Total Pressure Loss Coefficient Research Articles

Numerical studies on shock wave and boundary layer interaction in the high-load turbine rotor cascades

Design of transonic high-load turbine is the essential approach to improve thrust-to-weight ratio of gas turbine engines. The shock wave and boundary layer interaction (SWBLI) in high-load turbine is one of the important unsteady sources and the main source of loss in turbine stages. In order to study mechanism of SWBLI in high-load turbine, rotor blades with loading coefficients range from 1.6 to 2 were designed and Delayed Detached Eddy Simulations (DDES) were carried out. The influences of loading coefficient, exit isentropic Mach number and incidence angle on characteristics of shock wave, SWBLI as well as flow details in the blade passages were compared. Results indicated the shock wave was generated and enhanced as exit isentropic Mach numbers increased leading to the increase of 2.8% in the total pressure loss coefficient. There was not significant difference in the total pressure loss coefficient as loading coefficient increased, however the shock wave was stronger and the separation bubble was longer at the loading coefficient of 1.6. Meanwhile, the influence of strong separation near the leading edge of suction surface induced by variation of incidence angles on the characteristics of the shock wave and the loss was also obvious.

Experimental study on multi-angle pulsed jet actuators for controlling corner separation flow in a compressor cascade

Three-dimensional corner separation caused by the aerodynamic design of high-load compressor cascades is the main factor that limits their performance enhancement. Because of the unsteady characteristics of large-scale flow separation, injecting unsteady jets into the flow field has become a new flow control method. In this study, the essential flow characteristics of a compressor cascade were obtained through numerical simulations. Subsequently, the effect of pulsed jet angle on the aerodynamic performance of the cascade was experimentally investigated under different operating conditions using pulsed jet actuators. The results showed that at the design incidence angle of 0° and near-stall incidence angle of +4°, the starting point of the corner separation is located at 21% and 9% of the axial blade chord, respectively. The pulsation frequency downstream of the cascade trailing edge is 1454.5 Hz. Under the design conditions, the 20° pulsed jet exhibits the best control effect, reducing the total pressure loss coefficient by 13.4% compared with that of the original cascade. Moreover, pulsed jets demonstrate strong applicability under variable operating conditions. At near-stall incidence angles, when the jet angle is 30°, the loss-reduction ability of the pulsed jet is maximized, decreasing the loss by 25.2%. Pulsed jets reduce the cascade loss primarily by suppressing the secondary flow at the endwall and discretizing the vortex structures in the channel.

Multi-objective design optimization of a transonic axial fan stage using sparse active subspaces

In this paper, a multi-objective optimization strategy for efficient design of turbomachinery blades using sparse active subspaces is implemented for a turbofan stage design. The proposed strategy utilized sparse polynomial chaos expansion on a limited dataset to generate a function from which the differential and the covariance matrix can be obtained. Active subspace was used to compute the active variables via singular value decomposition and a hybrid polynomial correlated function expansion was used to construct a surrogate model on the active subspace. Coupled with freeform method and multi-objective genetic algorithm, an automated optimization loop was run at a single operating condition. An improvement in stage efficiency and total pressure ratio of 2.97% and 1.15% was achieved for the optimum design compared with the baseline. Additionally, total pressure loss coefficient decreased by 5.88%, exit flow angle by 34.65% and shock strength by 5.32%. The coupled effect of change in stagger angle, forward sweep, forward lean, and chord length reduced the recirculation at the hub, and blockage at the shroud due the tip leakage flow by decreasing the blade loading. The threshold value hyperparameter was found to be the most influential and must be accurately determined.

Extraction of geometric features and analysis of flow mechanism of high loaded compressor airfoils at low Reynolds number

When the aircraft cruising at high altitude, the aerodynamic performance of the compressor sharply decreases due to the enhanced boundary layer separation loss. Therefore, it is of great significance to design the high-performance airfoils suitable for low Reynolds numbers (Re) to improve the aerodynamic performance of compressor. In this paper, numerical simulations were carried out on a high-loaded subsonic compressor cascade V103 at low Re. Meanwhile, a multi-objective optimization of blade profile was performed to get optimal solutions, and data mining techniques were applied to extract valuable design knowledge from the optimal database. The results show that the Pareto-optimal airfoils achieve a maximum reduction of 27.32% in total pressure loss coefficient under design condition and 36.71% under near stall condition. Compared with the original airfoil, the curvature distribution law of the camber line with larger values at both ends and smaller values in the middle is advantageous for the performance of the compressor cascade under design condition. This distribution pattern causes a forward shift for the loading distribution of airfoils, accompanied by an earlier transition onset. Additionally, it reduces adverse pressure gradient at the rear part of suction surface, which suppresses the development and growth of laminar separation bubble and delays the separation of turbulent boundary layer. Under near stall condition, maintaining larger leading edge angle and uniform flow diffusion is beneficial for improving the performance of compressor cascade. This is because the larger leading edge angle reduces incidence angle, while the uniform flow diffusion inhibits a great flow separation.

CFD verification of total pressure loss coefficient for gas-liquid two-phase flow

Traditionally, we design gas-liquid two-phase flow systems with pressure loss that is assumed by homogeneous flow. However, in real phenomena, the flow rate of water or air through a system cannot be correctly predicted unless we evaluate by the total pressure loss. there are currently few methods for measuring total pressure loss, and the value of total pressure loss is not yet clearly. Therefore, we devised a method to measure total pressure loss by pump Q-H characteristic and succeeded in measuring total pressure loss indirectly. We compared the test results with the analysis to clarify the prediction accuracy of the analysis using commercial codes. We apply three two-phase flow models for the simulation: a VOF model, a homogeneous flow model, and a two-fluid model. The simulation results of comparison between each fluid model indicated that the VOF model predicted with the best accuracy, with the error of 46% compared to the test results. This study provides a benchmark for the current commercial code, and the prediction accuracy of the different models provides a direction for future improvements in the analysis.

Aerothermal Optimization of Film Cooling Hole Locations on the Squealer Tip of an HP Turbine Blade

Abstract This paper presents a time-efficient method to optimize the positions of film cooling holes on a gas turbine blade's squealer tip for cooling and aerodynamic performance. A computational approach is employed for the optimization, including validations against experiments. Five discrete film cooling holes are considered, and two different blowing ratios of 0.4 and 1.0 are studied. The positions of cooling holes on the tip along the tangential direction are varied as the input parameters of optimization. The multi-objective optimization uses an algorithm with an artificial neural network for fast fitness function predictions. The best cooling configuration found by the optimization achieves a 13.43% reduction in total heat flux and a 0.4% increase in aerodynamic loss when the blowing rate is 1.0. Including the casing relative motion in the computations results in a total pressure loss coefficient increase of about 8 % for both blowing ratios. For M=1.0, imposing the casing's motion results in a 10.2% reduction in total heat transfer to the tip compared to the stationary casing. For the lower blowing rate of 0.4, the total heat flux reduction to the tip is 12.0% because of imposed casing motion. Hence, the cooling effectiveness can be improved by employing the particular position optimization method presented in this study. The results suggest that experimental and computational heat transfer studies on cooled turbine blade tips, especially in cascade arrangements, need to consider the relative motion of the blade tip.

Flow mechanism of a highly loaded low pressure turbine cascade with integrated-optimized end wall contouring and root lean

The end wall loss of modern highly loaded low pressure turbine (LPT) has been greatly increased, due to the enhanced secondary flow loss and boundary layer separation loss. Thus, it is of great significance to develop effective flow control strategies to improve the end wall flow condition and aerodynamic performance of modern LPT. This research carried out a numerical investigation on the coupled flow control strategy, which combined non-axisymmetric end wall contouring (NEC) and root tangential lean (RTL), based on a highly loaded LPT cascade (Zweifel = 1.59). Meanwhile, the optimization process was used to get the optimal design parameters of the coupled method NEC&RTL. The results indicate that the optimal coupled configuration can reduce the total pressure loss coefficient by 12.68% and the non-dimensional secondary kinetic energy by 23.91%. Compared with the reference cascade without modification, the coupled method is found to improve the end wall flow conditions: the passage vortex is weakened both in size and strength, mainly attributed to the smaller cross-passage pressure gradient resulting from NEC; the closed separation bubble near end wall and the three-dimensional separation flow before trailing edge are eliminated, due to the great downward pressure gradient near end wall resulting from RTL; and the counter vortex is eliminated and the slender back flow is weakened under the additional coupling flow control effect of NEC&RTL. Therefore, the coupled flow control method can not only highlight the advantages of the independent methods, but also induce external flow control superiorities, demonstrating the application prospect of the coupled flow control strategy on the highly loaded LPT.

Performance Evaluation of Damaged Blades Using Numerical Simulation and Machine Learning

The formulation of maintenance strategies for gas turbines requires an accurate and rapid assessment of damaged blade performance. Variety and uncertainty in the three-dimensional geometric deformation of gas turbine vanes and blades increase the difficulty of degradation analysis. In this paper, the effects on the aerodynamic performance of partial damage to turbine nozzle guide vanes are investigated experimentally and numerically. A methodology for constructing a deformation–degradation model for high-pressure gas turbine vanes/blades is proposed based on machine learning. A database consisting of 670 numerical simulation results with various deformation patterns is constructed for the vane cascade. A mapping is constructed between the matrices defining geometric deformations and the total pressure loss coefficient of the cascades based on a convolutional neural network model. The accuracy of this deformation–degradation model is verified against experimental results. We find that deformation on the suction side of the blade has the most significant influence on aerodynamic performance. The deformation–degradation model constructed for a specific cascade provides a rapid evaluation of aerodynamic performance based on rough geometrical measurements without additional experiments or numerical simulations, providing a potential tool in the precise repair of gas turbines.

Effect of Reynolds number on the aerodynamic performance of highly loaded helium compressor cascade in high temperature gas-cooled reactor

Closed Brayton cycle power plant with High Temperature Gas-cooled Reactor is one of the most promising solutions for reduction in carbon emission. Helium is being used as a coolant in power conversion unit of these power plants. The output power of these systems is regulated by varying the density of helium in the power conversion unit in such a way that the compressor is required to operate over a wide range of Reynolds numbers. In this paper, the vortices structure in conventional and highly loaded helium compressor cascade are analyzed at low Reynolds numbers. Accordingly, the evolution mechanism of vortex structure in the highly loaded helium compressor cascade was studied and validated by experiments. It is concluded that with the decrease in Reynolds number, the vortex size and total pressure loss coefficient increase gradually. The loss weight factor of the passage vortex (PV) and concentrated shedding vortex (CSV) in the highly loaded helium compressor cascade increases by 262 % and 53.8 %, respectively, in comparison with the conventional compressor cascade.

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.

Proof of concept study on a self-driven pulsed jet on a compressor stator blade by numerical simulation

This study presents a new concept of self-driven pulsed jet flow control on a compressor stator blade. This passive unsteady flow control method has the advantage that neither external flow nor electrical source is needed. This study’s preliminary proof-of-concept study on a low-speed compressor stator blade is performed using numerical simulation. When the pulsed jet frequency is 100 Hz, the optimum control performance and control efficiency are reached, and the total pressure loss coefficient is reduced by 8.9%. As the valve’s rotational speed increases, the pulsed jet’s momentum coefficient decreases gradually. The analysis of the unsteady characteristics of the self-driven pulsed jet shows that the jet velocity is close to a periodic square wave signal, and the typical reduced jet velocity ranges from approximately 0.15 to 0.8. Moreover, the time-averaged driving torque on the valve depends on the rotational speed and is relevant to the self-starting and self-driven characteristics of this passive flow control method. Under different bearing resistance torque, the self-driven valve behaves differently in three cases that can self-start and be self-driven, cannot self-start but can be self-driven, and cannot self-start nor be self-driven.

Parametric optimization applied to design a high-performance vaneless-diffuser for CO2 centrifugal compressor

ABSTRACT Carbon-dioxide (CO2) centrifugal compressors are machines with high potential of usage in power generation plants and oil industry as it achieves high thermal efficiency in the Brayton cycles and contributes to oil production through the capture and storage of CO2 (CCS) in EOR (Enhanced Oil Recovery) systems. High levels of static pressure at the outlet are desired for EOR applications, which are usually obtained through the insertion of vanes in the diffuser. This work intends to increase vaneless-diffuser static pressure recovery by modifying only its meridional profile, ensuring a broader range of off-design operation when compared to vaned diffusers and attending to the fluctuations of mass flow and rotation expected in EOR practical applications. Therefore, a parametric optimization through surrogate model coupled to CFD was performed with three different objective functions that were submitted to single-optimization through the NSGA-II method: Maximize total-to-total polytropic efficiency, minimize total pressure loss coefficient or maximize static pressure recovery coefficient. Additionally, a sensitivity analysis was conducted using Morris Elementary Effects and SS-ANOVA. The results indicated that the optimized geometries increased the total-to-total polytropic efficiency by 2.9%, reduced the total pressure loss coefficient by 24.0% and increased the static pressure recovery coefficient at the design point by 11.4%, which is discussed in detail after a careful phenomenology assessment. The strategy adopted in the present work through a combination of Sensitivity Analysis, surrogate models and CFD increased the vaneless-diffuser static pressure recovery without the need of inserting vanes in the diffuser, which avoid instabilities in the equipment and would restrict its range of off-design operation.

Aerothermal optimization of turbine cascade squealer tip with non-uniform squealer height

The squealer tip has significant influence on both the aerodynamic and heat transfer characteristics of the high-pressure turbine blade. However, due to the complexity of parameterization and meshing of the squealer and the complicated flow structure within the over-tip region, the existing squealer designs in the open literature have constant squealer heights. In this paper, the design space to the squealer height with non-uniform squealer height is extended and the new flow features it may bring are investigated. A parameterization system specifically designed for the non-uniform squealer height using five control parameters is implemented to automatically generate the geometry and hybrid meshes. Combining it with the multi-objective optimization system using genetic algorithms, a transonic turbine cascade squealer tip is optimized employing Reynolds-averaged Navier–Stokes k–ω shear stress transport model. The main objective of this study is to obtain a squealer configuration with the lowest total pressure loss coefficient and heat transfer coefficient. The optimum configuration with non-uniform squealer height achieves improvements in both the aerodynamic efficiency and the heat transfer performance, relative to the baseline conventional squealer tip geometry with the constant squealer height. Additionally, this work demonstrates that a flow structure in which the main flow forms a “blanket” below the leakage flow in the squealer is beneficial for aerothermal performance, especially reducing heat transfer losses, which provides valuable insight into the squealer tip design of advanced high-pressure turbines.

Numerical study on the interaction between rotating detonation wave and turbine stator blades with multicomponent mixtures

Aiming at the coupling problem between rotating detonation combustor and turbine components, this paper uses methane recuperation gas as fuel and adopts two-dimensional numerical simulation to study the working process after coupling between rotating detonation combustor and turbine stator blades. The results show that with the increase of methane conversion, the propagation mode changes from single wave mode to unstable single wave mode. Under the two modes, after the oblique shock wave sweep, the turbine stator blade outlet Mach number suddenly rises to 2.5. In addition, the flow field structure of a single turbine blade is also analyzed, and it is found that after oblique shock wave sweep, the flow field structure changes, and boundary layer separation occurs at the trailing edge of turbine stator blades. Turbine stator blades have a certain inhibition effect on the pressure oscillation at the combustion chamber outlet, and the disturbance generated by detonation wave extinguishing in unstable single-wave mode will be transmitted through turbine stator blades to the outlet. With the increase of fuel activity, turbine stator blade pressure loss gradually increases. The total pressure loss coefficient of single-wave mode is 9%, and the total pressure loss coefficient of unstable single-wave mode is 4%–6% higher than that of single-wave mode.

Experimental and numerical investigation of an S-shaped duct using pulsed jet actuators with multiple angles

The complexity and twistiness of the ultra-compact S-shaped ducts with high curvature aggravate the internal flow separation, which affects the intake efficiency. An advanced flow control technology, pulsed jet actuators (PJAs), was implemented to improve the flow condition in the S-shaped duct at a Mach number of 0.4. Based on the considerable control effect obtained from the experimental studies, the influence of the pulsed jet angles on the S-shaped duct was further investigated by utilizing the unsteady Reynolds-averaged Navier–Stokes equations (URANS). The jet angles as a research parameter were set to 15°, 30°, 45°, 60° and 75°, respectively. Moreover, steady jet schemes were adopted to reveal the unsteady flow control mechanism of the pulsed jet. The results showed that the overall performance of the S-shaped duct with the pulsed jet was improved, mainly attributed to the additional kinetic energy and dispersion. The total pressure loss coefficient and distortion index at the aerodynamic interface plane (AIP) were decreased by 7 % and 2.9 % when the pulsed jet angle was 30°. Furthermore, the temperature distortion was significantly improved, with a maximum temperature rise rate of −87.59 K/s. It is concluded that the suitable penetration of the mainstream is the prerequisite for achieving the best flow control effect.

Оцінка впливу канавок у дифузорі на втрати повного тиску

The subject of the study is the total pressure in the diffuser channel with passive control of the boundary layer. The object of this research is a diffuser channel with riblets. The mechanism of total pressure reduction due to the presence of longitudinal grooves on the streamlined surface is related to the strengthening of viscous forces acting in the transverse direction caused by them. These forces generate a relatively calm flow inside the riblets, which pushes out the turbulent flow, which causes surface friction, pushing it away from the surface. Thus, the process of generation and growth of elongated wall turbulent formations. In addition, in the presence of riblets, a change in the pressure field in the transverse direction is possible, as well as the creation of a quasi-two-dimensional flow in the immediate vicinity of the wall. In this work, the aim of research to investigate the influence of riblets on total pressure loss in the diffuser channel with the purpose of evaluating the influence of riblets on the change in the nature of the flow and the change in pressure, using numerical modeling. To study the nature of the flow in the diffuser channel, we resolved the following tasks: a diffuser channel was modeled with the following dimensions: length L=50 mm, hydraulic diameter at the inlet X=25 mm, hydraulic diameter at the outlet Y=40 mm, riblets diameter d=4 mm, number of riblets n =4 and n=8. The study was compared with a diffuser with the following dimensions: length L=50 mm, diameter at the inlet X=25 mm, diameter at the outlet Y=40 mm. To research the flow in the diffuser, the method of numerical experiment was chosen. An unstructured calculation grid was used in the diffuser for the research. The calculation of the turbulent gas flow was performed by numerically solving the averaged Navier-Stokes equations. During the numerical simulation of the flow, the Reynolds stress turbulent viscosity model will be used. As a result, using a diffuser with riblets increases the coefficient of hydraulic resistance compared to a smooth diffuser. The resistance reduction mechanism does not work due to the relatively short length of the diffuser and small longitudinal vortices of the riblets scale do not have time to form. There are pressure increasing instead of laminization of the flow and pressure decreasing. Scientific novelty - received new data on the effect of riblets in the diffuser channel on the total pressure loss coefficient. Practical significance - the obtained results can be used to control the boundary layer in diffuser channels (in diffuser compressor grates) to improve their characteristics.

A characterization of unsteady effects for transonic turbine airfoil limit loading

To better understand the effect of vortex shedding on the nature of the trailing edge shock system during airfoil limit loading a computational fluid dynamic investigation was performed for a transonic turbine airfoil at sublimit, limit and supercritical conditions. Four modeling strategies were employed: steady state RANS, unsteady RANS, DDES, and turbulence model free. The influence of transient modeling approaches on the predicted mass-flow averaged total pressure loss coefficient, mass-flow averaged flow angles, and on the limit loading pressure ratio were found to be insignificant with the exception of the URANS model during critical loading. It was found that the URANS modeling approach failed to predict the transition from near trailing edge dominated vortex formation to base pressure vortex formation resulting in a drastic rise in predicted total pressure loss. Surface isentropic Mach distributions were predicted similarly for all modeling strategies, with the exception of the trailing edge base pressure region and points of shock impingement along the suction surface. A detailed review of the boundary layer states at the trailing edge was performed. It was found that all of the modeling approaches predicted laminar boundary layer profiles along the pressure surface trailing edge and turbulent profiles along the suction surface. The predicted base pressure distributions were also reviewed, showing the base pressure to decrease with increasing pressure ratio. The unsteady simulation approaches consistently predicted lower average surface pressures than the steady state RANS simulations. Qualitative images of the numerical Schlieren contours were presented and reviewed showing large differences in the prediction of vortex shape, size, and subsequent shock influence.

OH* Chemiluminescence Characteristics of the RP-3 Fueled Dual-Swirl Direct-Mixing Combustor.

The flame structure characteristics of the RP-3 fueled dual-swirl direct-mixing combustor are studied experimentally. The flame shape is marked by the OH* radical, which is captured by a CMOS camera with an image intensifier. The flow fields and spray distributions are obtained by particle image velocimetry. The variation of pilot/main-stage flame structure with global fuel-air ratio (FAR), fuel ratio (FR), pilot/main-stage swirl numbers (Sp and Sm), and Δp/p (total pressure loss coefficient) is further investigated. The typical flame structure, consisting of two main-stage combustion zones (pilot-stage and main-stage combustion zones), is first analyzed. Then, according to the relationship between integral OH* intensity and global heat release rate, increasing swirl number will weaken the effect of strain rate on OH* chemiluminescence. The global FAR has little impact on the flame structure, while modifying the FR will alter the flame mode. The influence of the Sp on the flame structure is more significant than that of the Sm. Within the range of experimental conditions, the greater the swirl number, the smaller the Δp/p is required to obtain the maximum OH* radical intensity. The addition of non-swirl flow in the main-stage swirler can improve the stability of the pilot-stage flame. The equivalent swirl number is further evaluated by the neural network.

Numerical investigation of tip clearance flow in a variable geometry turbine with non-uniform partial clearance

Abstract In variable geometry turbine vanes, tip clearance height and shape vary with the rotation of the vane, which affect the aerodynamic performance significantly. However, these issues are rarely considered in published studies. The current paper investigated the flow field features of transonic variable geometry turbine vanes with non-uniform partial clearance induced by the vane rotating. The results show that: The influence of guide vane rotation on the clearance height and its distribution cannot be ignored. At the same turning angle, the maximum clearance difference is up to 0.79 mm (0.8 % vane height). The height and shape variation of the non-uniform clearance leads to the change in the leakage flow rate, secondary flow structure, and aerodynamic loss of the variable guide vane. Under the combined effect of pressure difference on both sides of the clearance, axial and circumferential non-uniformity of clearance height, the total pressure loss coefficient is up to 9.44 % when the turning angle is −10°. The effect of the pivot on the clearance flow was also analyzed. The pivot increases the pressure in the gap flow field and reduces leakage flow velocity. However, a backflow region appears at the suction side of the pivot, which increases the aerodynamic losses.

Quasi-three-dimensional loss prediction model of subsonic compressor cascade based on bidirectional long short-term memory networks and multi-head self-attention

The prediction of compressor cascade loss is a crucial aspect of compressor design. Flow separation is an important flow structure and the main source of loss in subsonic cascades. In order to capture the flow separation and accurately evaluate flow loss, a data-driven quasi-three-dimensional (quasi-3D) subsonic compressor cascade loss prediction model based on bidirectional long short-term memory (BiLSTM) and multi-head self-attention is proposed. The model contains four sub-models to predict the pressure, temperature, axial velocity, and total pressure loss coefficient in two-dimensional slices along the axial direction, using Mach number, curved blade angle, solidity, camber angle, and incidence as inputs, respectively. For the purpose of adapting to cascade geometrical change, geometric reformulation is adopted before the model training. The model is trained and tested by validated computational fluid dynamics results, which contain symmetric separation and asymmetric separation samples. It is proved that the model is able to accurately predict flow parameters value in each slice. Then, four typical cases are mainly discussed, which shows that the model can effectively capture the characteristics of flow separation formation and development. Afterward, different models are compared, and it is found that the BiLSTM with multi-head self-attention model achieved the lowest mean squared error, which is because of its outstanding predicting ability in asymmetric separation cases. The work of this paper indicates that the quasi-3D loss prediction model proposed in this paper will be beneficial to the flow separation structure rapid prediction and cascade loss accurate evaluation in compressor design.