One of the issues of handling large CFD datasets and process them to derive important design correlations is the limitation in automating the post-processing of data. Machine learning techniques, developed to process large unlabelled dataset, can play a key role on this subject. In this work an unsupervised approach to isolate different flow features inside a 2D cascade is proposed and validated. The approach relies on machine learning methods and in particular on Exploratory Data Analysis (EDA) and Principal Component Analysis for the pre-processing of the data and on K-means clustering for the post-processing. The K-means algorithm was trained on a Design of Experiments (DoE) of over 140 cases of 2D linear cascade configurations to identify the boundary layer on the profiles and the wake downstream. Validation resulted in a perfect capability of identifying the regions of interest. Then a possible exploitation of this method is presented, to compute pressure losses downstream of the cascade and train an artificial neural network to make a regression able to extend data to all the possible combinations of geometrical and operating parameters of the cascade. The same algorithm was applied to 3D flow cascades of profiles with sinusoidal leading edges to stress its extrapolation capability in case of flow regimes not present in the training DoE.

The present paper discusses the results of a large experimental data set describing transitional boundary layers. Time resolved Particle Image Velocimetry (PIV) measurements have been adopted to survey the boundary layer developing over a flat plate under prescribed adverse pressure gradients typical of turbomachinery components. The tests have been performed while varying the pressure gradient, the Reynolds number and the inlet free-stream turbulence intensity (FSTI). Two exemplary cases, referring to bypass and separated flow transition, are discussed by means of principal axis analysis and proper orthogonal decomposition (POD). The POD is used to provide statistical representation of the flow structures and to compute the turbulence production (i.e., the mean flow energy dissipation) due to the dynamical features observed for the different transition types. Reduced order model representations of the flow field are provided and their contribution to the total turbulence kinetic energy production is isolated. This analysis is closed by the inspection of the eigenvectors of the strain rate and Reynolds stress tensors. For the separated flow case, it is shown that the eigenvectors of strain rate and shear tensor are almost perfectly aligned downstream of the maximum displacement of the bubble. The reduced order model reconstruction of the Kelvin-Helmholtz shed vortices provides the largest part of the overall TKE production. For the high FSTI induced transition, the eigenvectors of the shear and stress tensors do not have the same direction. The loss generation is related to the local maximum Reynolds normal stress in the streamwise direction, induced by the boundary layer streaks and their breakdown.

In the present work, 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 the 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. Additionally, keeping the same steps for the education of the kernel on another cascade, loss distribution and magnitude are still well captured. Thus, 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 paper discusses the impact of rotor purge flows on the unsteady flow field downstream of a two-stage, two-spool test turbine. The analyzed setup is representative of the second high-pressure turbine (HPT) and the first low-pressure turbine (LPT) stage in a modern turbofan aero-engine, with a turbine center frame (TCF) with non-turning struts in-between the two turbines. All measurements were carried out for an engine-representative test vehicle setup at the Transonic Test Turbine Facility at Graz University of Technology. The test rig features a secondary air system delivering five purge flows with independent temperature and mass flow control to the HPT and LPT cavities. This work extends the results shown in two recent publications analyzing the time-resolved flow through the same two-stage setup at fixed purge flow rates. The paper aims to provide additional input about the driving sources of unsteadiness in gas turbines for aeronautic applications, by isolating the HPT and LPT purge air contributions. The time-resolved flow field at the LPT exit was acquired with a Fast Response Aerodynamic Pressure Probe (FRAPP) for three different purge conditions (reference case, no HPT purge case, no LPT purge case), to separate and quantify the impact of HPT and LPT purge contributions on the main flow field. The so-called Rotor Synchronic Averaging (RSA) technique was used as phase-averaging approach, to account for the unsteadiness due to both rotors. Proper Orthogonal Decomposition (POD) was then applied to isolate the most important structures and identify their origins. The comparison of the three data-sets shows a significant influence of the HPT purge on the entire flow field at the LPT exit, even though the HPT is located far upstream, while the LPT purge impact appears to mostly affect the end-wall region.

In the case of aircraft engines, the fuel is injected as a liquid spray which may play a role in thermoacoustic instabilities through creating changes to the mixture fraction inside the combustion chamber. This study uses two-phase incompressible non-reacting large eddy simulation with Lagrangian particle tracking to show how spray droplets of different sizes can be affected by large scale hydrodynamic structures and acoustic forcing. The forcing is applied at the inlets of a truncated computational domain that only includes the geometry downstream of the fuel injector using the newly developed PODFS (proper orthogonal decomposition Fourier series) method. The PODFS is a model that can reproduce the effects of acoustic forcing by extracting planes of data from an auxiliary acoustically forced compressible unsteady Reynolds averaged Navier-Stokes simulation. A proper orthogonal decomposition analysis shows that fuel droplets of a typical size seen in jet engines are more sensitive to acoustic and hydrodynamic structures than droplets with an order of magnitude larger or smaller diameter, consistent with their Stokes number. Phase and azimuthally averaged results show that fluctuations of the spray mixture fraction represented by large droplets affect the total spray mixture fraction much more than fluctuations of the small droplets. An additional intermittent spray dispersion mechanism was identified that is due to intermittent vorticity being generated between the two outer injector flow passages. An injector design modification has been suggested that will reduce the prevalence of this mechanism.

This paper presents measurements of 10 kHz acetone planar laser induced fluorescence (PLIF) to study the behavior of effusion cooling fluid injected into a non-reacting gas turbine combustor flow at elevated pressure. This study was performed as part of a larger effort to understand potential interactions of the swirling flame with the cooling air. The combustor — which was representative of a rich-burn/quick-quench/lean-burn (RQL) configuration — consisted of a swirl nozzle, quench jets, and a modular liner that could be fitted with various effusion cooling panels and optical access windows. Primary air was seeded with acetone, and passed through the swirl nozzle. Unseeded secondary air was passed on the outside of the liner, entering the combustion chamber through the quench jets and effusion panels. The PLIF laser sheet was arranged parallel to the effusion panel at various offset distances to visualize the mixing between the core flow and effusion jets. The PLIF images were analyzed with a POD-based methodology to de-noise the images and identify patterns in the effusion jet characteristics. The results show that high blowing ratios produce individual effusion jets rather than a single, coalesced film. The effusion jets are highly unsteady, interacting strongly with the turbulent flow from the swirl nozzle and dilution jets. Furthermore, the average trajectories of effusion jets are non-uniform across the panel and are shaped by upstream features in the combustor, namely swirl and dilution jets.

Gas path diagnostics is a key aspect of the engine health monitoring (EHM) process that aims to detect, identify and predict engine component faults, using information from installed sensors, in order to guide maintenance action, maintain engine efficiency and prevent catastrophic failures. To achieve high prediction accuracies, current data-derived diagnostic models tend to be engine specific while the model-based methods are known to be time-consuming, especially for complex engine configurations. This paper proposes an integrated approach for accurate and accelerated isolation and prediction of multiple-degraded gas turbine component faults that comprises 3 steps — feature extraction using the Principal Component Analysis (PCA), machine learning classification with a multi-layer perceptron, artificial neural network (MLP-ANN) and model-based fault prediction via the non-linear Gas Path Analysis (GPA) technique. In this hybrid approach, the PCA first transforms the measurement fault signature into a fault-feature domain, which becomes an input to the multi-label ANN classifier used to isolate the potential faulty components. The non-linear GPA finally quantifies the magnitude of degradation that produced the recorded fault signature. Once trained and validated, the PCA-ANN model is deployed as part of the data processing mechanism prior to the actual GPA calculation. This method was assessed and validated using the thermodynamic performance model of a 2-shaft, high-bypass ratio, turbofan engine. For training and testing the PCA-ANN classifier, a total of 28,000 final samples for 14 measurement parameters, each averaged from 10 data points with Gaussian noise of zero mean and unit standard deviation, and implanted with single-, double- and triple-component fault cases of various magnitude, were generated by steady-state performance simulation of the engine model at its reference operating condition. Correlation analysis of this data set revealed the optimum sensor subset to be used for multi-component diagnostics. A quantitative analysis of the PCA-ANN fault isolation on the test set produced a classification accuracy of 96.6% and performed better on all metrics, compared to other multi-label classification algorithms. Finally, the proposed integrated approach achieved an average of 94.35% reduction in processing time, when compared to the conventional non-linear GPA by component-fault-cases (CFCs), while predicting implanted faults to the same accuracy.

An important noise source in modern high bypass ratio turbofans is from multiple pure tones produced by the fan during takeoff. An experiment conducted on a 1.5 pressure ratio fan in an internal flow facility provided dynamic pressure measurements to investigate multiple pure tone generation and propagation. Since multiple pure tones are generated by blade shock variation primarily due to the fan’s blade stagger angle differences, the blade stagger angles were measured with an array of over-the-rotor dynamic pressure transducers. Multiple pure tone measurements were made with 30 wall-mounted dynamic pressure transducers from 0.4 to 1.1 diameters upstream of the rotor. Measured blade stagger angle differences correspond to the the shock amplitude variation measured upstream. The acoustic field was extracted from the dynamic pressure signals using principal component analysis as well as duct mode beamforming. Shocks traveling out the inlet were found to couple to duct modes propagating at similar angles. Over-the-rotor acoustic liners appear to reduce rotor shock variation resulting in a reduction of sub-harmonic multiple pure tone sound pressure levels by 3–4 dB.

In the present work reduced order models (ROM) that are independent from the full order finite element models (FOM) considering geometrical non linearities are developed and applied to the dynamic study of a fan. The structure is considered to present nonlinear vibrations around the pre-stressed equilibrium induced by rotation enhancing the classical linearised approach. The reduced nonlinear forces are represented by a polynomial expansion obtained by the Stiffness Evaluation Procedure (STEP) and then corrected by means of a Proper Orthogonal Decomposition (POD) that filters the full order nonlinear forces (StepC ROM). The Linear Normal Modes (LNM) and Craig-Bampton (C-B) type reduced basis are considered here. The latter are parametrised with respect to the rotating velocity. The periodic solutions obtained with the StepC ROM are in good agreement with the solutions of the FOM and are more accurate than the linearised ROM solutions and the STEP ROM. The proposed StepC ROM provides the best compromise between accuracy and time consumption of the ROM.

Geometric uncertainties in the blade manufacturing process have important consequences in terms of dynamical properties of bladed disks. In this paper we address the problem of modeling a full bladed disk composed by blades having uncertain geometry. The geometric imperfection of the blades is represented and analyzed according to a procedure previously presented by the authors, based on the principal component analysis and the mesh morphing. The dynamical model of the full disk is constructed following the component mode synthesis approach. The blade geometry is represented using a probabilistic model constructed from an experimental dataset. The effect of the geometric uncertainties is assessed using a linear propagation approach, leading to a procedure that is fast enough to be embedded into a Monte Carlo simulation loop.

In the modern highly-loaded gas turbine, due to the large pressure difference between the suction side and the pressure side of the turbine blade, strong cross flow is formed and it strongly affects the aerodynamic and cooling performances in the end-wall region. The film cooling behavior in the environment of strong cross flow is different from the straight channel environment widely studied in the literature. In this research, the effect of cross flow on film cooling is investigated by Large Eddy Simulation (LES) using subgrid-scale (SGS) model. Numerical simulation is carried out in a curved passage to simulate the turbine blade passage. Shaped cooling hole with blowing ratio 1 is studied. The time-averaged friction line results are compared with existing experimental ink trace results. The vortex structures, both time-averaged and instantaneous, are analyzed to study the effect of cross flow on film cooling. At the exit of the cooling hole, the hanging vortices with negative y-vorticity are more flat in shape and closer to the wall in position in contrast to hanging vortex with positive y-vorticity, which is caused by cross flow and results in the asymmetry of hairpin vortices downstream as well as the asymmetry of the distribution of coolant. It has been shown that the vortices from mainstream have a significant impact on the field near the exit of the cooling hole. Those vortices interact with the hairpin vortices from the cooling hole and directly lead to the asymmetry of the hairpin vortices. Proper Orthogonal Decomposition (POD) analysis is further conducted to extract the dominant flow structures and the physical mechanisms of primary POD modes are given to explain the distribution of film cooling effectiveness affected by cross flow. Based on the specific situation in this work, a fast incremental POD (iPOD) approach is adopted since the rank of the field matrix is far less than the rows, which is caused by the tall and thin character of the matrix, which makes the analysis less costly and more effective. This research helps to understand the cooling performance in the real turbine blade passage and to explain the coolant mixing process based on the instantaneous flow field obtained using high precision LES simulation and powerful iPOD.

The adiabatic film cooling effectiveness behind a single row of circular holes fed by internal crossflow was measured by fast-response pressure-sensitive paint technique. During the experiment, the coolant flow was discharged from the coolant holes via either plenum or crossflow channel. The test model has a row of circular holes with 3D spacing, 6.5D entry length, and 35° inclination angle. Two blowing ratios (M = 0.40 and 0.80) were tested with a density ratio of 0.97. A numerical steady-state RANS simulation, using SST k-ω and Realizable k-ε turbulence models, was conducted to understand the internal crossflow behaviors. The unsteadiness caused by the flow structures (counter-rotating vortex pair (CRVP) and horseshoe vortex) was quantified by the root mean square and the cross-correlations. In addition, the proper orthogonal decomposition was used to identify the large-scale unsteady coherent structures and their contributions. The fluctuations of the crossflow feed were asymmetric, which were significantly weaker compared with the plenum case. The CRVP, as the most significant coherent structures, were demonstrated to play the main role in the unsteadiness of the crossflow feed.

With increasing complexity of aerodynamic devices such as gas turbine fuel swirl nozzles (FSN) and combustors, the need for time-resolved full volume flow characterisation is becoming greater. Even with modern advancements in both numerical and experimental methods, this remains a challenging area. The work presented in this paper combines multiple non-synchronous planar measurements to reconstruct an estimate of a synchronous, instantaneous flow field of the whole measurement set. Temporal information is retained through the linear stochastic estimation (LSE) technique. The technique is described, applied and validated with a simplified combustor and FSN geometry flow for which 3-component, 3-dimensional (3C3D) flow information is known from published tomographic PIV[1]. Using the tomographic PIV data set, multiple virtual ‘planes’ may be extracted to emulate single planar PIV measurements and produce the correlations required for LSE. In this example, multiple parallel planes are synchronised with a single perpendicular plane that intersects each of them. As the underlying data set is volumetric, the measured velocity is known a priori and therefore can be directly compared to the estimated velocity field for validation purposes. The work shows that when the input time-resolved planar velocity measurements are first POD (proper orthogonal decomposition) filtered, high correlation between the estimations and the validation velocity volumes are possible. This results in estimated full volume velocity distributions which are available at the same time instance as the input field — i.e. a time resolved velocity estimation at the frequency of the single input plane. While 3C3D information is used in the presented work, this is necessary only for validation; in true application planar technique would be used. The study concludes that provided the number of sensors used for input LSE exceeds the number of POD modes used for pre-filtering, it is possible to achieve correlation greater than 99%.

Swirling flow is often employed in gas turbine combustion chambers for the sake of improving flame stability. Swirling flow induces not only recirculation zones but also large coherent structures which show close relationship with flow dynamics and combustion instability. The flow dynamics including Precessing Vortex Core (PVC) in simple swirlers are extensively studied, while the flow instability characteristics in a multi-swirler combustor are not fully reported. In the present paper, Large Eddy Simulation (LES) of non-reacting turbulent swirling flow is conducted in a multi-swirler burner which comprises a pilot stage and a main stage. Flow dynamics in the multi-swirler combustor are analyzed based on phase-averaged evolution of instantaneous flowfield. Proper Orthogonal Decomposition (POD) is employed to identify the coherent structures in the multi-swirling flow. Results show that the main stage and pilot stage flow interact with each other generating highly turbulent swirling flow. PVC is successfully captured at the boundary of Main recirculation zone (MRZ) in the pilot stage with a dominant frequency of 1915 Hz. The PVC leads to periodic azimuthal flow instability. POD analyses for the velocity fields show dominant high-frequency modes (mode 1 and mode 2) in the pilot stage. However, the dominant energetic flow is damped rapidly downstream of the pilot stage that it has little effect on the main stage flow.

Many industrial combustion systems, especially power generation gas turbines, use fuel-lean combustion to reduce NO x emissions. However, these systems are highly susceptible to combustion instability, the coupling between combustor acoustics and heat release rate oscillations of the flame. It has been shown in previous work by the authors that a precessing vortex core (PVC) can suppress shear layer receptivity to external perturbations, reducing the potential for thermoacoustic coupling. The goal of this study is to understand the effect of combustor exit boundary condition on the flow structure of a swirling jet to increase fundamental understanding of how combustor design impacts PVC dynamics. The swirling jet is generated with a radial-entry, variable-angle swirler, and a quartz cylinder is fixed on the dump plane for confinement. Combustor exit constriction plates of different diameters are used to determine the impact of exit boundary condition on the flow field. Particle image velocimetry (PIV) is used to capture the velocity field inside the combustor. Spectral proper orthogonal decomposition, a frequency-resolved eigenvalue decomposition that can identify energetic structures in the flow, is implemented to identify the PVC at each condition in both energy and frequency space. We find that exit boundary diameter affects both the structure of the flow and the dynamics of the PVC. Higher levels of constriction (smaller diameters) force the downstream stagnation point of the vortex breakdown bubble upstream, resulting in greater divergence of the swirling jet. Further, as the exit diameter decreases, the PVC becomes less energetic and less spatially defined. Despite these changes in the base flow and PVC coherence, the PVC frequency is not altered by the exit boundary constriction. These trends will help inform our understanding of the impact of boundary conditions on both static and dynamic flame stability.

Liquid fuel spray characterisation is essential for understanding the mechanisms underlying fuel energy release and pollutant formation. Careful selection of operating conditions can promote flow instabilities in the fuel spray which can enhance atomisation and fuel mixing, thereby resulting in more efficient combustion. However, the inherent instabilities present in the spray could have adverse effect on the combustor dynamics. Hence, it is important to better understand the dynamical behaviour of the spray, and particularly at representative operating conditions. This work describes an experimental investigation of dynamical behaviour of pressure-swirl atomisers used in Siemens industrial gas turbine combustors, at a range of chamber pressures and fuel injection pressures, using high speed laser planar imaging. Two modal decomposition techniques — Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition (DMD) — are applied and compared to assess the spray dynamics. Results indicate that both POD and DMD are able to capture periodic structures occurring in the spray at different spatial length scales. The characteristic frequencies estimated from both the methods are in good agreement with each other. Both techniques are able to identify coherent structures with variable size, shape and level of staggering, which are observed to be dependent on the pressure difference across the atomiser and the chamber pressure. The spatio-temporally resolved data and the results could be used for spray model development and validation. Furthermore, the methods employed could be applied to other fuel atomisers, and more complicated conditions involving cross flow and higher chamber temperatures.

Recent improvements in computational and experimental combustion have led to an increased availability of high-fidelity data sets. With the surge of high-quality data, the reduction and analysis of this information become a key aspect in improving our understanding of highly dynamic, reacting flow fields. This work utilizes simultaneous high-speed stereoscopic particle image velocimetry (sPIV), OH-planar laser induced fluorescence (PLIF) and fuel-PLIF measurements in a high pressure, liquid fueled swirl combustor. This work extends one of the most commonly used modal analysis techniques, proper orthogonal decomposition (POD). Instead of decomposing the data as a series of time coefficients and spatial modes, which is the result of snap-shot POD, the data is re-oriented such that the coefficients are a function of the transverse distance and the modes are a function of axial distance and time. This reoriented POD method, referred to as spatio-temporal POD, allows us to directly target convective flow structures while still extracting the same types of information as from snap-shot POD. Additionally, since the modes contain temporal information, phase speeds associated with the convective structures can easily be calculated from the phase portraits. This paper presents results of this analysis, allowing us to track the growth of coherent structures, their phase speeds, and their subsequent decay.

This work is part of a research initiative that aims at increasing the overall gas turbine efficiency by means of constant volume combustion (CVC). For that purpose, flow control in the compressor becomes important, since unsteady combustion effects that may occur in a CVC are very likely to affect stability and efficiency of the compressor negatively due to flow disturbances. Active Flow Control (AFC) often has to deal with uncertain flow conditions, e.g., due to turbulence, varying operating ranges, or simply environmental effects. By that, system parameters such as gain or time constants of the system model also become uncertain, making it difficult for control algorithms to ensure optimality or even stable behavior. Robust control in the sense of ℋ ∞ control tackles these problems using an uncertainty description and a nominal model of the system. In this contribution, robust control applied to a linear stator cascade is addressed when only a binary control output from solenoid valves is available. Moreover, a surrogate control variable is proposed, describing the extent of the velocity deficit. By means of a principal component analysis, this control variable is reconstructed from a single measurement input. AFC is realized via trailing edge blowing. In comparison to proportional valves, solenoid valves are cheaper and offer faster switching times with the drawback of a restricted range of the control output to integer or even binary values. Since the ℋ ∞ controller, as well as most other control algorithms, results in a real-valued signal u(t) ∈ ℝ, a sum-up rounding strategy is applied to the controller output, forming a binary control output u b (t) ∈ {0, 1}. Although it is impossible for the two outputs to completely match, unless both are integer-valued, there is proof that the difference of real-valued to binary output is bounded in its integral value. The investigations show that a switching frequency of the valves of 100 Hz is sufficient to ensure that the control error via binary control matches its expected equivalent via real-valued control for the presented system.

This two-part paper presents the unsteady flow interactions between an engine-representative high-pressure turbine (HPT) and low-pressure turbine (LPT) stage, connected by a turbine center frame (TCF) duct with non-turning struts. The setup was tested at the high-speed two-spool test turbine facility at the Institute for Thermal Turbomachinery and Machine Dynamics at Graz University of Technology and includes relevant purge and turbine rotor tip leakage flows. Due to the complexity of such a test, the unsteady component interactions in an HPT-TCF-LPT module have not received much attention in the past and require additional analysis to determine new approaches for further performance improvements on the system level. The flow downstream of an HPT is highly unsteady and dominated by statorrotor interactions, which affect the flow behavior through the downstream TCF and LPT. To capture the unsteady flow structures, time-resolved aerodynamic measurements were carried out with a fast-response aerodynamic pressure probe (FRAPP) at three different measurement planes. In this first part of the paper, the time-resolved and phase-averaged flow fields with respect to the HPT and LPT trigger are studied. Since the two rotors are uncorrelated, the applied method allows the identification of the flow structures induced by either of them. Upstream of the LPT stage, the HPT flow structures evolving through the TCF duct dominate the flow fields. Downstream of the LPT stage, the flow is affected by both the HPT and the LPT secondary flow structures. The interactions between the various stator rows and the two rotors are detected by means of time-space plots and modal decomposition. To describe the fluctuations induced by both rotors, particularly the rotor-rotor interaction, the Rotor Synchronic Averaging (RSA) is used to analyze the flow field downstream of the LPT. The second part of the paper decomposes the flow fields to gain additional insight into the rotor-rotor interactions using the Proper Orthogonal Decomposition (POD) and RSA methods. The paper highlights the need to account for the HPT-induced unsteady mechanisms in addition to the LPT flow structures and the interaction of both to arrive at improved LPT designs.

The complex leakage flow structure in the tip region of unshrouded rotor is a main source of turbine aerodynamic loss. Due to the complex turbulence characteristics of the tip leakage flow, the widely used Reynolds Averaged Navier-Stokes (RANS) approach may fail to accurately predict the multi-scale turbulent flow and the related loss. In order to effectively improve the turbine efficiency, more insights into the turbulence characteristics and the loss mechanism in the tip leakage flow are required. In this work, a Delayed Detached Eddy Simulation (DDES) study is conducted to simulate the flow inside a high pressure turbine blade, with emphasis on the tip region. DDES results are in good agreement with the experiment and the comparison with RANS results verifies the advantages of DDES in resolving finer flow structures of leakage flow, also in capturing the complex turbulence characteristics. The snapshot Proper Orthogonal Decomposition (POD) method is used to extract the dominant flow features. The flow structures and the distribution of Reynolds stress help to reveal the process of leakage flow and its interaction with the secondary flow. Meanwhile, it is found that the separation vortex (SV) forms from leading edge to trailing edge, and the strong interactions between tip leakage vortex (TLV) and passage secondary vortex (PSV) significantly enhance the turbulence intensity. Based on the DDES results, loss analysis of tip leakage flow is conducted based on entropy generation rates. For the leakage flow related loss, the largest local entropy generation rate occurs at 50 % of axial chord, and the interaction between the leakage vortex and up passage vortex promotes the loss generation. To sum up, the current DDES study about the tip leakage flow provides helpful information about the loss generation mechanism and may guide the design of low-loss blade tip.