This study proposes a latent-space framework to support model tuning, where geometrical features and detailed aerodynamic loss information of Low Pressure Turbine (LPT) blades are jointly represented. The model developed here is aimed at associating geometric shape modifications to spatial and modal characterization of losses, providing local analysis of different loss sources. First, Proper Orthogonal Decomposition (POD) is applied to Large Eddy Simulations (LES) of different LPT cascades to analyze local entropy production, identifying different contributions to loss generation due to different coherent flow structures. Then, POD is used to encode both blade geometries and loss distributions, condensing geometricand efficiency information in two reduced-order latent spaces. The corresponding coefficients, obtained by projection on POD modes, allowed the identification of a low-dimensional representation of both predictors and targeted functions. This enables the development of a simple yet efficient model that correlates the compressed representations of geometry and loss distributions, allowing blade shape optimization while maintaining a high degree of interpretability of the results. Despite the simplicity of the model and the limited data available, the application of the POD encoder demonstrates to provide physic-consistent response surface, also showing promising predictive capability once tested on literature blade, the T106A cascade, that did not participate to the tuning phase. The model allows foran easy interrogation of the profile geometry design space, providing related loss mechanisms in different regions of the examined domain.

This study presents a joint experimental and numerical investigation on the impact of endwall ridging on the efficiency of a turbine nozzle guide vane (NGV). Experiments are conducted on a large-scale, low aspect ratio cascade replicating a high-diffusion stage between the high-pressure turbine (HPT) and the low-pressure turbine (LPT). The cascade is tested in its baseline configuration and with two endwall ridging devices, one of which is optimized through parametric computational fluid dynamics (CFD) simulations by varying ridge height, spacing, and position within the blade channel. A Pareto front analysis identifies the optimal configuration that maximizes flow uniformity at the cascade exit while minimizing losses. The flow field is experimentally investigated in a radial-tangential plane at the cascade exit using a five-hole probe to assess total pressure losses and flow angle uniformity, key factors for evaluating carry-over effects on the downstream row and refining CFD predictions. Additionally, CFD streamline visualizations provide further insight into the ridging device operation. The findings demonstrate that the optimized endwall ridging effectively mitigates the cross-stream flow induced by the passage vortex. As a result, flow uniformity is enhanced, leading to increased cascade efficiency compared to the baseline configuration.

The trend towards more compact and efficient low-pressure turbine (LPT) designs can benefit significantly from advanced numerical predictive tools. Faithfully capturing the complex transitional and turbulent nature of unsteady flows in LPTs often necessitates high-fidelity methods such as Large Eddy Simulations (LES) for accurate predictions of turbine efficiency and loss generation. Integrating high-fidelity simulations into design cycles, which are currently predominantly driven by rapid low-fidelity Unsteady Reynolds-Averaged Navier-Stokes (URANS) calculations, requires cutting-edge numerical tools that leverage modern high-performance computing architectures. In Part II of this paper, we present results from a newly established, highly-resolved LES and state -of-the art URANS database for three newly designed LPT profiles. These are a conventional standard lift profile, a front-loaded high -lift profile, and an aft-loaded profile. These profiles are evaluated individually within a repeating 1.5 stage LPT configuration operating under engine -like conditions at an isentropic exit Mach number of 0.3. A Reynolds number sweep, ranging from 70,000 to 320,000, captures a broad spectrum of engine-relevant flow conditions. The LES study incorporates time-resolved, time-averaged, and phase -locked averaged results, enabling a detailed examination of unsteady flow dynamics driven by the rotor blade passing frequency. This analysis provides deep insights into blade-wake interactions, unsteady boundary layer evolution, and loss generation mechanisms across the profiles and Reynolds numbers, highlighting the influence of loading distribution on aerodynamic performance and stage efficiency. On top of that, state-of-the-art URANS of the same configurations are undertaken. While trends are largely recovered, there are important differences with the LES data. These are especially present for the aft-loaded profile, being a radical blade design compared to conventional profiles, as a result of the large flow separation on the blades' suction side which cannot be adequately captured by URANS. This reveals regions where improvements in turbulence and transition modeling are necessary in a mean sense, and in a phase-locked averaged sense, relevant for capturing the by-nature unsteady flow phenomena in LPT stages. By establishing this high-fidelity database, the work advances the understanding of unsteady aerodynamic phenomena in realistic LPT modules and lays the foundation for the development of more accurate predictive models to improve URANS given the cost of LES. The findings contribute to the design of more compact, efficient, and high-performance turbine systems, addressing critical challenges in modern turbomachinery.

The complex transitional and turbulent nature of unsteady flows seen in Low-Pressure Turbines (LPTs) often demands high-order methods such as Large Eddy Simulations (LES) for accurate predictions of turbine efficiency and loss generation. This study presents results from a highly resolved LES and state-of-the art Unsteady Reynolds-Averaged Navier-Stokes (URANS) database for three newly designed LPT profiles. These are a conventional standard lift profile, a front-loaded high-lift profile, and an aft-loaded profile. Each profile is evaluated individually within a repeating 1.5-stage LPT configuration operating under engine-like conditions at an isentropic exit Mach number of 0.3. A Reynolds number sweep, ranging from 70,000 to 320,000, captures a broad spectrum of engine-relevant flow conditions. The LES study incorporates time-resolved, time-averaged, and phase-locked averaged results, enabling a detailed examination of unsteady flow phenomena such as blade-wake interactions, unsteady boundary layer evolution, and loss generation mechanisms. Complementary URANS calculations of the same configurations are undertaken and compared with the LES data. While trends are largely recovered, important differences with the LES data can be identified. These are especially present for the aft-loaded profile, it being a radical blade design compared to conventional profiles, highlighting the necessity for turbulence and transition modelling improvements when considering more aggressive blade designs. Ultimately, this work advances the understanding of unsteady aerodynamic phenomena in LPTs via LES and lays the foundation for the development of more accurate URANS models.

To reduce pollutant emissions, modern aeroengines adopt combustors that work with lean premixed flames. These generate significant flow distortions, and due to the compact engine architecture, combustor-turbine interaction becomes a crucial design aspect. From an industrial perspective, achieving design targets while minimizing time to market requires effective and efficient design tools. This study employs a state-of-the-art in-house CFD solver, extensively validated for combustor-turbine interaction, to investigate the aerodynamics of an engine-representative high-pressure turbine (HPT) stage tested in the DLR NG-Turb facility within the European FACTOR project. The test case consists of a 1.5 stage cooled transonic turbine, with distorted inlet conditions coming from a combustor simulator. In detail, steady/unsteady RANS (Reynolds-Averaged Navier-Stokes) simulations were carried out to analyze two clocking positions between the swirling hot spot and nozzle guide vanes (leading-edge clocking, passage clocking). Numerical setups combined Roe's upwind, central difference, and AUSM+ - up schemes with high-Reynolds Wilcox k-w and Menter k-w SST turbulence models, both in baseline and helicitycorrected formulations. Comparison with experimental data shows that time-accurate simulations improve flow-field predictions downstream of the rotor and that the helicity-based correction can significantly enhance the results. To the best of the author's knowledge, this is the first application of helicity-corrected turbulence models in the context of hot-streak interaction with an aeronautical cooled HPT stage. This work demonstrates that URANS simulations with advanced turbulence closures can effectively estimate the complex aerodynamics of realistic HPT and hot-streaks migration, while ensuring computational requirements that are in line with industrial design practices.

This paper presents the design of a transonic low-pressure turbine (LPT) for the next-generation fighter air-breathing engines. The study focuses on the design of a cascade profile representative of an LPT Nozzle Guide Vane (NGV), that follows conventional literature guidelines for transonic turbine airfoils. The paper reports on the numerical and experimental methods that will be employed for a detailed understanding of the flow physics for this baseline solution. The experimental setup includes a high-speed linear cascade that can be operated at a wide range of inlet turbulence levels (Tu = 5% - 8%) and outlet Mach numbers (M = 0.8 - 1.2). The test section inlet is equipped for hot-wire anemometry measurements, while a purposely designed multi-hole probe is traversed to measure the aerodynamic flow quantities at the cascade outlet. The central passage airfoils feature arrays of pneumatic pressure taps to evaluate the blade loading and hot films to study the status of the boundary layer. Optical sidewalls enable full-field Schlieren and Background-Oriented Schlieren imagery to study cascade shock patterns and unsteady shock-boundary layer interactions. The test section is designed with provision for time-resolved stereo-PIV measurements to cross-validate the cascade velocity field and quantify the turbulence statistics and transport mechanisms through the transonic LPT passage. A detailed planning for high-fidelity flow simulations (LES and DNS) is presented in the second part of the paper. State-of-the-art computational methodologies will be employed along with advanced post-processing techniques including mode decomposition to enhance the understanding of the flow physics, assess the limitations of traditional numerical methods and complement the experimental findings.

The reduction of the leakage flow rate and the optimization of sealing systems is today a key point for improving the performance of modern gas turbines, since the losses due to the interaction process between the main flow and those entering /leaving the cavities is among the most relevant for the case of stator/rotor cavity systems. In the present work, numerical simulations are employed to model and describe the flow behaviour within a typical aeroengine turbine cavity system and to investigate the interaction process between the main flow and the flow entering/exiting from a cavity system representative of LPT for aeroengine applications. Experimental results acquired in a cold-flow facility reproducing one-and-half Low Pressure Turbine axial flow stage equipped with an engine-like cavity system and upstream and downstream rotor rows have been used to validate the simulations. Several turbulence models have been tested to investigate their effect on the flow field into the cavity and on overall performance parameters. Comparison with experimental results will also identify the best model also for future investigations. Moreover, the effect related to the rotor/stator interaction, provoking ingestion into the cavity of the wake generated by the upstream rotor bars and blockage effect related to the presence of the downstream bars will be inspected. To this aim, both steady and unsteady calculations were carried out through the commercial solver Numeca. Several calculations, varying the flow rate of cooling air injected inside the cavity, were carried out to identify the value of the flow rate capable of sealing the cavity, avoiding the ingestion of flow from the main channel. Cavity discharge coefficient and the total pressure loss distribution have been computed for the different sealing flow rate to understand the effect that this parameter has on the stage efficiency and leakage flow through the cavity. Finally, results presented into the work provide a clearer picture on the mechanisms responsible for producing additional losses and how this mechanisms are affected by the sealing flow rate, thus helping the designer in further improvement the LPT stage efficiency.

In this paper novel machine-learnt transition and turbulence models are applied to the prediction of boundary-layer transition and wake-mixing in a Low-Pressure Turbine (LPT) cascade, including unsteady inflow cases with incoming wakes. A Laminar Kinetic Energy (LKE) transport approach is employed as baseline transition framework. The application of the machine learnt EARSMs takes advantage of a zonal strategy based on a newly developed sensing function that allows automated wake demarcation. This work compares the performance of several approaches that are based on the application of improved transition models without machine learnt EARSMs, baseline transition model with EARSMs trained for improved wake mixing, and new transition and turbulence closures that are simultaneously developed in a fully coupled way and aimed at improving both transition and wake mixing predictions in LPTs. The investigation is carried out on a cascade of industrial footprint, representative of modern LPT bladings. First of all, the various modelling frameworks are evaluated, without bar wakes (steady conditions), over a wide range of Reynolds number values. URANS analyses have also been carried out to investigate the predictive capabilities of machine-learnt closures in the presence of incoming bar wakes. RANS/URANS results are scrutinized against LES calculations and experimental data. It will be demonstrated that machine learnt EARSMs are capable of producing realistic wake mixing when simply applied on top of the baseline transition models. However, the most accurate predictions will be shown to be provided by fully integrated machine-learnt transition/turbulence closures.

The LM2500 gas turbine was derived from General Electric TF39 and CF6 aircraft engines and has been a best-selling model for over 40 years. The fourth generation LM2500+G4 introduced in 2005 offers an output greater than 34.6 MW at ISO conditions with a thermal efficiency of more than 39%, supporting a wide range of power generation and mechanical drive Land and Marine applications. A fifth generation LM2500+G5 uprate of this successful product has achieved new levels of output shaft power and efficiency, enabled by a more efficient core and a new Universal Power Turbine (UPT) design. The new power turbine module is briefly described in this document from the basic 1D parameters up to the design of the airfoils, which have high lift characteristics. In addition to the traditional goal of increasing performance, the design of the UPT also focused on achieving robustness and versatility. The UPT is now the best choice for a wide range of applications, with only minor hardware changes needed to adapt it to support the LM2500+, LM2500+G4 and the uprated LM2500+G5 gas generators for both 50Hz and 60Hz applications. The compliance testing campaign for the LM2500+G5 UPT was conducted on ten engines, using development instrumentation to validate overall gas turbine performance in terms of power, efficiency, and emissions, as well as UPT module performance. The test objectives focused on different power ranges, combustor burner modes and mapping, compressor bleeds, and UPT speeds. The test exceeded expectations, demonstrating the overall engine and component performance of the UPT and, providing valuable data that will be used to further improve the design and performance of the gas turbine.

In the present work, Large Eddy Simulations (LES) of a low pressure turbine cascade have been carried out to understand the different sensitivity to incidence angle variation observed under steady and unsteady inflow conditions. Indeed, experimental evidence shows poor sensitivity to incidence angle variation for the steady case in a range of −9° ≤ i ≤ +9°, while losses grow rapidly in the unsteady flow environment. It will be shown that this is due to wake migration into the core flow region, producing an additional source of losses for the unsteady flow case. Both steady and unsteady simulations have been carried out at nominal and two positive and negative incidence angles. The LES data have been inspected to identify the physical reasons behind the different loss trends, after being validated by means of blade loading and loss coefficient distributions. To this end, Proper Orthogonal Decomposition (POD) has been applied to each dataset. POD modes clearly identify the different structures responsible for loss production in the steady and the unsteady operation of the cascade. The structures carried by upstream wakes have been isolated and identified both in the boundary layer and in the core flow. The process of bowing, tilting and reorientation of the wake filaments and related structures is shown to be the dominant mechanism responsible for the strong loss sensitivity to the incidence angle variation observed under unsteady inflow. Otherwise, structures responsible for the transition of the blade boundary layer do not exhibit significant variation for both the steady and the unsteady flow cases, thus further strengthening that the sensitivity to incidence variation is mainly due to wake migration into the core flow region. This loss source is not classically considered in literature correlations and actual design processes.

The effects due to streamwise oriented riblets on the free-stream turbulence (FST) induced transition of a flat plate boundary layer (BL) are studied by means of Particle Image Velocimetry (PIV) and Laser Doppler Velocimetry (LDV). The experiments are performed on a flat plate installed between adjustable endwalls providing an adverse pressure gradient typical of turbine blades for aeronautical applications. Four different ribbed surfaces are compared with a smooth reference case to test the effects of riblet dimension and location on turbulent spot nucleation. Riblets are installed in the laminar, the transitional, and the fully turbulent BL. Their effects are studied for three Reynolds numbers (Re = 70000, 150000, 220000) under fixed pressure gradient and free-stream turbulence intensity (Tu = 5%). Varying the flow Reynolds number leads to different non-dimensional riblet height and spacing values, providing an overall test matrix of 13 different conditions. LDV measurements allowed the computation of the wall shear stress and the BL statistical moments at selected spatial positions. On the other hand, PIV data are used to characterize the riblet effects on turbulent spot nucleation and transition: the higher the non-dimensional riblet height, the more the breakup events occur upstream with respect to the smooth case. Momentum losses significantly increase when the non-dimensional riblet spacing becomes larger than 30 wall-units. Otherwise, beneficial effects are obtained for smaller riblet spacing, regardless of their streamwise extension and position with respect to the BL transition region.

In this work, streamwise oriented riblets were installed on a flat plate exposed to an adverse pressure gradient simulating the diffusing part of a low-pressure turbine (LPT) blade and, successively, on the suction side of an LPT cascade operating under unsteady inflow condition. Different riblets heights, spacings and positions have been tested, and complementary measuring techniques have been used to quantify their effects on the boundary layer evolution, and on losses. Experiments carried out on the flat plate allowed the detailed description of the riblet effects on the coherent structures developing within the boundary layer affecting transition, thus providing a rationale for the identification of the optimal riblet geometry once scaled in wall units. For riblet heights equal to about 20 wall-units, a maximum loss reduction of 8% was observed with respect to the smooth case. Otherwise, for larger riblet dimensions, earlier transition occurs due to enhanced boundary layer instability and losses significantly increase. Interestingly, the streamwise extension of the ribbed surfaces with respect to the transition region was found to play a minor role compared with the riblet dimension. The riblet configurations providing the highest reductions of viscous losses on the flat plate were then tested in the LPT blade cascade for different Reynolds numbers. In this case tests have been carried out with impinging upstream wakes. An overall reduction of the profile losses comparable to that observed in the flat plate case has been confirmed also in the unsteady operation of the turbine cascade. Low sensitivity of the profile losses to the riblet streamwise extension was also observed in the cascade application. This confirms that positive effects in terms of loss reduction can be obtained even when the exact transition position is not known a priori.

Particle image velocimetry measurements have been carried out in a low-pressure turbine cascade operating under unsteady inflow to deeply investigate reduced frequency and flow coefficient effects on flow dynamics, and, consequently, on loss generation in the boundary layer and in the core flow region. Two independent measuring setups have been used for the purpose. The first one captured a large view of the entire blade passage, thus allowing the observation of the incoming wakes and related large-scale vortices developing in the core flow region. The second setup was instead focused on the rear part of the blade suction side to analyze the boundary layer development and to observe the mechanisms dominating the wake-boundary-layer interaction. Tests were performed for four flow cases, varying the reduced frequency and the flow coefficient independently. Proper orthogonal decomposition has been applied to quantify the turbulent kinetic energy production in the core flow, due to wake dilatation and distortion, and in the boundary-layer region. Upstream wake migration and boundary-layer-related losses are consequently quantified from particle image velocimetry data and compared with total pressure measurements for the different combinations of the inflow parameters, providing a clear view of the different loss sources affecting the unsteady operation of low-pressure turbine cascades.

In the present work linear and non-linear regression functions have been tuned with an extensive database describing the unsteady aerodynamic efficiency of low-pressure-turbine cascades. The learning strategy has been first defined using a dataset published in a previous work concerning the loss coefficient measured in a large-scale cascade for a large variation of the Reynolds number, the reduced frequency, and the flow coefficient. Linear models have been educated accounting for the Occam’s razor parsimony criterion, condensing the effects due to the parameter variation in few predictors. Then, these predictors have been used to generate an extended polynomial-base function, and an ℓ1-norm constrain has been included into the optimization process to promote sparsity. Different non-linear models have been also evaluated, introducing the formulation of Gaussian Processes for regression for different kernel functions. The capabilities of the models here tuned are compared by means of cross-validation global and local criteria. Cross-validated error and leverage distribution have been analyzed. The proper compromise between model accuracy and generalizability is identified as a Pareto front in the space of the cross-validation indicators. In addition, a new variance-based indicator for the identification of the best model among the candidate ones has been introduced and its capability in complementing the cross-validation analysis is here discussed. The learning strategy has been finally replicated adopting an extremely large database, experimentally acquired in the framework of the extensive collaboration between the University of Genova and AvioAero. For confidentiality, results concerning this database are not shown in terms of response surface, but model performances are discussed in order to strengthen the strategy here adopted, that could be useful also to other research groups adopting their own data.

This paper presents an assessment of machine-learned turbulence closures, trained for improving wake-mixing prediction, in the context of LPT flows. To this end, a three-dimensional cascade of industrial relevance, representative of modern LPT bladings, was analyzed, using a state-of-the-art RANS approach, over a wide range of Reynolds numbers. To ensure that the wake originates from correctly reproduced blade boundary-layers, preliminary analyses were carried out to check for the impact of transition closures, and the best-performing numerical setup was identified. Two different machine-learned closures were considered. They were applied in a prescribed region downstream of the blade trailing edge, excluding the endwall boundary layers. A sensitivity analysis to the distance from the trailing edge at which they are activated is presented in order to assess their applicability to the whole wake affected portion of the computational domain and outside the training region. It is shown how the best-performing closure can provide results in very good agreement with the experimental data in terms of wake loss profiles, with substantial improvements relative to traditional turbulence models. The discussed analysis also provides guidelines for defining an automated zonal application of turbulence closures trained for wake-mixing predictions.

In the present work URANS simulations are presented to describe the unsteady interaction process between the flow ingested/ejected from a cavity system and the main flow evolving into a Low Pressure Turbine stage. Particular care is posed on the analysis of the loss generation mechanisms acting outside the stator row and in the rear part of the axial gap separating the cavity flow ejection section and the leading edge plane of the downstream rotor row. The simulated geometry reproduces a typical engine cavity configuration, with upstream and downstream rotor rows reproduced by means of moving bars. Experimental results have been used to validate the simulations. This experimental data cannot explain and quantify alone the overall interaction process between the cavity flows and the main flow. The results of a simulation made by removing the domain of the cavity have been employed in order to better highlight and quantify the effects due to main flow and cavity flows interaction on total pressure loss. A deep inspection of the loss amount along the axial direction makes evident that losses generated in the vane row are basically increased prior to enter into the downstream rotor bars, due to cavity main flow interaction.

In this study, proper orthogonal decomposition (POD) has been applied to a large dataset describing the profile losses of low-pressure turbine (LPT) cascades, thus allowing (i) the identification of the most influencing parameters that affect the loss generation; (ii) the identification of the minimum number of requested conditions useful to educate a model with a reduced number of data. The dataset is constituted by the total pressure loss coefficient distributions in the pitchwise direction. The experiments have been conducted varying the flow Reynolds number, the reduced frequency, and the flow coefficient. Two cascades are considered: the first for tuning the procedure and identifying the number of really requested tests, and the second for the verification of the proposed model. They are characterized by the same axial chord but different pitch-to-chord ratio and different flow angles, hence two Zweifel numbers. The POD mode distributions indicate the spatial region where losses occur, the POD eigenvectors provide how such losses vary for different design conditions and the POD eigenvalues provide the rank of the approximation. Since the POD space shows an optimal basis describing the overall process with a low-rank representation (LRR), a smooth kernel is educated by means of least-squares method (LSM) on the POD eigenvectors. Particularly, only a subset of data (equal to the rank of the problem) has been used to generate the POD modes and related coefficients. Thanks to the LRR of the problem in the POD space, predictors are low-order polynomials of the independent variables (Re, f+, and ϕ). It will be shown that the smooth kernel adequately estimates the loss distribution in points that do not participate to the education. In addition, keeping the same steps for the education of the kernel on another cascade, loss distribution and magnitude are still well captured. Thus, the analysis show that the rank of the problem is much lower than the tested conditions, and consequently, a reduced number of tests are really necessary. This could be useful to reduce the number of hi-fidelity simulations or detailed experiments in the future, thus further contributing to optimize LPT blades.

The boundary layer developing over the suction side of a low pressure turbine cascade operating under unsteady inflow conditions has been experimentally investigated. Time-resolved Particle Image Velocimetry (PIV) measurements have been performed in two orthogonal planes, the blade to blade and a wall parallel plane embedded within the boundary layer, for two different wake reduced frequencies. Proper Orthogonal Decomposition (POD) has been used to analyze the data and to provide an interpretation of the most significant flow structures for each phase of the wake passing cycle. To this purpose, a POD based procedure that sorts the data synchronizing the measurements of the two planes has been developed. Phase averaged data are then obtained for both cases. Moreover, once properly sorted, POD has been applied to sub-ensembles of data at the same relative phase within the wake passing cycle. Detailed information on the most energetic turbulent structures at a particular phase are obtained with this procedure (called phased POD), overcoming the limit of classical phase average that just provides a statistical representation of the turbulence field. Furthermore, the synchronization of the measurements in the two planes allows the computation of the characteristic dimension of boundary layer structures that are responsible for transition. These structures are often identified as vortical filaments parallel to the wall, typically referred to as boundary layer streaks. The largest and most energetic structures are observed when the wake centerline passes over the rear part of the suction side, and they appear practically the same for both reduced frequencies. The passing wake forces transition leading to the breakdown of the boundary layer streaks. Otherwise, the largest differences between the low and high reduced frequency are observed in the calmed region. The post-processing of these two planes further allowed us to compute the spacing of the streaks and make it non-dimensional by the boundary layer displacement thickness observed for each phase. The non-dimensional value of the streaks spacing is about constant, irrespective of the reduced frequency.

A Bayesian method has been used to identify the best model strategy to describe the profile losses of low pressure turbine (LPT) cascades operating under unsteady inflow. The model has been tuned with experimental data measured in a large scale cascade facility, equipped with a moving bar system. Tests have been carried out on two different cascades, investigating three different reduced frequencies, three mass flow coefficients and several Reynolds numbers (up to eight) per condition, accounting for an overall amount of 51 different combinations of these parameters for each cascade. The predictor functions included into the model have been varied starting from a classic polynomial formulation for each influencing parameter, and then with functional relationships mimicking physical constrains and loss tendencies. Different combinations of the predictors, also including different types and orders of the cross-terms, have been evaluated by means of a Bayesian model selection method searching for the maximum probability of the model in fitting the cloud of experimental data. In particular, the evaluation of the Model Evidence (ME) using the Bayesian Information Criterion approximation (BIC) has allowed obtaining sufficient accuracy and avoiding overfitting at the same time. The best model here identified will be shown to be able to well reproduce the loss surface of a third different cascade that does not participate to the model selection. Realistic profile loss evolutions outside of the design space tested are provided, thus also allowing for a generalization of the structure of the model for other applications and future works.

In the present work the profile losses of a highly loaded low pressure turbine cascade, operating under realistic unsteady conditions, have been measured for different incoming wake parameters. A moving bar system has been used for wake generation, allowing the variation of bar diameter, bar count and axial gap between the bars and the cascade leading edge. The overall test matrix spans three different bar diameters, three different reduced frequencies and two axial gaps, for a total of 18 different conditions tested. For each of them, the time dependent cascade inflow has been characterized by means of phase-locked hot-wire measurements in order to determine the effects of the wake parameter modification on the turbulence intensity peak and on the incoming wake momentum deficit, that are well known to affect the generation of losses in the downstream cascade. The procedure recently developed by the authors based on the simultaneous acquisition of the signals of two kiel probes, located upstream and downstream of the cascade, has been used to accurately quantify the profile losses for each condition, identifying the contribution due to the wake bowing, tilting and dilation process into the cascade channel and that due to the wake-boundary layer interaction. Additionally, Proper Orthogonal Decomposition (POD) has been applied to the ensemble data matrix constructed from the loss coefficient distributions in the pitchwise direction, measured for the different conditions, with the aim of highlighting the loss trend vs the design space parameter variation. The POD modes obtainedfrom the cross-correlation matrix provide a direct information on the flow region where losses are prevalently produced, while the corresponding POD coefficients give the weight of each parameter in the loss generation. The paper shows the potentiality of this procedure in providing a rapid identification of the main causes of losses, and the dominant parameter affecting the process of loss generation.