See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352341681

Stator re-stagger optimization in multistage axial compressor

Article in Propulsion and Power Research · June 2021

DOI: 10.1016/j.jppr.2021.03.002

CITATIONS READS

5 128

5 authors, including:

Jinguang Yang Michele Ferlauto

Dalian University of Technology Politecnico di Torino
46 PUBLICATIONS 317 CITATIONS 96 PUBLICATIONS 534 CITATIONS

SEE PROFILE SEE PROFILE

Yan Liu
Dalian University of Technology
56 PUBLICATIONS 461 CITATIONS

SEE PROFILE

All content following this page was uploaded by Michele Ferlauto on 06 May 2023.

The user has requested enhancement of the downloaded file.

Propulsion and Power Research 2021;10(2):107e117

http://ppr.buaa.edu.cn/

Propulsion and Power Research

w w w. s c i e n c e d i r e c t . c om

ORIGINAL ARTICLE

Stator re-stagger optimization in multistage axial

compressor
Jinguang Yanga, Min Zhanga, Cheng Penga, Michele Ferlautob,
Yan Liua,*

a
School of Energy and Power, Dalian University of Technology, Dalian, 116024, China
b
Department of Mechanical and Aerospace Engineering, Politecnico di Torino, Turin, 10129, Italy

Received 24 August 2020; accepted 16 March 2021

Available online 11 June 2021

KEYWORDS Abstract As a widely applied technique in multistage axial compressors, variable stator
vanes (VSV) can flexibly rematch the blade rows to fulfil a variety of aerodynamic performance
Multistage axial
compressor; requirements, such as high efficiency and wide surge margin. The purpose of this paper is to
Mean-line calculation; develop an optimization method to quickly determine VSV settings during the preliminary
Loss model calibration; design phase. A mean-line method with a model calibration procedure is adopted to evaluate
Variable stator vane; compressor performance, and the NSGA-II algorithm is employed for automatic optimization.
Re-stagger optimization The developed optimization system is then employed to determined re-stagger arrangement in a
multistage compressor. A single-speed optimization with performance constraints of specific
operating point and a multi-speed optimization with different control laws are conducted. Re-
sults are compared with available experimental re-stagger scheme, which verifies the effective-
ness of the re-stagger optimization method. Moreover, method is proposed to determine
operating parameters of a working point with a user-defined pressure ratio or mass flow rate
after variable geometry.
ª 2021 Beihang University. Publishing services by Elsevier B.V. on behalf of KeAi Communications
Co. Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

*Corresponding author.
E-mail address: [email protected] (Yan Liu).
Peer review under responsibility of Beihang University.

Production and Hosting by Elsevier on behalf of KeAi

https://doi.org/10.1016/j.jppr.2021.03.002
2212-540X/ª 2021 Beihang University. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article
under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
108 Jinguang Yang et al.

Nomenclature ori original

opt optimized
f objective function ref reference value
g constraint condition
i incidence angle (unit:
)
K proportionality coefficient
Acronyms
CDA controlled diffusion airfoil
m_ mass flow rate (unit: kg/s)
CP constrained point
n revolution speed (unit: rpm)
CFD computational fluid dynamics
Nrow blade row number
DCA double circular arc
DF diffusion factor
Greek letters EMC empirical model calibration
g stagger angle (unit:
) GA genetic algorithm
h efficiency MSOpt multi-speed re-stagger optimization
ha adiabatic efficiency SM surge margin
pt total pressure ratio SSOpt single-speed re-stagger optimization
c design variable SV stator vane
u total pressure loss VSV variable stator vane

Subscripts
c constrained operation condition
s surge condition For multistage compressors, variable inlet guide vane
(VIGV) or VSV can improve stage matching, as illustrated
in Figure 1. Stagger angles of IGV or each stator vane (SV)
can be adjusted, and hence inlet flow angles or incidence
1. Introduction angles (i) at rotor inlet are changed according to operating
conditions. The re-stagger magnitude is often decreased
With the progress of gas turbines and aero engines, re- from upstream IGV to downstream SVs [8]. This finally
quirements such as high-efficiency and high-loading multi- compromises stage mismatching to ensure a wide high-
stage axial compressors are urgent. Meanwhile, the performance operating range and a large surge margin.
matching problem is also significant since each blade row Therefore, it is necessary to develop a reasonable criterion
has its optimum performance at a specific inlet flow angle to determine re-stagger scheme.
condition. This condition is termed as design point or match In order to determine VSV scheme without using
point, which is determined by the design mass flow rate and experimental or CFD methods that are highly costly and
design revolution (m_ d and nd). However, a compressor al- time-consuming, many researchers have conducted optimi-
ways encounters off-design conditions, where mass flow (m) _ zation studies using reduced-order models. Roh and Daley
and speed (n) deviate from their design values. As intro- [8] optimized re-stagger settings of IGV and SVs of an axial
duced by Aungier [1], at nd condition, unstable operations compressor using the particle swarm optimization (PSO)
referred as surge and choke are commonly occurred when m_ algorithm, where the objective was composed of surge mass
is lower and larger than m_ d respectively. Meanwhile, front flow, polytrophic efficiency and pressure ratio based on a
stages often experience stall and then surge, and rear stages weighting scheme and two speed conditions were consid-
are choked when n is lower than nd. An opposite process ered. Their results indicated that the developed PSO method
will happen when n is higher than nd [2]. These can induce is an efficient and rapid tool for on-line re-stagger optimi-
high aerodynamic loss and even structural damage due to zation. Regarding a seven-stage aircraft compressor, Sun
flow oscillation. and Elder [2] carried out a direct research for optimal re-
In order to avoid stall and subsequent surge operations,
various control techniques [3,4] including inter-stage
bleeding, multi-spool arrangement, casing treatment and
variable stator vanes (VSV) are utilized in multistage axial
compressors. Among these performance improvement
methods, VSV approach can reduce stage mismatching
level and hence satisfy different performance requirements.
This technique is also beneficial during engine start-up and
acceleration processes [5] or when encountering environ-
mental changes caused by season and flight altitude [6].
It can also be considered in initial design phase of com-
pressors [7], where design condition is generally different
from operating condition. Figure 1 Schematic diagram of re-stagger scheme.
Stator re-stagger optimization in multistage axial compressor 109

stagger setting based on a sequential weight increasing conducted. A six-stage axial part of a combined compressor
factor technique (SWIFT). Like the study of Roh and Daley is chosen for re-stagger optimization. The obtained optimized
[8], they performed several optimization cases under settings are compared with available experimental ones, and
different single revolution speed, and results also proved verify the correctness of the developed method.
that VSV is beneficial to improve performance and realize
stage re-matching at part-speed operations. Later, using the
same optimization scheme as in Ref. [2], White et al. [9] 2. Re-stagger optimization method
further demonstrated possibility of VIGV or VSV to
improve the compressor performance. 2.1. Optimization procedure
Re-stagger optimization works for VIGV or VSV of
Gallar et al. [10] and Reitenbach et al. [11] were conducted As already mentioned, the purpose of the re-stagger opti-
from a whole engine cycle point of view. Gallar et al. [10] mization is to adjust stagger settings of IGV or stator for
considered coupled efficiency of compressor and turbine, multistage compressors. The optimization procedure developed
while Reitenbach et al. [11] integrated engine specific fuel in this paper is shown in Figure 2. Key elements include design
consumption analysis. Both two studies realized appropriate variables (c), a performance evaluation solver and an optimi-
re-stagger schedules that had better surge margin for their zation algorithm. The performance indexes are also used in the
compressor. Similarly, Wang et al. [12] numerically evalu- definition of objective function, i.e., f(c). Since c and f(c) are
ated impacts of VSV on the efficiency and fuel consumption case-dependent considering application requirements and
of a three-shaft gas turbine. In additions, Li et al. [13] working conditions of a compressor, they will be introduced in
conducted VSV optimization based on artificial neural net- detail in Section 4. This section mainly discusses the perfor-
works and genetic algorithm. They studied two operating mance solver and the optimization algorithm.
conditions at design-speed and carried out a sensitivity
analysis using an established ANN to analyze influence of 2.2. One-dimensional mean-line method based on
geometric parameters on compressor performance. empirical model calibration
Above-mentioned researches have proved the effective-
ness of re-stagger setting to improve the compressor perfor- Generally, essential performance parameters for com-
mance. Most of them define multi-objectives in their pressors contain total pressure ratio (pt), adiabatic efficiency
optimization procedures, however, only a set of operating (ha) and surge margin (SM). In order to evaluate these
conditions on one speed line are specified during the opti- variables, a 1D mean-line method is adopted in current
mization process. In addition, one-dimensional (1D) mean- study. Calculation stations are located on the mean line of
line method combined with proprietary loss models is usually blade rows. With the specified inlet conditions and geo-
employed for performance calculation. Therefore, prediction metric parameters, compressible fluid relations are
accuracy of 1D method is significant for feasibility of opti- employed to solve aerodynamic parameters at the inlet and
mized VIGV or VSV. This paper does not intend to make outlet of each blade row. Different rotational speeds and
comments on accuracy of 1D tools in above studies, but various operating conditions at a specific speed are
attempt to increase predictive accuracy via a model calibra- considered to get characteristic curves of a compressor.
tion technique. Based on this, a re-stagger optimization sys- In order to involve real flow effects, various empirical
tem is developed. Moreover, an optimization considering loss models are integrated in the 1D mean-line analysis
both single-speed and multiple-speed conditions is code. Empirical models associated with minimum loss

Figure 2 Flow chart of stator re-stagger optimization.

110 Jinguang Yang et al.

incidence [14], reference deviation [15], off-design devia- general, but the finally calibrated loss models are case
tion [14], profile loss [16], off-design loss [17,18] and end- dependent.
wall blockage [19] are adopted for the compressor studied in
this paper. In addition, since one of main applications of re- 2.3. Optimization algorithm
stagger setting is to increase SM of multistage compressors,
models for estimating stall and surge are also integrated. In The re-stagger optimization in the current paper is a
the present mean-line code, surge prediction models, such as multi-objective problem, where the objectives will also be
McKenzie [9], Aungier [1], Koch [20] et al., are included. discussed thoroughly in Section 4. Therefore, the NSGA-II
They can help to determine a relatively reasonable surge (non-dominated sorting genetic algorithm II) algorithm
boundary line, and hence aid to get an appropriate influence [24] is adopted. This algorithm calculates fitness function
tendency of re-stagger on surge boundary. Mathematical according to magnitude of objective functions and con-
formulations of these empirical loss models are not pre- straints. Based on fitness function of parent population,
sented here for the sake of brevity, please refer to the work selection, crossover, mutation, and recombination opera-
of Peng [21] for detail. tors are executed to generate offspring populations. After
Empirical loss models can efficiently assess irreversible multiple generations of screening, solutions termed as
loss in compressor cascades. However, most of them [22] Pareto-optimal front are finally obtained. Due to superi-
are established based on early experimental results of some orities in simplicity, good robustness and global conver-
standard cascades, such as NACA-65 series, C series and gence, the NSGA-II algorithm has been widely applied in
DCA series. Meanwhile, some models do not take 3D flow engineering and academic fields.
effects into account. With development of aerodynamic In terms of multiple constraints and objective functions,
design techniques, lots of modern blades, such as geomet- the non-domination principle is employed here to determine
rically parameterized blades and controlled diffusion air- the rank of individuals in parent population. In addition, the
foils, have been widely used in high-performance crowded-comparison operator is introduced to ensure di-
compressors. Therefore, feasibility and accuracy of existing versity of population members, and the elite-preserving
models needs to be further verified for modern blades operator is applied to accelerate convergence of the opti-
with different loading levels and under various working mization algorithm.
conditions.
Due to the above mentioned issues, an empirical model
calibration (EMC) is developed by Peng [21] to improve 3. Compressor studied and method validation
accuracy of the 1D mean-line solver. The process is pre-
sented in the right-hand side of Figure 2. It is also based on 3.1. The compressor studied
an optimization-based scheme, whose design variables are
coefficients of empirical models. As a general empirical A P&WA combined compressor [25,26] with VIGV and
model system, there are 23 coefficients to be optimized. The VSV is selected as the test case. This compressor is designed
objective is to minimize the discrepancy between the per- for a 10 MW class gas turbine. It consists of a six-stage axial
formance value predicted by the 1D mean-line solver and compressor and a single-stage centrifugal compressor on the
that obtained via experiments or 3D CFD predictions. same spool, as shown by the flow-path in Figure 3. In this
Models associated with minimum loss incidence, deviation paper, the re-stagger optimization is applied for the six-stage
and total pressure loss are mainly considered in the cali- axial compressor part. Table 1 lists some key performance
bration process. Meanwhile, the calibration is executed parameters at the design point.
using performance data on the whole characteristic curves of During experiments in Refs. [25,26], the inlet guide vane
a compressor. This means that loss models are modified to and all stators of the P&WA axial compressor are equipped
accurately evaluate performance for as more working con- with mechanical devices to reset stagger angles. Varying
ditions as possible. During the calibration optimization, the ranges of re-stagger angles for the IGV and S1eS6 are
Hausdorff distance is applied to calculate the objective presented in Table 2 [25]. Meanwhile, Figure 4 shows the
function, and the Nelder-Mead [23] algorithm is adopted to experimental re-stagger scheme of the IGV and S1eS2 as a
optimize the model coefficients. Feasibility of this 1D mean- function of revolution. Positive angle value indicates that
line solver based on EMC will be validated in Section 3.2. vane is turned to close, while negative value is just the
It is admitted that calibrated coefficients are case opposite. For this combined compressor, the purpose of re-
dependent, and this is a natural result. The standard loss and stagger is not only to provide an adequate off-design effi-
deviation models are aimed to be applied in a general pur- ciency and surge margin characteristic, but also to achieve
pose. However their accuracy may be low when applied to a flow matching of the axial and centrifugal compressor at
particular application. Since the calibration acts as an part speed. Meanwhile, it is worth mentioning that to satisfy
improved process of the standard model, it is natural that the the latter requirement, an inter-stage bleeding technique is
calibrated models have a narrower application range adopted in experiments, which is not considered since only
compared to the standard ones. So the calibration method is axial part is modeled.
Stator re-stagger optimization in multistage axial compressor 111

the experimental data [25]. It should be mentioned that after

a comparison of the three surge estimation models intro-
duced in Section 2.2, the McKenzie model [9] is finally
chosen to determine the surge boundary line of the P&WA
compressor.
Figure 5 shows the total pressure ratio (pt) and adiabatic
efficiency (ha) predicted by the 1D solver using the stan-
dard and calibrated models respectively. From the com-
parison results of pt in Figure 5(a), values predicted by
standard models (1D) are higher at all operating conditions,
and the 1D calibrated data (1D_C) are nearly consistent
Figure 3 Flow path of the combined compressor. with experimental values (EXP). For ha in Figure 5(b),
prediction results using calibrated models are in better
agreement with the measured data than those using stan-
3.2. Validation of the 1D mean-line solver dard models, except those at speeds of 95%nd and 50%nd.
Furthermore, predicted mass flow rate at peak efficiency
In order to validate the 1D method developed in Section points after calibration are consistent with experimental
2.2, performance of the P&WA 6-stage compressor is pre- value. These results prove that the 1D solver with cali-
dicted. Coefficients of empirical models are calibrated using brated models can accurately predict the compressor per-
formance. Therefore, the solver and calibrated models can
be employed to evaluate objective functions during the re-
Table 1 Performance parameters of the P&WA 6-stage. stagger optimizations.
Parameters Symbols Values

Design revolution speed nd 15,000 rpm

Design mass flow rate m_ d 31.572 kg/s
Total pressure ratio pt 6.67
Adiabatic efficiency ha 89.0%

Table 2 Adjustable ranges of re-stagger settings.

Vane g/(
)

IGV 10, þ40

S1 5, þ15
S2 5, þ10
S3 5, þ5
S4 5, þ5
S5 5, þ5
S6 10, þ10

Figure 4 Re-stagger scheme of IGV and S1eS2 in experiments. Figure 5 Overall performance of the P&WA compressor.
112 Jinguang Yang et al.

4. Re-stagger optimization and discussion optimized. The mathematical expression of the SSOpt
problem is formulated as below:
For the re-stagger optimization of the P&WA axial
compressor, single-speed and multi-speed conditions are


considered, and both involve multiple working points cal- minimize f1 ðcÞZabs pt;opt;c  pt;ori;c
culations. It should be mentioned that the centrifugal part of


the P&WA compressor is not modeled, but flow angle minimize f2 ðcÞZ 1:0  ha;opt;c þ 1:0  ha;opt;s
8

matching issue is considered in the optimization. >
> g1 ðcÞZ pt;opt;s  pt;ori;s > 0:0 ð1Þ
<

4.1. Single-speed re-stagger optimization (SSOpt) s:t: g2 ðcÞZ DF;opt;c  0:60 < 0:0
>
>

:
g3 ðcÞZ DF;opt;s  0:60 < 0:0
4.1.1. Design parameters and objective functions
The single-speed re-stagger optimization (SSOpt) means where subscripts ori and opt are the original and optimized
that optimization is conducted under a specific revolution. values, c and s stand for the constrained point and near-
The objective is to maximize both the efficiency and surge surge point respectively. In addition, DF demonstrates blade
margin. As noted in Ref. [25] and implied by Figure 5, the diffusion factor. It is introduced to avoid results with un-
P&WA compressor has poor performance at speed of 70% reasonable aerodynamic conditions.
nd. Therefore, this speed condition is selected for the SSOpt. Equality constraints in Eq. (1) can decrease the stability
The operating point which has the maximum efficiency of the optimization and lead to local optimal solution. Due
under this speed is regarded as the constrained point (CP). to this, the condition that pt is unchanged is converted to an
This means that the total pressure ratio (pt) of the CP should objective function, as indicated by f1(c) in Eq. (1). More-
be unchanged during the optimization, while the efficiency over, the requirement of not altering pt may reduce the
(ha) of the CP is maximized. feasible solution space during initial random sampling. In
Since pt of the CP is not altered after optimization, the order to alleviate the issue, the revolution speed is added as
essence of the SSOpt is to rotate the characteristic curves an additional design variable. Since performance parameters
around the CP. Figure 6 shows the feasible and unfeasible at constrained and surge points are equally important, f2(c)
directions of this rotation. Compared to the original char- is a weighted sum of the efficiency at these two conditions.
acteristic curve, i.e., Cori, surge margin will be broadened if The subtractive expression (1eha) is to make f2(c) be
the Cori curve revolves to the Cfea curve, while it will be minimized. The design variables for the SSOpt are stagger
narrowed if the Cori curve revolves to the Cunf curve. Since angles of the IGV and SVs, also the revolution.
variation of surge mass flow is small, optimization direction
is significantly affected by the estimation accuracy of surge 4.1.2. SSOpt results
boundary models. So a multi-point optimization strategy is Using the SSOpt approach described above, two re-
integrated into SSOpt, and SM is avoided in objective stagger arrangements are conducted. The first one optimizes
function, instead the efficiency at a near-stall point (SP) is stagger angles of IGV, S1 and S2 (IGV-S2), while the sec-
ond optimizes those of IGV and S1eS5 (IGV-S5). For the
second scheme, the stagger angle of S6 is fixed in order to
guarantee the inlet flow angle of the subsequent centrifugal
compressor. The optimization process is carried out for 70
generations, with 600 individuals for each generation. Re-
sults of both two SSOpt arrangements are shown in
Figure 7, where Pareto fronts can be clearly identified. Since
pt is regarded as the primal criterion for selecting the
feasible solution, so points with minimum values of f1(c)
are chosen as the optimal solution, as indicated by
SSOpt_IGV-S2 and SSOpt_IGV-S5 in Figure 7.
Detailed re-stagger information of the SSOpt_IGV-S2
and SSOpt_IGV-S5 are listed in Table 3. Values of g are
gradually decreased from the upstream IGV to downstream
S5 at the nominal 70%nd. It can also be seen from Table 3
that the SSOpt_IGV-S5 arrangement is more appropriate for
multi-stage compressor. This is due to a fact that the front
stages are easily surged while the rear stages are to be
choked at part speed. The IGV, S1eS2 are closed and
Figure 6 Illustration of optimization direction on one speed line. S3eS5 are opened in the SSOpt_IGV-S5 scheme. This is
Stator re-stagger optimization in multistage axial compressor 113

Figure 7 Objectives of the SSOpt.

Figure 8 Comparison of characteristic curves for the P&WA 6-stage

Table 3 Re-stagger schemes of the SSOpt.
axial compressor at 70%nd.
SSopt_IGV-S2 SSOpt_IGV-S5
nopt/rpm 11062.8 10951.2  
gIGV/(
) 40.000 39.999 pt;s mori
gS1/(
)
SM Z  1  100% ð2Þ
14.998 14.998 pt;ori ms
gS2/(
) 10.000 10.000
gS3/(
) e 0.361
It is found that compared to the non-re-stagger
gS4/(
) e 5.000
gS5/(
) e 4.933 scheme, SSOpt_IGV-S2 and SSOpt_IGV-S5 improve the
SM by 19.146% and 19.155% respectively, and increase
with maximum efficiency by 6.015% and 6.805%
beneficial to avoid the surge condition in front stages and to respectively.
alleviate choke condition in rear stages. Figure 9 presents incidences (i) and off-design loss (u)
Figure 8 plots the characteristic curves of the P&WA of blade rows from R1 to S6 at the CP. Both i and u are
axial compressor at the nominal speed of 70%nd. Non-re- calculated using empirical loss models. It can be observed
stagger indicates performance data without re-stagger that the re-stagger optimization rematches the blade
scheme, while re-stagger_EXP represents those with rows to keep the incidences away from the stall ones, this
experimental re-stagger scheme [25]. All data are predicted is the reason why the surge margin and efficiency are
using the 1D mean-line solver. It can be clearly seen that increased at part speed in Figure 8. Although Figures 8
relative to the non-re-stagger curve, the experimental and and 9 prove that the IGV-S5 optimization results are
both two optimized performance curves are revolved in slightly better than those of IGV-S2, the structural cost
clockwise around the constrained point. This is consistent and influence of vane end gaps can be significantly
with the analysis described in Figure 6. In order to evaluate increased in this scheme. Therefore, the IGV-S2 results
the operating range, the SM is defined as: may be a more reasonable solution, which is consistent
114 Jinguang Yang et al.
0 1
X m_ s  m_ ref
minimize f1 ðcÞZ @ A; iZ1; 2; 3; 4
m_ s
X
minimize f2 ðcÞZ ð1:0  ha;i;max Þ; iZ1; 2; 3; 4
ð3Þ
8 DgIGV;i  DgIGV;100%
>
< g1 ðcÞZ  KIGV Z0:0
ni  nd
s:t:
>
:
g2 ðcÞZDgSV;i  KSV DgIGV;i Z0:0

where the subscript i represents each speed condition.

Similar to the definition of f2(c) in Eq. (1), both f1(c) and
f2(c) in Eq. (3) are weighted sum objective functions of the
performance at the four speed conditions. For the function
f1(c), m_ ref is a reference mass flow, and m_ s is predicted using
the surge model. The subtractive formula in f2(c) is also to
solve a minimization problem.
The two constraint functions in Eq. (3) are defined
considering VSV controlling laws. For g1(c), a constant
parameter is introduced to specify a linear relationship be-
tween the re-stagger angle and the revolution speed. This is
due to a situation that g is proportionally altered with n. The
parameter KIGV is a linear factor of experimental re-stagger
schedules for IGV only. In addition, under multi-speed cir-
cumstances, a common rod is installed to simultaneously
control the re-stagger setting of IGV and SVs to reduce
complexity, weight, and cost of the structure [10]. This
single-rod arrangement has been utilized in the LM2500
compressor [27], GE-E3 compressor [28] and KHI 14-stage
compressor [29]. Therefore, the second constraint function,
g2(c), imposes a linear constraint for the variations in g
Figure 9 Incidences of blade rows at the operating point. between IGV and stators, as implied by the factor KSV. This
setting is not same to the strategy in the SSOpt, where re-
stagger angles of IGV and SVs are independently adjusted
with the experimental scheme. All results verify that the by separate mechanical structures. To summarize this sec-
re-stagger optimization can effectively identify re-stagger tion, it is easy to verify that there are 3 design variables for
direction of variable vanes in preliminary design phase the MSOpt, namely DgIGV,100%, KIGV and KSV.
of compressors.
4.2.2. MSOpt results
4.2. Multi-speed re-stagger optimization (MSOpt) For the MSOpt, two optimization arrangements are also
considered. They are based on the re-stagger of IGV-S2 and
4.2.1. Design parameters and objective functions IGV-S5, as termed as MSOpt_IGV-S2 and MSOpt_IGV-S5
The last section just considers one single-speed during respectively. During the MSOpt process, 70 generations and
the re-stagger optimization. In order to further improve the 200 populations are set respectively. Results show that for
ability of VIGV&VSV for performance enhancement, the MSOpt_IGV-S2 case, the convergence trend of f1(c) is
multi-speed re-stagger optimization (MSOpt) is conducted the same as that of f2(c), so the Pareto-optimal front is
for the PW&A 6-stage axial compressor. Four speed con- converged to a point. For the MSOpt_IGV-S5 case, the
ditions are considered in MSOpt, and they are 70%nd, 80% Pareto front is shown in Figure 10. An optimal point is
nd, 90%nd and 95%nd respectively. Unlike the requirements selected to realize the maximum efficiency, as shown by the
in SSOpt, pressure ratio in MSOpt is not constrained, so the red circle in the Figure 10.
multi-point strategy in Section 4.1 is not adopted, and only The optimized re-stagger angles at various speed con-
the peak efficiency (ha,max) of each speed condition is ditions are shown in Figure 11, while comparisons in
evaluated. Meanwhile, surge mass flow rate (m_ s ) is to be performance are presented in Figure 12. Similar to the
minimized. The mathematical expressions of optimization SSOpt, changing trend in g from IGV to S2 designed by
can be written as: both two MSOpt cases is the same as that arranged in
Stator re-stagger optimization in multistage axial compressor 115

Figure 10 Objectives of the MSOpt_IGV-S5.

Figure 12 Characteristic curves of the P&WA 6-stage compressor

after the MSOpt.

extent. The maximum difference in peak efficiency be-

tween the two optimized schemes is 1.071% at 70%nd. It
implies that re-stagger the rear stages has a relatively small
effect on performance improvement. This phenomenon is
also observed in many compressors [27e29]. All results
illustrate the effectiveness of the developed re-stagger
optimization schedule.
It is noticed that results of the MSOpt at 70%nd is
basically the same as that of the SSOpt, and the optimal
setting is achieved at bounds set by Table 2. This indicates
Figure 11 Re-stagger schedules obtained by the MSOpt. that the performance at 70%nd dominates the multi-speed
optimization, also serves as a validation of each other. The
adjustable ranges are set by real structures, which also limit
experiments. The re-stagger tendency of the the achieved performance. Also, the obtained re-stagger
MSOpt_IGV-S5 also reflects the part-speed operating setting is different with that of the experiment (shown in
features of multi-stage compressors, i.e., alleviating stall in Figure 4). This is may due to that the experimental setting is
front stages and choking in rear stages. In addition, not aimed at aerodynamic performance only, or is the result
Figure 12 demonstrates that both MSOpt schedules have of a comprise between VSV, bleeding and/or matching with
smaller surge mass flow rate, larger efficiency, and wider the centrifugal compressor stage. However, the predicted
surge margin than the experimental re-stagger settings. effects, as shown in Figures 8 and 12, verify the correctness
Compared with the MSOpt_IGV-S2 scheme, the of both the experiment and optimization results in the pre-
MSOpt_IGV-S5 improves the performance only by a small sent study.
116 Jinguang Yang et al.

4.3. Method of locating a working point for a specific In order to get the working point which can realize the
pressure ratio or mass flow rate maximum efficiency for the specific pt, an optimization
using the genetic algorithm is conducted. The revolution
It can be seen from Figures 8 and 12 that the character- speed is regarded as the design parameter. Results indicate
istic curves are obviously altered after the re-stagger opti- when the efficiency reaches to the maximum value, the
mization. Therefore, it is necessary to determine reasonable revolution speed and the mass flow rate are 86.1%nd and
working points to get required off-design performance for 23.696 kg/s respectively. This is nearly consistent with the
the re-staggered compressor. Due to this, a method to point P4 in Figure 14. Therefore, this method can be utilized
ascertain the working point under a certain pressure ratio is to determine working points for other pressure ratios to get a
proposed. The re-stagger scheme illustrated in Figure 4 is reasonable operating line.
used here, and the pressure ratio is fixed to be 4.18. Cor- The same philosophy can also be applied when a mass
responding characteristic curves are shown in Figure 13. flow rate, instead of total pressure ratio, is specified. Those
There are many operating points that can achieve the procedures may be used in the matching axial compressors
required pt, e.g., P1, P2 and P3 in Figure 13. For these with other components.
points, mass flow rates are increased with enlarging the
revolution speed. Meanwhile, the adiabatic efficiency is first
increased and then decreased with raising the mass flow 5. Conclusions
rate, see Figure 14. This corresponds to the condition
changing from stall to the choke, so the working point is A re-stagger optimization method for variable stator
determined according to the maximum efficiency, as vanes is proposed in this paper to improve off-design
demonstrated by P4. aerodynamic performance of multistage axial compressors.
It integrates a mean-line performance analysis method and
the multi-objective NSGA-II optimization algorithm. Pre-
diction accuracy of the mean-line analysis method is
improved via integrating a model calibration procedure.
Two common applications, namely single-speed and
multiple-speed re-stagger optimizations, are carried out for a
multistage axial compressor with VIGV and VSV. Stagger
angles of rows from IGV to second stage stator (IGV-S2),
and from IGV to five-stage stator (IGV-S5) are optimized.
In the single-speed optimization, mass flow rate and
pressure ratio of the operating point are constrained, and
“multi-point” processing is employed to guide the optimi-
zation direction. Results show that compared with the fixed
geometry, the IGV-S2 scheme can improve the surge margin
and maximum efficiency by 19.146% and 6.015% respec-
tively. In addition, optimized surge margin and efficiency of
Figure 13 Schematic diagram of characteristic curves when VIGV/ IGV-S5 approach are only 0.008% and 0.790% higher than
VSV used. those of IGV-S2 scheme. For the multi-speed optimization,
off-design performance of the compressor with re-staggered
vanes is significantly improved, and performance improve-
ments of the IGV-S5 scheme is also small when compared
with the IGV-S2.
Moreover, the tendency of re-stagger schemes obtained
by both single-speed and multi-speed optimizations is
consistent with those of experimental test. But the optimized
schemes have higher efficiency and wider surge margin. A
final achievement of this paper is that a screening method is
established to identify the operating point of a compressor
with variable geometry under a user-defined pressure ratio
or mass flow rate.
In summary, the above results demonstrate effectiveness
and stability of the re-stagger optimization method. This
method is based on 1D performance solver, so it can be
easily combined with other techniques to determine a more
Figure 14 Relationship between mass flow rate and efficiency with comprehensive and reasonable active control scheme for
a specified total pressure ratio when VIGV/VSV used. compressors in engineering applications.
Stator re-stagger optimization in multistage axial compressor 117

Acknowledgements [14] R.O. Bullock, I.A. Johnsen, Aerodynamic design of axial flow com-
pressors, National Aeronautics and Space Administration Report, No.
NASA-SP-36, 1965.
The authors would like to thank the National Natural [15] A.D.S. Carter, H.P. Hughes, A Theoretical Investigation into the
Science Foundation of China (Grant No.51606026) for Effect of Profile Shape on the Performance of Aerofoils in Cascade,
funding this work. Aeronautical Research Council Reports and Memoranda, No. 2384,
1946.
[16] C.C. Koch, L.H. Smith, Loss sources and magnitudes in axial-flow
References compressors, J. Eng. Power 98 (3) (1976) 411e424.
[17] W.C. Moffatt, W. Jansen, The off-design analysis of axial-flow com-
[1] R.H. Aungier, Axial-Flow Compressors, ASME Press, New York, pressors, J. Eng. Power 89 (4) (1967) 453e462.
2003. [18] H.F. Creveling, R.H. Carmody, Computer Program for Calculating
[2] J.J. Sun, R.L. Elder, Numerical optimization of a stator vane setting in Off-Design Performance of Multistage Axial-Flow Compressors,
multistage axial-flow compressors, Proc. Inst. Mech. Eng. Part A: J. National Aeronautics and Space Administration Report, No. CR-
Power Energy 212 (4) (1998) 247e259. 72427, 1968.
[3] L. Gallar, M. Arias, V. Pachidis, R. Singh, Stochastic axial compressor [19] W.R. Davis, A Computer Program for the Analysis and Design of
variable geometry schedule optimisation, Aero. Sci. Technol. 15 (5) Turbomachinery-Revision, Carleton University, Revised Edition,
(2011) 366e374. Ottawa, 1971.
[4] D.B. Alone, S.S. Kumar, S.M. Thimmaiah, J.R.R. Mudipalli, [20] C.C. Koch, Stalling pressure rise capability of axial flow compressor
A.M. Pradeep, S. Ramamurthy, V.S. Iyengar, Stability management of stages, J. Eng. Power 103 (4) (1981) 645e656.
high speed axial flow compressor stage through axial extensions of [21] C. Peng, One-dimensional Performance Modeling and Variable Ge-
bend skewed casing treatment, Propul. Power Res. 5 (3) (2016) ometry Optimization of Multistage Axial Compressor, Master thesis,
236e249. Dalian University of Technology, Dalian, Liaoning, 2019.
[5] Q.G. Zheng, H.B. Zhang, Z.F. Ye, L.Z. Miao, Acceleration process [22] M. Cetin, C. Hirsch, G.K. Serovy, Application of Modified Loss and
optimization control of turbofan engine based on variable guide vane Deviation Correlations to Transonic Axial Compressors, Advisory
adjustment, J. Aero. Power 31 (11) (2016) 2801e2808. Group for Aerospace Research and Development Report, No.
[6] H. Chang, W.G. Zhao, D.H. Jin, Z.Y. Peng, X.M. Gui, Numerical AGARD-R-745, 1987.
investigation of base-setting of stator’s stagger angles for a 15-stage [23] T. Rowan, Functional Stability Analysis of Numerical Algorithms,
axial-flow compressor, J. Therm. Sci. 23 (1) (2014) 36e44. Ph.D. thesis, University of Texas at Austin, Austin, 1990.
[7] L. Shi, B. Liu, P. Zhang, J. Li, L. Wang, One-dimensional charac- [24] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist
teristic optimization design for ten-stage high pressure compressor in multiobjective genetic algorithm: NSGA-II, IEEE Trans. Evol. Com-
commercial engine, J. Aero. Power 28 (7) (2013) 1564e1569. put. 6 (2) (2002) 182e197.
[8] H. Roh, S. Daley, The online optimisation of stator vane settings in [25] J.K. Schweitzer, B.T. Brown, Advanced Industrial Gas Turbine
multi-stage axial compressors, Int. J. Adv. Mechatron. Syst. (IJA- Technology Readiness Demonstration Program: Final Report, Phase I:
MechS) 1 (4) (2009) 266e279. Design Study, USA Department of Energy Report No. HCP/T 5035,
[9] N.M. White, A. Tourlidakis, R.L. Elder, Axial compressor perfor- 1977.
mance modelling with a quasi-one-dimensional approach, Proc. Inst. [26] J.K. Schweitzer, J.D. Smith, Advanced Industrial Gas Turbine Tech-
Mech. Eng. Part A: J. Power Energy 216 (2) (2002) 181e193. nology Readiness Demonstration Program: Phase II Final Report:
[10] L. Gallar, M. Arias, V. Pachidis, P. Pilidis, Compressor variable ge- Compressor Rig Fabrication Assembly and Test, USA Department of
ometry schedule optimisation using genetic algorithms, in: Proceedings Energy Report No. DOE/OR/05035-T2, 1981.
of ASME Turbo Expo: Power for Land, Sea, and Air, Orlando, Florida, [27] A.R. Wadia, D.P. Wolf, F.G. Haaser, Aerodynamic design and testing
USA, June 8-12, Paper No. GT2009-60049, 2009. of an axial flow compressor with pressure ratio of 23.3:1 for the
[11] S. Reitenbach, M. Schnoes, R.G. Becker, T. Otten, Optimization of LM2500þ gas turbine, J. Eng. Power 124 (3) (2002) 331e340.
compressor variable geometry settings using multi-fidelity simulation, in: [28] S.J. Cline, W. Fesler, H.S. Liu, R.C. Lovell, S.J. Shaffer, High Pres-
Proceedings of ASME Turbo Expo: Power for Land, Sea, and Air, sure Compressor Component Performance Report, National Aero-
Montreal, Quebec, Canada, June 15e19, Paper No. GT2015-42832, 2015. nautics and Space Administration Report, No. NASA-CR-168245,
[12] Z.T. Wang, J. Li, K. Fan, S.Y. Li, The off-design performance simu- 1983.
lation of marine gas turbine based on optimum scheduling of variable [29] T. Ikeguchi, A. Matsuoka, Y. Sakai, Y. Sakano, K. Yoshiura, Design
stator vanes, Math. Probl Eng. (2007) Article ID 2671251. and development of a 14-stage axial compressor for industrial gas
[13] B. Li, C.W. Gu, X.T. Li, T.Q. Liu, Numerical optimization for stator turbine, in: Proceedings of ASME Turbo Expo: Power for Land, Sea,
vane settings of multi-stage compressors based on neural networks and Air, Copenhagen, Denmark, June 11-15, Paper No. GT2012-
and genetic algorithms, Aero. Sci. Technol. 52 (2016) 81e94. 68524, 2012.

View publication stats