposed methodology provides a physically grounded and computationally efficient framework to understand, predict, and mitigate losses in a generic gas turbine component, with direct implications for the improvement of gas turbine performance in marine propulsion applications.