Turbomachinery Flows Research Articles

Fast prediction of compressor flow field based on a deep attention symmetrical neural network

Accurate and rapid prediction of compressor performance and key flow characteristics is critical for digital design, digital twin modeling, and virtual–real interaction. However, the traditional methods of obtaining flow field parameters by solving the Navier–Stokes equations are computationally intensive and time-consuming. To establish a digital twin model of the flow field in a transonic three-stage axial compressor, this study proposes a novel data-driven deep attention symmetric neural network for fast reconstruction of the flow field at different blade rows and spanwise positions. The network integrates a vision transformer (ViT) and a symmetric convolutional neural network (SCNN). The ViT extracts geometric features from the blade passages. The SCNN is used for deeper extraction of input features such as boundary conditions and flow coordinates, enabling precise flow field predictions. Results indicate that the trained model can efficiently and accurately reconstruct the internal flow field of the compressor in 0.5 s, capturing phenomena such as flow separation and wake. Compared with traditional numerical simulations, the current model offers significant advantages in computational speed, delivering a three-order magnitude speedup compared to computational fluid dynamics simulations. It shows strong potential for engineering applications and provides robust support for building digital twin models in turbomachinery flow fields.

Regression modeling of Bödewadt slip flow dynamics involving Reiner-Rivlin nanofluid based on a modified Buongiorno approach

Regression models are useful in analyzing rotational flows as they enable accurate predictions of wall shear and heat transfer coefficient. In addition, Bödewadt flow is of paramount importance in fluid dynamics of rotating systems such as turbomachinery and geophysical flows. Moreover, nanofluid’s enhanced heat transfer properties can improve cooling efficiency in applications involving turbines and electronic systems. This study delves into the Bödewadt boundary layer flow of a Reiner-Rivlin fluid containing nanoparticles over a stationary porous disk under slip conditions. The two-phase Buongiorno model is employed, incorporating temperature-dependent diffusion coefficients for enhanced accuracy. To facilitate numerical simulations, the transport equations are converted into an ordinary differential system comprising four unknowns. In the present work, a highly reliable Keller-Box methodology is adopted which agrees very well with the MATLAB built-in program ‘bvp4c’. The computed 2-D and 3-D streamlines vividly capture the Bödewadt flow scenario with Reiner-Rivlin nanofluid. The principle aim to investigate the impact of non-Newtonian behaviour and slip on the flow pattern, while also examining the behavior of temperature/concentration field for nanoparticle working fluids. As thermophoretic diffusion increases, the thermal boundary layer thickens considerably, leading to a notable decrease in the cooling rate of the disk. In contrast, Brownian diffusion has only a minimal impact on the heat transport. In addition, wall suction effect is observed to significantly boost the disk’s cooling rate, though at the expanse of increasing skin friction coefficients. This study introduces linear and quadratic regression models designed to precisely predict both the surface drag and disk cooling rate, which are crucial factors in engineering processes.

Graphic processing unit accelerated time-domain harmonic balance method for multi-row turbomachinery flow simulation

Graphic processing unit accelerated time-domain harmonic balance method for multi-row turbomachinery flow simulation

Hydraulic loss analysis of turbodrill blade cascades based on entropy production theory

Turbodrill is a kind of high-temperature resistant downhole motor which is commonly used in drilling deep and high-temperature wells. It belongs to multi-stage axial flow turbomachinery. There are complex flow losses in turbodrill blade cascades, but conventional flow-loss evaluation methods cannot obtain the location of losses in cascades. The entropy production method constructs the relationship between flow field parameters and flow losses, so the location where flow losses generate can be obtained, which is commonly used to analyze the flow losses in various flows. In this paper, entropy production method is used to obtain the magnitude and position of hydraulic losses in turbodrill blade cascades. The single passage calculation model of three-stage stator and rotor is established, and CFD method is employed to compute the flow field parameters at different viscosities and rotary speeds. Different kinds of entropy production rates and values are calculated. Finally, the total flow loss calculated by entropy production method is compared with that calculated by pressure difference method, and the contrastive result is very similar, which indicates the correctness of entropy production method. The results show that the high entropy production regions are mainly located near the walls and the position changes with variations in rotary speed and viscosity, and the stator entropy production value is always lower than that in rotor. In addition, when the fluid viscosity is low, the main types of entropy production are turbulent entropy production and wall friction entropy production, while when the fluid viscosity is high, the main types of entropy production are direct entropy production and turbulent entropy production. Furthermore, the rotary speed variation has a greater impact on turbulent entropy production and wall friction entropy production compared to viscosity change, while the viscosity change has a greater impact on direct entropy production than the variation rotary speed. This paper introduces entropy production method into flow losses analysis for turbodrill, presenting a new approach for studying flow losses and providing a reference for improving turbodrill design.

Application of a Novel Weighted Essentially Non-Oscillatory Scheme for Reynolds-Averaged Stress Model and Reynolds-Averaged Stress Model/Large Eddy Simulation (RANS/LES) Coupled Simulations in Turbomachinery Flows

In numerical simulations, achieving high accuracy without significantly increasing computational cost is often a challenge. To address this issue, this paper proposes an improved finite volume Weighted Essentially Non-Oscillatory (WENO) scheme for structured grids. By employing a single-point quadrature rule to perform flux integration on the control volume faces, this scheme is designed for use in NUAA-Turbo three-dimensional fluid solvers based on structured grids, utilizing RANS and RANS/LES coupling to simulate turbomachinery flows. Firstly, the new WENO scheme is validated against classical numerical test cases to evaluate its stability and reliability in handling discontinuities, double Mach reflection problems, and Rayleigh–Taylor (RT) instability. Compared to the original scheme, this improved finite-volume WENO scheme demonstrates better stability near discontinuities and more effectively resolves flow features at the same grid resolution. Next, for engineering applications related to turbomachinery, such as compressor and turbine characteristics, calculations using RANS are performed and the results obtained with WENO-ZQ3 and WENO-JS3 are compared. Finally, the new fifth-order WENO scheme is applied to RANS/LES coupling simulations of turbine wake and film cooling. The results indicate that the improved finite-volume WENO scheme provides better stability and accuracy in engineering applications. For instance, the average error in calculating compressor efficiency characteristics is reduced from 0.76% to 0.05%, the error in turbine vane pressure distribution compared to the experimental values is within 1%, and the error in film cooling efficiency centerline distribution compared to the experimental values is within 3%. Additionally, the qualitative results of turbine wake and film cooling show that even with a small number of grid points, more detailed flow physics can be captured, thereby reducing computational costs in aerodynamic applications.

Block Spectral Method Analysis for Turbomachinery Applications

Abstract The block-wise spectral reconstruction of turbomachinery flows has been successfully used in the past to reduce the computational cost of full-annulus turbomachinery simulations. The rationale of the method is based on an explicit representation of the long-wavelengths of the problem using a Fourier expansion of the blocks, which in turn contain the short-wavelength content of the spectrum. This work analyzes the accuracy of this approach and shows that this approximation retains not only the explicit Fourier representation of the long wavelengths and finite volume approximation of the short wavelengths but their interaction as well. This interaction due to the scattering associated with the nonuniformity of the flow is referred to as Tyler–Sofrin modes by turbomachinery practitioners and is highly relevant for many practical applications. The solution’s harmonic content is illustrated by resorting to the linear wave equation with varying propagation speed and the NASA rotor 67 fan.

Simulation of Indexing and Clocking with a New Multidimensional Time Harmonic Balance Approach

Alongside the capability to simulate rotor–stator interactions, a central aspect within the development of frequency-domain methods for turbomachinery flows is the ability of the method to accurately predict rotor–rotor and stator–stator interactions on a single-passage domain. To simulate such interactions, state-of-the-art frequency-domain approaches require one fundamental interblade phase angle, and therefore it can be necessary to resort to multi-passage configurations. Other approaches neglect the cross-coupling of different harmonics. As a consequence, the influence of indexing on the propagation of the unsteady disturbances is not captured. To overcome these issues, the harmonic balance approach based on multidimensional Fourier transforms in time, recently introduced by the authors, is extended in this work to account for arbitrary interblade phase angle ratios on a single-passage domain. To assess the ability of the approach to simulate the influence of indexing on the steady, as well as on the unsteady, part of the flow, the proposed extension is applied to a modern low-pressure fan stage of a civil aero engine under the influence of an inhomogeneous inflow condition. The results are compared to unsteady simulations in the time-domain and to state-of-the-art frequency-domain methods based on one-dimensional discrete Fourier transforms.

Tip-leakage-flow excited unsteadiness and associated control

Tip leakage flow in turbomachinery inherently generates intense unsteady features, named self-excited unsteadiness, which significantly affects the operating stability, aerodynamic efficiency, and noise but has not been well understood. A zonalized large eddy simulation is employed for a linear cascade, with wall-modeled large eddy simulation active only in the tip region. The simulation is well validated with advantages demonstrated for effectively reducing the computational cost while maintaining an equivalent prediction accuracy in the region of interest. The time-averaged and spatial-spectral characteristics of tip leakage vortex (TLV) structures are systematically discussed. The self-excited unsteady processes of TLV include the tip gap separation, the tip leakage and jet-mainstream interaction, the primary tip leakage vortex (PTLV) wandering motion, and the induced separation near end wall. The Spectral Proper Orthogonal Decomposition (SPOD) is used to examine the dominant frequencies and their coherent structures. It is found that these unsteady features change from a single high frequency to multiple lower frequencies due to the PTLV breakdown. The SPOD and correlation analyses reveal that the self-excited unsteadiness originates initially from unsteady vortex separation in the tip gap and is then fed by the interactions between the tip leakage jet and mainstream. The associated unsteady fluctuations are convected along the tip leakage jet trajectory, causing the wandering motion of PTLV core. Based on the revealed unsteadiness sources, a micro-offset tip design is proposed and shown to be an effective solution to reducing the tip flow unsteadiness. This work improves the understanding of tip-leakage-flow dynamics and informs the control of the associated unsteady fluid oscillation and noise.

Spectral proper orthogonal decomposition of harmonically forced turbulent flows

Many turbulent flows exhibit time-periodic statistics. These include turbomachinery flows, flows with external harmonic forcing and the wakes of bluff bodies. Many existing techniques for identifying turbulent coherent structures, however, assume the statistics are statistically stationary. In this paper, we leverage cyclostationary analysis, an extension of the statistically stationary framework to processes with periodically varying statistics, to generalize the spectral proper orthogonal decomposition (SPOD) to the cyclostationary case. The resulting properties of the cyclostationary SPOD (CS-SPOD for short) are explored, a theoretical connection between CS-SPOD and the harmonic resolvent analysis is provided, simplifications for the low and high forcing frequency limits are discussed, and an efficient algorithm to compute CS-SPOD with SPOD-like cost is presented. We illustrate the utility of CS-SPOD using two example problems: a modified complex linearized Ginzburg–Landau model and a high-Reynolds-number turbulent jet.

Unsteady Flows and Component Interaction in Turbomachinery

Unsteady component interaction represents a crucial topic in turbomachinery design and analysis. Combustor/turbine interaction is one of the most widely studied topics both using experimental and numerical methods due to the risk of failure of high-pressure turbine blades by unexpected deviation of hot flow trajectory and local heat transfer characteristics. Compressor/combustor interaction is also of interest since it has been demonstrated that, under certain conditions, a non-uniform flow field feeds the primary zone of the combustor where the high-pressure compressor blade passing frequency can be clearly individuated. At the integral scale, the relative motion between vanes and blades in compressor and turbine stages governs the aerothermal performance of the gas turbine, especially in the presence of shocks. At the inertial scale, high turbulence levels generated in the combustion chamber govern wall heat transfer in the high-pressure turbine stage, and wakes generated by low-pressure turbine vanes interact with separation bubbles at low-Reynolds conditions by suppressing them. The necessity to correctly analyze these phenomena obliges the scientific community, the industry, and public funding bodies to cooperate and continuously build new test rigs equipped with highly accurate instrumentation to account for real machine effects. In computational fluid dynamics, researchers developed fast and reliable methods to analyze unsteady blade-row interaction in the case of uneven blade count conditions as well as component interaction by using different closures for turbulence in each domain using high-performance computing. This research effort results in countless publications that contribute to unveiling the actual behavior of turbomachinery flow. However, the great number of publications also results in fragmented information that risks being useless in a practical situation. Therefore, it is useful to collect the most relevant outcomes and derive general conclusions that may help the design of next-gen turbomachines. In fact, the necessity to meet the emission limits defined by the Paris agreement in 2015 obliges the turbomachinery community to consider revolutionary cycles in which component interaction plays a crucial role. In the present paper, the authors try to summarize almost 40 years of experimental and numerical research in the component interaction field, aiming at both providing a comprehensive overview and defining the most relevant conclusions obtained in this demanding research field.

Condensation shock topologies in carbon dioxide and a non-condensable gas mixture in supersonic nozzles

Condensation shock waves may occur in many flow expansion devices such as turbomachinery, gas ejectors, micro thrust-nozzles, and supersonic gas flow separators. However, their experimental analysis has been barely addressed as condensation shocks comprise complex phenomena such as compressible flow behavior, a shock-like phase transition, and a two-phase flow expansion. This work characterizes experimentally some condensation shock topologies of a mixture of carbon dioxide and dry air at several compositions in a Laval nozzle. Experiments were carried out in a test-rig instrumented with high-response pressure transducers installed along the Laval nozzle wall along with a Schlieren setup equipped with a high-speed video camera imaging the flow behavior within the nozzle. The nozzle wall profile was built by using the method of characteristics developed from a real equation-of-state suited for the testing mixture. Results revealed the influence of the nozzle wall profile on the condensation shock location and topology. Moreover, there types of flow behavior were captured and named as conventional, transition, and Mach wave condensation shocks. The transition from one topology to another occurred due to the interaction between cancelation waves originated from the nozzle wall and the phase-change phenomenon, giving rise to two distinct regions characterized by certain observable droplet population density. The current investigation presents an in-depth phenomenological discussion of the three types of condensation shock topologies as such assessment has not been previously developed.

Development of unsteady reduced-order methods for multi-row turbomachinery flow simulation based on the computational fluids laboratory three-dimensional solver

Development of unsteady reduced-order methods for multi-row turbomachinery flow simulation based on the computational fluids laboratory three-dimensional solver

Analyzing Unsteady Turbomachinery Flow Simulations With Mixing Entropy

Abstract The prediction of unsteady aerodynamic loads is a central problem during the design of turbomachinery. Over the last 20 years, harmonic balance methods have been shown to be highly efficient for this task. A CPU-cost optimal setup of a harmonic balance simulation, however, requires knowledge of relevant harmonics. In the case of a single blade row with a periodic disturbance this question amounts to the classical problem of harmonic convergence, a problem which is solely due to the nonlinearity of the unsteady flow physics. In contrast, for multi-stage configurations, the choice of harmonics is further complicated by the fact that the interactions of disturbances with blade rows may give rise to a vast spectrum of harmonics that possibly have important modal content, e.g., Tyler–Sofrin modes. The aim of this paper is to show that the mixing entropy attributed to circumferential modes of a given harmonic can serve as a disturbance metric on the basis of which a criterion could be derived whether a certain harmonic should be included or not. The idea is based on the observation that the entropy due to the temporal and circumferential mixing of the flow at a blade row interface may be decomposed, up to third-order terms, into independent contributions from different frequencies and mode orders. For a given harmonic balance (and steady) flow result, the mixing entropy attributed to modes that are simply mixed out, rather than resolved in the neighboring row, is shown to be a natural indicator of a potential inaccuracy. We present important features of the mixing entropy for unsteady disturbances, in particular a close relationship to sound power for acoustic modes. The problem of mode selection in a 1.5-stage compressor configuration serves as a practical example to illustrate our findings.

A compressible flow solver for turbomachinery of the real gases with strongly variable properties

Accurate prediction of real gas flows with strongly variable properties is an essential prerequisite for turbomachinery performance analysis. This paper presents a density-based solver for real gas flows in turbomachinery. The solver employs a third-order discretization scheme to improve computational accuracy. Real gas equations of state and look-up table are utilized in the solver to account for the strong nonlinear variations of thermophysical properties. The preconditioning method is adopted for simulations, including low-Mach flow regions, to address convergence difficulties. The implicit Time Consistent Preconditioned Gauss-Seidel Scheme (TCPGS) is employed for the time-stepping. The solver is validated by transonic compressor cascade wind tunnel experiments. Numerical test cases are adopted to evaluate the solver performance, including NASA Rotor 67, NASA Rotor 37, real gas flows in the Laval nozzle, and compressor seal. It has been proved that the numerical test results are consistent with published technical data. The advantages of this solver include numerical stability, computational efficiency, and physical accuracy, which indicate its applicability in the turbomachinery design process for real gases.

Experimental test of a compressor cascade with non-axisymmetric end-wall contouring at off-design conditions

End-wall losses in axial flow turbomachinery, particularly in compressors, constitute a significant challenge due to their adverse impact on overall efficiency. Pioneering works in this field have demonstrated the potential for end-wall loss reduction by improving end-wall profiles. However, many previous efforts have concentrated solely on design conditions, largely disregarding the behavior of these contoured end-walls under off-design conditions—where compressors often operate. This study investigates the adaptability of a non-axisymmetric end-wall contour, performing well under design conditions, to off-design scenarios, focusing particularly on the influence of Mach numbers on its ability to control corner separation effects. Experimental research was conducted using a combination of surface static pressure taps and aerodynamic probes to quantitatively and qualitatively analyze the effect of the non-axisymmetric end-wall contour on reducing losses. Quantitative outcomes demonstrate a notable decrease in overall losses in the blade cascade due to this end-wall contour. For all tested Mach numbers, the average loss reduces by 4% at 0 degrees of incidence, with a 0.9° decrease in deviation angle, and an 8% average loss reduction at 5 degrees of incidence, accompanied by a 1.5° decrease in deviation angle. Qualitative findings suggest that the mechanism by which the non-axisymmetric end-wall contour reduces blade cascade losses involves the enhancement of pitchwise pressure gradients. This enhancement redirects low-momentum fluid toward the blade's suction surface, diminishing its reach in the pitchwise direction. However, it also intensifies loss in the low-momentum fluid core. Additionally, the presence of a convex feature in the rear section of the blade passage mitigates pitchwise static-pressure gradients, effectively ameliorating secondary flows and preventing their further decline. Moreover, it was observed that the control effectiveness of the end-wall contour on corner separation increases with higher Mach numbers.

Improved prediction of turbomachinery flows using Reynolds stress model with [formula omitted] transition model

Transition is a prominent flow phenomenon in turbomachinery, especially at low and moderate Reynolds numbers. The turbomachinery internal flow often encounters complex vortex structures, typically accompanied by strong turbulence anisotropy and non–equilibrium. Conventional Reynolds–averaged Navier–Stokes (RANS) modeling with linear eddy–viscosity models (LEVMs) may fail short in capturing certain complex flow features. The Reynolds stress model (RSM) based on a higher turbulence closure level has more potential to reasonably consider turbulence anisotropy and non–equilibrium for such complex flows within turbomachinery than the LEVMs. However, the current RSM lacks an effective representation of transition phenomena. This paper presents an improved prediction of turbomachinery flows using RSM with transition modeling. The transition effects are incorporated into the RSM (SSG/LRR–ω model) via Menter's γ transport equation. The blending of the SSG/LRR–ω model with the γ transition model (SSG/LRR–ω–γ model) is firstly validated based on the T3 series flat plate cases with zero pressure gradient (T3A, T3B, T3A–) and variable pressure gradients (T3C1–C5). Comparisons with experiments show that the SSG/LRR–ω–γ model predicts corner separation flows with transition more accurately, although there is still room for further improvement. Then, the performance of the SSG/LRR–ω–γ model is assessed in a linear highly loaded prescribed velocity distribution (PVD) cascade and a NACA65 K48 compressor cascade. Comparisons with experiments indicate that, in both compressor cascades, the SSG/LRR–ω–γ model can predict more accurately corner flows with boundary layer transition on suction surfaces than the SSG/LRR–ω and SST model. Nevertheless, it still exhibits a tendency to overestimate the size of corner separation, especially in cases characterized by strong adverse pressure gradients. Comparisons of skin friction and velocity profiles predicted by the SSG/LRR–ω model with and without transition modeling illustrate the rationale behind the improved prediction. Further improvement of the SSG/LRR–ω–γ model for predicting turbomachinery corner flows should be conducted in future.

Direct Numerical Simulation of Transitional and Turbulent Flows Over Multi-Scale Surface Roughness—Part II: The Effect of Roughness on the Performance of a High-Pressure Turbine Blade

Abstract Turbine blades generally present surface roughness introduced in the manufacturing process or caused by in-service degradation, which can have a significant impact on aero-thermal performance. A better understanding of the fundamental physical mechanisms arising from the interaction between the roughness and the turbine flow at engine-relevant conditions can provide insights for the design of blades with improved efficiency and longer operational life. To this end, a high-fidelity numerical framework combining a well-validated solver for direct numerical simulation and a second-order accurate immersed boundary method is employed to predict roughness-induced aero-thermal effects on an LS89 high-pressure turbine (HPT) blade at engine-relevant conditions. Different amplitudes and distributions of surface roughness are investigated and a reference smooth-blade simulation under the same flow conditions is conducted for comparison. Roughness of increasing amplitude progressively shifts the blade suction side boundary layer transition upstream, producing larger values of the turbulent kinetic energy and higher total wake losses. The on-surface data-capturing capabilities of the numerical framework provide direct measurements of the heat flux and the skin friction coefficient, hence offering quantitative information between the surface topology and engineering-relevant performance parameters. This work may provide a benchmark for future numerical studies of turbomachinery flows with roughness.

Statistical investigations of profile error impact on flow and performance of a low-pressure turbine cascade

Profile error impacts on turbomachinery flow and blade performance have been attracting widespread attention. In the study, the characteristics of profile error of about one thousand real low-pressure turbine blades are extracted. Sensitivities of total pressure loss coefficient (ζ), outflow angle (β), and Zweifel lift coefficient (zw) of the blade to the basis modes of profile error are calculated. Flow solutions of the blades considering specified basis modes with high sensitivities illustrate that profile error contributes much to the variations of transition onset and flow acceleration on the suction side and flow mixing intensity in the wake. Uncertainty quantification of performance changes is then implemented by the method of moment (MM) using second-order sensitivities. With only 5% computational cost of that by Monte Carlo simulation (MCS), the MM-based statistical results are close to MCS ones with maximum relative error not exceeding 1.07%. The statistical results reveal that the variations of both β and zw are linearly dependent, whereas the variation of ζ is nonlinearly dependent on profile error. As the variation range of profile error increases, the standard deviation and skewness increase, indicating that the performance is more dispersive and the nonlinear dependence of ζ on profile error is intensified. Finally, the MCS flow fields are analyzed. Statistical shear stress near the leading edge and transition onset, statistical boundary layer momentum thickness near transition onset, statistical intermittency near transition onset, and statistical entropy in the wake are more considerable. The impact mechanisms of profile error on turbine flow and performance changes are demonstrated.

NUMERICAL ANALYSIS AND DIFFUSER VANE SHAPE OPTIMIZATION OF A RADIAL COMPRESSOR WITH THE OPEN-SOURCE SOFTWARE SU2

In recent years, the usage of open-source computational fluid dynamics tools is on a rise both in industry and academia. SU2 is one of these open-source tools. Unlike other open-source alternatives, SU2 is equipped with boundary condition types, solvers and methods that are especially developed for the analysis and design of turbomachinery. The aim of this work is to explore and investigate the capabilities of SU2 in the prediction of performance parameters of radial compressors. Two different single stage shrouded compressor geometries, one with a vaneless diffuser and the other with a vaned diffuser have been investigated with steady state CFD. The compressors were designed by MAN Energy Solutions Schweiz AG. Computational results with SU2 showed a satisfactory agreement with both the experimental data and reference CFD solutions obtained with Fidelity Flow, which is formerly known as Numeca Fine TURBO. Only at the relatively higher mass flow rates the difference between references and SU2 were higher compared to other operating points. After performance parameters were successfully calculated with SU2, the optimization tools that come with SU2 were also used. A 2D adjoint optimization study on the vane of the vaned diffuser was carried out. The study was carried out at a single operating point that is close to choke conditions. The loss generated by the large separated flow region at the suction side of the diffuser vane was reduced by 0.55 % in the optimized geometry using minimal modifications on the existing vane geometry to keep the performance of the compressor intact at other operating points. However, the resulting modification increased the total pressure loss by 0.86 % at one of the design operating points. This performance penalty could be due to the discontinuity in the vane geometry generated by the optimizer. Overall, the study shows that SU2 has the basic numerical schemes and models that are required for the analysis of radial turbomachinery flows and geometry optimization.

Particle image velocimetry in the impeller of a centrifugal pump: A POD-based analysis

The flow field within the channels of a centrifugal pump impeller is usually complex, containing turbulent structures in a wide range of time and length scales. Identifying the different structures and their dynamics in this rotating frame is, therefore, a difficult task. However, modal decomposition can be a useful tool for detecting coherent structures. In this paper, we make use of proper orthogonal decomposition (POD) of time-resolved flow fields in order to investigate the flow in a centrifugal pump. For that, we carried out experiments using time-resolved particle image velocimetry (TR-PIV) in a pump of transparent material operating at different conditions and obtained the statistical characteristics of the turbulent flow from phase-ensemble averages of velocities and turbulent kinetic energy. The results reveal that at the pump’s best efficiency point (BEP) the flow is well-organized, with no significant flow separation. For flow rates below the BEP, flow separation and vortex structures appear in the impeller channels, making the flow unstable. At flow rates above the BEP, intense jets appear close to the suction blades, while small instabilities occur on the pressure side. The POD analysis shows that at low flow rates, the flow is dominated by large-scale structures with intense energy levels, while at the BEP and higher flow rates, the flow is dominated by small-scale structures. Our results shed light on the turbulence characteristics inside the impeller, providing relevant information for reduced-order models capable of computing the flow in turbomachinery at much lower costs when compared to traditional methods.