Compressor Performance Prediction Research Articles

An Efficient Uncertainty Quantification Method Based on Inter-Blade Decoupling for Compressors

Abstract A compressor usually contains multiple blade rows, and its uncertainty dimensionality grows proportionally with the number of blade rows, leading to a rapid increase in required sample size for uncertainty quantification analysis. This paper proposes a new model decomposition method based on inter-blade decoupling, by analyzing uncertainty propagation in compressors. Traditionally, a surrogate model with uncertainty variables of all blade rows as input is directly established and the dimensionality is high. To solve this problem, this study decomposes the compressor domain into subdomains, each containing one blade row. For each subdomain, a submodel introduces aerodynamic uncertainties at the interfaces connecting different subdomains. The dimensionality of a submodel is roughly equal to the uncertainty factors in a single row, significantly reducing the required sample size. The uncertainties in the rotor and stator blade rows of a one-stage compressor are investigated to verify this method. Using principal component analysis and machine learning, the projection amplitudes of the interface aerodynamic flow field onto the principal modes are extracted, and submodels are established. Results show that the original 25-dimensional model can be decoupled into a 13-dimensional submodel for the rotor and a 16-dimensional submodel for the stator, reducing the required sample size from 600 to 90 with similar accuracy. This model decomposition method greatly reduces the cost of predicting compressor performance with uncertainty, laying a foundation for comprehensive analysis and effective control of uncertainty factors in engineering applications.

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.

Application of a semi-empirical modelling approach to a Two-Stage Rotary CO2 compressor

2-stage Rotary CO2 compressors are currently in use in refrigeration and heat pump applications, including occasionally including vapour injection. The need of correctly assessing the compressor’s efficiency by means of a computationally fast model arises for cycle performance prediction and optimization. In response, an innovation on an existing semi-empirical modelling approach has been introduced in this paper to describe 2-stage Rotary compressors. The model has been validated with experimental data of a CO2 compressor available in open literature, giving maximum errors of around ±3% for mass flow and ±1.82% for power prediction. Calibration of a simplified 1-stage model is also performed to analyse to which extent it is necessary to complexify the model, and to identify key or neglectable modelling elements. It is found that doing this increases the error of the compressor performance prediction. Model analysis has been done to identify the inefficiencies introduced by each physical effect considered.

Performance prediction and design optimization of a transonic rotor based on deep transfer learning

Deep transfer learning is frequently employed to address the challenges arising from limited or hard-to-obtain training data in the target domain, but its application in axial compressors has been scarcely explored thus far. In this paper, a multi-objective optimization framework of a transonic rotor is established using deep transfer learning. This framework first pre-trains deep neural networks based on the peak efficiency condition of 100% design speed and then fine-tunes the networks to predict the performance of off-design conditions based on the small training dataset. Finally, the design optimization of the transonic rotor is carried out through non-dominated sorting genetic algorithm II. Compared to neural networks that are trained directly, transfer learning models can achieve higher prediction accuracy, particularly in scenarios with small training datasets. This is because the pre-trained weights can offer a better initial state for transfer learning models. Moreover, transfer learning models can use fewer samples to obtain an approximate Pareto front, making the optimized rotor increase the isentropic efficiency at both peak efficiency and high loading conditions. The efficiency improvement of the optimized rotor is attributed to the reduction of the loss associated with the tip leakage flow by adjusting the tip loading distribution. Overall, this study fully demonstrates the effectiveness of transfer learning in predicting compressor performance, which provides a promising approach to solving high-cost compressor design problems.

Enhancing robustness and accuracy of supercritical CO2 compressor performance prediction in closed Brayton cycles: A thermodynamic properties-based numerical method

The optimization of the closed Brayton cycle results in the inlet condition of supercritical CO2 (sCO2) compressors being near the critical point. However, this proximity leads to decreased robustness and accuracy during numerical simulations. To enhance computational fluid dynamics (CFD) robustness and accuracy, a method utilizing Cubic B-spline smoothing is proposed to reconstruct thermodynamic properties using property look-up tables (LUTs). After revealing the characteristics and action mechanism of sCO2 thermodynamic properties on the solver, effects of the method on solution robustness, accuracy, and the relationship with resolution are systematically investigated. The results demonstrate that this method effectively addresses the deficiencies arising from the inability of original LUTs to adapt to the high resolution and the presence of pseudo-convergence. The residuals achieve the convergence requirement in half the iteration steps compared to the original method, resulting in an over one-fold reduction in calculation time. The solution robustness is significantly enhanced by over tenfold, and the calculation accuracy is improved. This improvement increases the sensitivity of friction loss and secondary flow loss to drastic changes in thermodynamic properties. As a result, a highly robust and accurate performance prediction scheme for centrifugal compressor components is established.

Co-optimization method for injection strategy of underground natural gas storage integrating aboveground and underground parts

Underground natural gas storage (UNGS) is an important infrastructure to overcome the imbalance between supply and demand of natural gas. During the gas injection period, gas is compressed aboveground and then transported to injection wells for injection into the underground reservoir. Reasonably formulating the compressor operation scheme and injection scheduling scheme is a crucial step. However, “Compressor-pipeline-injection well-reservoir" constitutes a synergistic whole. More efficient and practical operation schemes can only be proposed through the establishment of an optimization method that coordinates the entire gas injection process. A novel two-layer multi-objective collaborative optimization method is proposed to optimize the monthly operation scheme during the injection period. The upper-layer model focuses on optimizing injection volume allocation to minimize reservoir pressure deviation and total compressor energy consumption. The under-layer model optimizes the compressor operation scheme by considering the multi-stage compression (MC) and clearance adjustment (CA) equipped in reciprocating compressors. The synergistic relationship during the injection process is integrated into the traditional model through the interlayer iteration method. Furthermore, a high accuracy and efficiency proxy model based on machine learning algorithms (ML) is developed to predict compressor performance. The results of the case study indicate that, compared with traditional optimization method and the two-stage co-optimization method, the proposed method achieves an average energy consumption reduction of 9.54 × 106 kWh (10.79%) and 2.12 × 106 kWh (2.4%) respectively. The effective utilization of CA and MC can significantly alleviate the deviations between displacement and processing requirements, eliminating 22,120.5 × 104 Nm3(85.62%). The two-layer co-optimization method effectively integrates the aboveground and underground parts of UNGS, providing a solution for safe and efficient operation during the injection period.

An Improved Body Force Model to Simulate Blade Loading Variation Under Various Operating Conditions in Transonic Axial Flow Compressor

Abstract In order to analyze the unsteady aerodynamic response of compressors with limited computational resources, the development of the body force (BF) model has been pursued for decades. This model can simulate the effects of the compressor blade on the flow by applying a three-dimensional (3D) force field. Indeed, the accuracy of the BF model simulation highly depends on the construction strategy of BF field. However, for a transonic compressor, the influence of shock waves on the spatial distribution of blade loading is significant, which makes the construction of an appropriate BF field challenging. To solve this issue, an accurate BF model is proposed to describe the effects of the blade on the flow with high spatial fidelity in transonic compressors. For verification, a typical transonic fan rotor, NASA-Rotor 67, is selected. The numerical results calculated by the new BF model are compared with those obtained by the Reynolds-averaged Navier–Stokes (RANS) solver. According to the results, the new model can accurately capture the distribution of blade loading in the transonic blade channel at various operating conditions, eventually leading to a good prediction of compressor performance and the radial distribution of flow parameters downstream of the rotor. Moreover, this study discusses the causes of modeling errors and presents the application of the model in inlet distortion simulation at the end.

Performance prediction and Bayesian optimization of screw compressors using Gaussian Process Regression

Optimizing the performance of screw compressors is critical for achieving high efficiency and reducing costs in various industrial and engineering applications. Often, the design and optimization processes are time-consuming owing to the underlying iterative complex analyses. In this context, the present research investigates the potential of Gaussian Process Regression (GPR) and Bayesian optimization for the prediction and optimization of the performance of an oil-flooded screw compressor. Specifically, the GPR-based surrogate model is developed to predict the compressor performance characteristics based on its four main geometrical design parameters such as wrap angle, relative length, tip speed of the male rotor and built-in volume ratio. The model is trained using a dataset comprising 19,200 data points relating the input design parameters with the compressor performance, obtained using physics-based multi-chamber thermodynamic models. While four different learning algorithms such as Support Vector Machine (SVM), Artificial Neural Network (ANN), Polynomial regression and GPR are explored, the GPR performed the best resulting in an R2 value of 0.99 for the test dataset after hyperparameter tuning. Further, the model is also experimentally validated on a completely unseen dataset, showing very good predictions with a maximum error of 5%. The resulting surrogate model is then used to optimize the compressor design parameters using Bayesian optimization. The results are compared with optimization using Genetic Algorithm (GA) and physics-based multi-chamber thermodynamic model. It was shown the proposed approach results in similar optimal design parameters but with a significantly less optimization time by a factor of 7. The study highlight the potential of machine learning-based prediction and optimization of screw compressors in engineering applications.

Quantitative analysis of energy loss and performance prediction of regenerative compressor

The intricate internal flow structure within the regenerative compressor leads to substantial energy loss, presenting a long-standing challenge in accurately predicting its performance under various operating conditions. This paper establishes a performance prediction model that considers the impact of blade side exit angle and generalizes energy loss coefficient to accurately predict the performance of regenerative compressor under different operating conditions. Through numerical simulation analysis, the internal flow structure of the regenerative compressor is visualized, and the distribution and variation patterns of internal energy loss are studied. The magnitudes of loss rates for various components are quantitatively analyzed, revealing the underlying energy loss mechanism. A link between loss coefficients and the operating condition parameters of the regenerative compressor is established. Finally, the prediction model is applied to various similar compressor models for performance prediction, and compared with numerical simulation results. The results indicate that the prediction model can effectively predict the performance of various regenerative compressor models under different operating conditions, with maximum errors in total pressure and efficiency of 3.8% and 3.4%, and at rated conditions of 1.13% and 1.78%, respectively.

AKGNN-PC: An assembly knowledge graph neural network model with predictive value calibration module for refrigeration compressor performance prediction with assembly error propagation and data imbalance scenarios

Performance prediction plays a key role in product quality control and quality optimization. However, due to the error transmission in the refrigeration compressor assembly process and the imbalance of data collected during the refrigeration compressor assembly process, existing performance prediction methods cannot accurately extract key features of the assembly process resulting in low prediction accuracy. In this paper, a knowledge graph neural network (AKGNN) model with predictive value calibration is proposed for refrigeration compressor performance prediction in two stages. First, a sequence graph attention feature extraction module is designed to extract key features of the assembly process data to cope with the assembly error propagation. Second, to deal with the imbalanced distribution of compressor data, a predictive value calibration (PC) module is designed to calibrate the output of the first stage. Experimental results demonstrate the proposed method outperformed all compared cutting-edge methods in four evaluated indicators.

Centrifugal compressor performance prediction and dynamic simulation of natural gas hydrogen blended

It is great significant to explore the changes of compressor performance for hydrogen transportation. We found that a Gray-RBF neural network prediction model improved prediction accuracy. The prediction error of the model is 2.34 % by data verification. It was found that the hydrogen blending ratio was increased from 0 % to 30 %, the pipeline transmission power was reduced by 9.3 %. Simultaneously injecting hydrogen to natural gas causes the performance curve of the compressor to shift downwards. The pressure ratio decreases by 18.9 % and shaft power decreases by 28.6 % when the hydrogen blending ratio increases from 0 % to 30 %. The injection of hydrogen to natural gas causes the surge range of the compressor to increase, so the stable range to narrow. In general, this research may provide reference and guidance for the transportation research of hydrogen blended natural gas and compressor operation about gas industry.

Performance prediction of the centrifugal air compressor for fuel cells considering degradation characteristics based on the hierarchical evolutionary model

The performance of the centrifugal air compressor is closely related to the efficiency and energy density of the fuel cell system, and extensive work has been done in this study to achieve precise performance predictions of the compressor considering degradation characteristics. The durability test of the compressor was carried out for 5000 h, and degradation characteristics with break-in time were discussed in detail. On this basis, the hierarchical evolutionary model was constructed to perform the performance predictions based on the Light Gradient Boosting Machine (LightGBM) and Multi-objective Whale Optimization Algorithm (MOWOA). The results demonstrate that the outlet pressure of the compressor decreases by 9.4 kPa and the power consumption increases by 0.71 kW at rated volume flow rate after the break-in of 5000 h, with attenuation ratios of 4.7 % and 12.2 %, respectively. The final non-dominated solution sets have excellent uniformity, diversity and convergence, proving that the developed MOWOA has high adaptability and strong optimization capability. The hierarchical evolutionary model achieves exact predictions for the outlet temperature, flow rate and pressure of the compressor, with absolute relative errors (AREs) below 1 % in outlet temperature and pressure while 5 % in outlet flow rate. All of these provide an efficient multi-objective optimization algorithm and an auto-calibrated hierarchical evolutionary model for further performance optimization of the air compressor.

Investigation of turbulence models for aerodynamic analysis of a high-pressure-ratio centrifugal compressor

The aerodynamic design and optimization of modern centrifugal compressors are inseparable from computational fluid dynamics (CFD) tools. In a design environment, CFD simulations are expected to predict compressor performance characteristics with reasonable accuracy. Turbulence models and model simplification can play a significant role in the discrepancy between analysis and reality. To investigate the influence of turbulence models on aerodynamic analysis results of centrifugal compressors, the National Aeronautics and Space Administration's high efficiency centrifugal compressor stage is selected as the test vehicle. Six eddy-viscosity models, namely, Spalart–Allmaras (SA), standard k–epsilon (k−ε), renormalization group k–epsilon (RNG k−ε), Wilcox k–omega (k−ω), shear stress transport (SST), and SST-reattachment modification (SST-RM) are considered. In all numerical analyses, measured surface roughness is included with the aim of improving numerical solution accuracy. Although some discrepancies exist, the compressor performance characteristics predicted by the k−ω model are in satisfactory agreement with the test data over the whole speedline. The k−ε model overpredicts the isentropic efficiency than the other models, while the SST-RM model underestimates the total pressure ratio in the near-stall region. The RNG k−ε model matches high flow characteristics but overestimates the choked mass flow rate. The SST model obtains a similar operating range with the test data but underestimates the compressor's isentropic efficiency and total pressure ratio, while the SA model totally overestimates the compressor characteristics due to the wall function. Furthermore, variations caused by turbulence models are discussed together with the flow field analysis. Finally, to verify the impact of the impeller backplate bleed cavity, an additional simulation is conducted that specifically accounts for this feature.

Development of a surrogate model for uncertainty quantification of compressor performance due to manufacturing tolerance

In gas turbines and jet engines, stagger angle and tip gap variations between adjacent blades lead to the deterioration of performance. To evaluate the effect of manufacturing tolerance on performance, a CFD-based uncertainty quantification analysis is performed in this work. However, evaluating dozens of thousands of rotor assembly through CFD simulations would be computationally prohibitive. A surrogate model is thus developed to predict compressor performance given an ordered set of manufactured blades. The model is used to predict the influence of tip gap and stagger angle variations on maximum isentropic efficiency. The results confirm that the best arrangement is obtained by minimizing the stagger angle variation between adjacent blades, and by maximizing the tip gap variation. Another finding is that the best arrangement yields the lowest variability, the range of maximum efficiency being 4 times sharper (resp. 2 times) than worst arrangement for stagger angle variations (resp. tip gap variations). Not measuring manufacturing tolerance, or not specifying any strategy for the blade arrangement, lead to variability as large as the worst arrangement.

Assessment of advanced RANS turbulence models for prediction of complex flows in compressors

Reynolds-Averaged Navier-Stokes (RANS) Computational Fluid Dynamics (CFD) has been widely used in compressor design and analysis. However, reasonable prediction of compressor flow and its impact on compressor performance remains challenging. In this study, Menter’s Shear Stress Transport (SST) model and its variants, as well as the ω-based Reynolds stress model (Stress-BSL) are assessed. For a single rotor (Rotor 67), under the peak efficiency operating condition, all studied turbulence models predict its performance with reasonable accuracy; under the off-design conditions, SST with Helicity correction (SST-Helicity) shows superiority in predicting the effect of flow on the spanwise distribution of aerodynamic parameters. For Darmstadt’s 1.5-stage transonic axial compressor, SST-Helicity outperforms SST, SST with the Quadratic Constitutive Relation (SST-QCR) and Stress-BSL in predicting the performance as well as the spanwise distribution of aerodynamic parameters. At the design rotating speed, the stall margin given by SST-Helicity (20.90%) is the closest to the experimental measurement (24.81%), which is more than twice that by SST (8.71%) and 1.5 times that by SST-QCR (14.14%). This paper demonstrates that SST-Helicity model, together with a good quality and sufficiently refined grid, can capture the compressor flow features with reasonable accuracy, which results in a credible prediction of compressor performance and stage matching.

Compressor Performance Prediction Based on the Interpolation Method and Support Vector Machine

Compressors are important components in various power systems in the field of energy and power. In practical applications, compressors often operate under non-design conditions. Therefore, accurate calculation on performance under various operating conditions is of great significance for the development and application of certain power systems equipped with compressors. To calculate and predict the performance of a compressor under all operating conditions through limited data, the interpolation method was combined with a support vector machine (SVM). Based on the known data points of compressor design conditions, the interpolation method was adopted to obtain training samples of the SVM. In the calculation process, preliminary screening was conducted on the kernel functions of the SVM. Two interpolation methods, including linear interpolation and cubic spline interpolation, were used to obtain sample data. In the subsequent training process of the SVM, the genetic algorithm (GA) was used to optimize its parameters. After training, the available data were compared with the predicted data of the SVM. The results show that the SVM uses the Gaussian kernel function to achieve the highest prediction accuracy. The prediction accuracy of the SVM trained with the data obtained from linear interpolation was higher than that of cubic spline interpolation. Compared with the back propagation neural network optimized by the genetic algorithm (GA-BPNN), the genetic algorithm optimization of extreme learning machine neural network (GA-ELMNN), and the genetic algorithm optimization of generalized regression neural network (GA-GRNN), the support vector machine optimized by the genetic algorithm (GA-SVM) has a better generalization, and GA-SVM is more accurate in predicting boundary data than the GA-BPNN. In addition, reducing the number of original data points still enables the GA-SVM to maintain a high level of predictive accuracy.

Determination of Effective Flow and Force Areas for Reed Type Compressor Valve Systems: Part 1—Model Derivation Using CFD Analysis Results

The dynamic behavior of reed valves during suction and discharge is very important in predicting compressor performance and valve fatigue fracture. For an accurate dynamic analysis of reed valves, mass flow rate and effective force area should be accurately modeled. In this study, mass flow rate and effective force area models were developed based on CFD analysis results using the well-known commercial software ANSYS Fluent. CFD analyses were carried out for various values of port radius, valve radius, valve lift, and pressure difference. The mass flow rate and pressure force on valve were obtained from the analysis. Then, effective flow and force area models were proposed and the model coefficients were determined using the CFD results. The average prediction error of the effective flow area and the effective force area were 1.90% and 2.9%, respectively. It was shown that the effective flow area is dependent only on the port radii and valve lift; it is not dependent on the valve radii. The proposed effective force area model can accurately describe the decrease and increase in pressure force with valve lift.

Quantitative investigation of the turbulence model effect on high-pressure-ratio centrifugal compressor performance prediction

The accuracy of centrifugal compressors performance and flow field prediction by Computational Fluid Dynamics (CFD) is a serious problem facing in the design and analysis process due to its complex flow feature inside. Different turbulence models choosing in the Reynolds-Averaged Navier-Stokes Equations (RANS) leads to different simulation results. In this paper, a high-pressure-ratio centrifugal compressor was used as the simulation object to discuss the effect of common turbulence models, including a one-equation model (Spalart-Allmaras model), a two-equation linear model (Shear Stress Transport model) and a two-equation nonlinear model (Explicit Algebraic RSM model), on performance prediction under design speed. By introducing the calculation method of local entropy generation rate, an improved method of dividing different types of loss in centrifugal compressors was proposed. After comparing with the experiment results, the effect of turbulence models on loss prediction inside the compressor was quantitative investigated. Further, the flow mechanism underlying was discussed. The results show that the secondary flow loss have the largest impact on the components' prediction performance, also, the reasons for different magnitudes of secondary flow loss inside the impeller and the diffuser are different. These comparisons hope to help designers choose an appropriate turbulence model in further compressor design process and may benefit future turbulence modelling development.

Modification and Validation of 1D Loss Models for the Off-Design Performance Prediction of Centrifugal Compressors with Splitter Blades

One-dimensional (1D) aerodynamic performance predictions are very often conducted by researchers and designers during the preliminary design of centrifugal compressors. This paper focuses on a 1D prediction method for centrifugal compressors with splitter blades, which is rarely seen in the open literature. One-dimensional prediction of aerodynamic overall performance is made for centrifugal compressors with different technical design specifications. However, the aerodynamic overall prediction accuracy relies on the accuracy of the 1D-loss-models used. Therefore, an optimum combination of loss models is proposed by summarizing a variety of loss models presented in the public literature. In addition, an optimization method is utilized to optimize some coefficients involved in loss models in order to improve the generality of the combined model. The modified models obtained in this study are proved to have good predictive accuracy.

Performance prediction of centrifugal compressor based on a new volute loss model and corrected theoretical work

AbstractPerformance prediction plays an important role in the predesign stage of centrifugal compressors. A set of loss models is usually selected and grouped to fulfill the above prediction. To improve the defects of traditional prediction, we modified the outlet velocity and slip coefficient of the impeller by means of the energy average method to obtain theoretical work. A new volute loss model is proposed to replace the traditional semiempirical loss model by adopting a suitable loss model for each flow passage component. A proper performance prediction method is established for the centrifugal compressor studied in this paper. The overall performance of a single‐stage centrifugal compressor is predicted and compared with experimental and numerical simulation results. It is found that the prediction method proposed can better predict the overall performance of the centrifugal compressor. And the volute loss is the third largest loss after impeller loss.