Beruflich Dokumente
Kultur Dokumente
M. Fadl1
Modeling for Transient Natural
Department of Engineering Science,
University of Oxford,
Oxford OX1 3PJ, UK
Convection
e-mail: m.s.fadl@lboro.ac.uk Turbine flexible operations with faster startups/shutdowns are required to accommodate
2 emerging renewable power generations. A major challenge in transient thermal design and
L. He analysis is the time scale disparity. For natural cooling, the physical process is typically in
Department of Engineering Science, hours, but on the other hand, the time-step sizes typically usable tend to be very small (sub-
University of Oxford, seconds) due to the numerical stability requirement for natural convection as often
Oxford OX1 3PJ, UK observed. An issue of interest is: What time-step sizes can and should be used in terms of sta-
e-mail: Li.He@eng.ox.ac.uk bility as well as accuracy? In this work, the impact of flow temporal gradient and its model-
ing is examined in relation to numerical stability and modeling accuracy for transient
P. Stein natural convection. A source term-based dual-timing formulation is adopted, which is shown
GE Power, to be numerically stable for very large time-steps. Furthermore, a loosely coupled procedure
Baden 5400, Switzerland is developed to combine this enhanced flow solver with a solid conduction solver for solving
unsteady conjugate heat transfer (CHT) problems for transient natural convection. This
G. Marinescu allows very large computational time-steps to be used without any stability issues, and thus
GE Power, enables to assess the impact of using different time-step sizes entirely in terms of a temporal
Baden 5400, Switzerland accuracy requirement. Computational case studies demonstrate that the present method can
be run stably with a markedly shortened computational time compared to the baseline
solver. The method is also shown to be more accurate than the commonly adopted quasi-
steady flow model when unsteady effects are non-negligible. [DOI: 10.1115/1.4037721]
1 Introduction commonly used procedures, have been based on heat transfer coef-
ficient (HTC), with an assumption that HTC is largely invariant
The growing shares of renewable energy sources in the market
with respect to wall temperature. A direct HTC-based approach
and emerging solar power applications have set higher require-
could be in considerable errors under certain conditions for forced
ments on steam turbine operations. In order to design reliably a
convection heat transfer in a gas turbine blade passage, e.g., as
steam turbine for flexible operations, it is essential to understand
observed by Maffulli and He [5,6], Zhang and He [7]. For a natural
dynamic thermal characteristics of steam turbine components dur-
convection problem of the present interest, the near-wall flow is
ing a transient process. Nowadays, a power plants competitive-
dominantly driven by the wall temperature, making a HTC-based
ness is measured not only on the performance per se but also on
approach even more difficult. This strong dependence of flow on
the overall plant operational economy. This includes the ability to
the wall thermal condition in natural convection presents a clear
start up and shut down quickly, and frequently for some applica-
case for pursuing conjugate heat transfer (CHT) solutions for
tions, while having an adequate high cycle thermal fatigue life.
coupled fluidsolid domains.
For instance, a cooling down process will have to take into
For general CHT method developments and applications, the
account material strength limits and usage factors [1,2], and the
time scale disparity between fluid and solid parts as a fundamental
residual temperature gradients when restarted may have some
feature must be recognized. The ratio between the two time scales
considerable impact on turbine life span. An adequate prediction
can be up to 104 for gas turbine blades, as pointed out by He and
of component metal temperatures before a restart becomes
Oldfield [8]. This time scale disparity presents a challenge for
increasingly important, and is one of the key capabilities required
numerical solution methods adopted for the two domains, as well
for a transient steam turbine cooling process. Note also that steam
as for the interface treatment between them. A direct simultaneous
turbines in solar power plants would face many more startups and
coupling is a seemingly straightforward option, but can be hugely
shutdowns compared to typical base-load machines.
time consuming. On the other hand, the time scale disparity can
A number of efforts have been made to develop modeling-
also serve as a justification for using a quasi-steady flow model
based methods for prediction of turbine shutdown and cooling
for transient CHT, i.e., coupling an unsteady solid solver with a
processes. Marinescu et al. [3,4] developed a new procedure for
steady fluid solver at each physical time-step. The quasi-steady
assessing the natural cooling of steam turbine components, vali-
flow-based loosely coupled CHT methods have been actively
dated with experimental measurements. The numerical cooling
developed and applied to forced convection heat transfer problems
calculation is effectively based on a concept to approximate the
for turbine configurations and others, e.g., Refs. [911]. For natu-
fluid gross buoyancy during a natural cooling by an equivalent
ral convection, there have been some attempts to use the direct
fluid conductivity, calibrated against experimental data. The pro-
coupled CHT solutions, e.g., Refs. [2], [12], and [13], known to
cedure, though with notable empiricism, was shown to work quite
be computationally time consuming. However, the present authors
well for natural cooling predictions of steam turbines.
are not aware of any published efforts of using a loosely coupled
In terms of higher fidelity, computational fluid dynamics-based
CHT approach for natural convection.
convective heat transfer analyses for turbine applications, most
A particularly relevant issue as has been observed is related to
the time-step size allowable in natural convection simulations
1
Present address: CREST, Loughborough University, Loughborough, UK, LE11 with common commercial codes (e.g., FLUENT). The numerical sta-
3TU. bility requirement tends to restrict the time-step size to a very
2
Corresponding author. small value (typically subseconds), e.g., Refs. [1214]. When a
Contributed by the Turbomachinery Committee of ASME for publication in the transient CHT solution for turbine flexible operations is pursued,
JOURNAL OF ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received July 6,
2017; final manuscript received July 11, 2017; published online September 26, 2017. this time-step limit can be particularly restrictive, given the time
Editor: David Wisler. scale disparity between the fluid and solid domain.
Journal of Engineering for Gas Turbines and Power JANUARY 2018, Vol. 140 / 012605-1
C 2018 by ASME
Copyright V
Journal of Engineering for Gas Turbines and Power JANUARY 2018, Vol. 140 / 012605-3
Table 1 Heat flux with different mesh sizes much bigger than those of the baseline FLUENT solutions while still
giving an excellent accuracy. And because of this much-enlarged
Grid size 100 25 200 50 300 75 400 100 time-step range, the corresponding trade-off between temporal
Heat flux (ID) 1382.6 1372.8 1395.9 1395.1 accuracy and computational efficiency can now be considered.
Heat flux (outer diameter) 691.3 686.4 698.0 697.5
3.4 Impact of Quasi-Steady Flow Model. Having identified
the levels of differences in terms of the time-step size, we now
solutions and the direct solutions, though with hugely different examine the impact of using a quasi-steady model for natural con-
time-step sizes. In Figs. 5 and 6, the instantaneous temperature vection. A quasi-steady model can be easily realized in the present
and velocity contours are compared for t 10 s and 160 s, respec- procedure by simply switching off the temporal gradient term
tively. Clearly, both field solutions are also in excellent (i.e., the UDF source term in Eq. (6)), when the fluid solution pro-
agreement. ceeds in the physical time, subject to corresponding transient
In terms of computing effort, the direct solution for a 10 s boundary conditions in time. Here, we consider the solution
period needs 1000 iterations (a physical time-step of 0.2 s with 20 behavior when the boundary condition is subject to two different
subiterations per step). The present solution for the same physical temporal gradients. In the first case, the boundary temperature is
time (one step) needs about 100 iterations. So, for this case, the linearly varied by 20 K in 0.2 s. In the other, the gradient is
present solution is ten times faster than the direct solution. This is reduced by a factor of 4. It is 5 K in 0.2 s.
already quite considerable. When the time-step of 160 s is taken, Figure 7 shows the instantaneous heat flux distributions with
the number of subiterations required is increased to 400500. But the different models for the high temporal gradient case. The
this increase is not proportional to the increase in the physical impact of the temporal errors associated with the quasi-steady
time-step size, thus the net gain in computational effort then model is clearly shown to be non-negligible for this case. Note
becomes a factor of 3240. that for the low gradient case (Fig. 8), although the instantaneous
Clearly, we shall need to consider the temporal errors for the heat flux distributions tend to show only small differences, the
cooling process particularly for the situation where the temporal accumulated effects at a later time may be much larger. The tran-
gradient is higher as at the start of a transient process. Overall, the sient load in the solid domain at a later time will depend on the
present results clearly illustrate that time-step sizes can be taken total integrated heat transferred across all the boundaries during
Fig. 2 Residual histories for different time-step sizes (direct Fig. 3 Residual histories for different time-step sizes (present
unsteady solver of FLUENT) source term-based unsteady solver)
the entire time period prior to that time instant. Thus, care will 4 Unsteadiness Modeling Characteristics for Natural
have to be taken with a quasi-steady fluid model for predicting Convection (FluidSolid Conjugate Heat Transfer)
transient thermal loads, even for some rapid ramping in a very
short period of time in an otherwise seemingly slow process. 4.1 Steady CHT for Coupled Annular FluidSolid
It should also be pointed out that the present method including Domains. The first conjugate heat transfer test case is for a simple
the unsteady source terms should consume largely the same 2D concentric annulus domain configuration as shown in Fig. 9.
amount of computing time as a quasi-steady transient approach The fluid domain is the same as that used for the fluid-only analy-
without the unsteady fluid source term. Thus, the present method ses. The outer boundary of the fluid domain is an isothermal wall
does have a clear advantage with a much-expanded validity at lit- at 400 K. The solid domain has an ID of 350 mm, and its outer
tle or no extra cost compared to a quasi-steady model, particularly boundary is the interface with the fluid domain. The inner bound-
when integrated to a loosely coupled CHT procedure. ary of the solid domain is an isothermal wall at 800 K. The solid
Fig. 5 Instantaneous temperature fields for the present solu- Fig. 6 Instantaneous velocity fields for the present solutions
tions ((a) Dt 5 10 s and (b) 160 s, respectively) and the direct ((a) Dt 5 10 s and (b) 160 s, respectively) and the direct baseline
baseline solutions (Dt 5 0.2 s for both cases) solutions (Dt 5 0.2 s for both cases)
Journal of Engineering for Gas Turbines and Power JANUARY 2018, Vol. 140 / 012605-5
Numerical tests are conducted for the direct CHT solver in FLU- have some considerable implications on the computational effi-
ENT by using Dt 0.2 s, 0.4 s, 0.5 s, 1.0 s, and 10 s. It is observed ciency comparison as will be discussed in Sec. 4.3.
that the solutions again can only be stable for a time-step less than
0.2 s. Figure 12 shows the convergence histories of the direct 4.3 Effect of Time-Step Size on Accuracy and Speed of
solver with different time-steps sizes. In clear contrast, Fig. 13 Loosely Coupled CHT Solution. Now, we examine the conju-
shows the convergence histories of the present loosely coupled gate heat transfer capability of the present loosely coupled CHT
procedure. The results confirm that the time-step limit is more or implementation for transient natural convection. We first look at
less the same as that for the fluid-domain only case. the effect of time-step size on both accuracy and stability of simu-
Thus, the results of the present cases indicate that it is the lation results for the given transient wall temperature variation
numerical instability of the flow solver that completely dictates (thus the same temporal gradient). The loosely coupled CHT solu-
the allowable time-step size, independent of the fluidsolid CHT tion is obtained for matching in time at different physical time-
coupling. The test results also suggest that the time-step restriction steps to check the accuracy and stability of the simulation results.
on the baseline flow solver should remain largely the same regard- For consistency of the comparisons, all solutions are restarted
less of the temporal gradient of the flowfield. from the same initial solution with Ti 600 K obtained at the end
On the other hand, the present loosely coupled CHT methods of the 10 s period of the linear variation of the inner solid wall
show no upper limit on the time-steps sizes, similarly to the fluid- boundary temperature.
domain only cases, e.g., as shown Fig. 3. The stark contrast can Figure 14 shows the effects of different time-step sizes on the
numerical predictions of wall temperature distributions at
Journal of Engineering for Gas Turbines and Power JANUARY 2018, Vol. 140 / 012605-7
Fig. 20 Comparison of instantaneous heat flux contours at fluidsolid interface (at t 5 200 s),
between the direct solution (Dt 5 0.2 s) and the present loosely coupled solution (Dt 5 10 s): (a)
direct CHT and (b) loosely coupled CHT
Journal of Engineering for Gas Turbines and Power JANUARY 2018, Vol. 140 / 012605-9
Acknowledgment
The present work is sponsored by UK Engineering and Physical
Science Research Council (EPSRC) and GE Power.
References
[1] Topel, M., Genrup, M., J ocker, M., Spelling, J., and Laumert, B., 2015,
Operational Improvements for Startup Time Reduction in Solar Steam
Turbines, ASME J. Eng. Gas Turbines Power, 137(4), p. 042604.
[2] Born, D., Stein, P., Marinescu, G., Koch, S., and Schumacher, D., 2016,
Thermal Modeling of an Intermediate Pressure Steam Turbine by Means of
Conjugate Heat TransferSimulation and Validation, ASME J. Eng. Gas Tur-
bines Power, 139(3), p. 031903.
[3] Marinescu, G., Mohr, W. F., Ehrsam, A., Ruffino, P., and Sell, M., 2013,
Experimental Investigation in to Thermal Behavior of Steam Turbine
ComponentsTemperature Measurements With Optical Probes and Natural
Cooling Analysis, ASME J. Eng. Gas Turbines Power, 136(2), p. 021602.
[4] Marinescu, G., Stein, P., and Sell, M., 2015, Natural Cooling and Startup of
Steam Turbines: Validity of the Over-Conductivity Function, ASME J. Eng.
Gas Turbines Power, 137(11), p. 112601.
[5] Maffulli, R., and He, L., 2014, Wall Temperature Effects on Heat Transfer
Coefficient for High Pressure Turbines, AIAA J. Propul. Power, 30(4), pp.
10801090.
[6] Maffulli, R., and He, L., 2017, Impact of Wall Temperature on Heat Transfer
Fig. 21 Comparison of instantaneous temperatures on a mid- Coefficient and Aerodynamics for 3-D Turbine Blade Passage, ASME J.
domain cut plane (at t 5 200 s) between the direct solution Therm. Sci. Eng. Appl., 9(4), p. 041002.
(Dt 5 0.2 s) and the present loosely coupled solution (Dt 5 10 s): [7] Zhang, Q., and He, L., 2014, Impact of Wall Temperature on Turbine Blade
(a) direct CHT and (b) loosely coupled CHT Tip Aerothermal Performance, ASME J. Eng. Gas Turbines Power, 136(5),
p. 052602.
[8] He, L., and Oldfield, M. L. G., 2011, Unsteady Conjugate Heat Transfer Mod-
adiabatic. The corresponding inner surfaces of these side walls are eling, ASME J. Turbomach., 133(3), p. 031022.
subject to the fluidsolid interface treatment. [9] Sun, Z., Chew, J. W., Hills, N. J., Volkov, K. N., and Barnes, C. J., 2010,
Efficient Finite Element Analysis/Computational Fluid Dynamics Thermal
Figure 20 shows the comparison in instantaneous heat flux on Coupling for Engineering Applications, ASME J. Turbomach., 132(3),
the fluidsolid interface at t 200 s between the direct CHT solu- p. 031016.
tion (Dt 0.2 s) and the present loosely coupled CHT (Dt 10 s). [10] Errera, M., and Baque, B., 2013, A Quasi-Dynamic Procedure for Coupled
The corresponding comparison in the temperature contours Thermal Simulations, Int. J. Numer. Methods Fluids, 72(11), pp. 11831206.
[11] Wang, Z., Corral, R., Chaquet, J. M., and Pastor, G., 2013, Analysis and
between the two solutions is given in Fig. 21 for a cut plane nor- Improvement of a Loosely Coupled Fluid-Solid Heat Transfer Method, ASME
mal to the axis at the middle point of the domain. Paper No. GT2013-94332.
All the comparisons demonstrate excellent agreement between [12] Altac, Z., and Ugurlubilek, N., 2016, Assessment of Turbulence Models in
the present loosely coupled solutions and the baseline direct CHT Natural Convection From Two- and Three-Dimensional Rectangular
Enclosures, Int. J. Therm. Sci., 107, pp. 237246.
solutions for this 3D configuration. The hugely different time- [13] Ma, J., and Xu, F., 2015, Transient Flows Around a Fin at Different Positions,
steps usable for the two methods indicate a similar computational Procedia Eng., 126, pp. 393398.
efficiency gain in a 3D case to that in a 2D case. [14] Xu, F., and Saha, S. C., 2014, Transition to an Unsteady Flow Induced by a
Fin on the Sidewall of a Differentially Heated Air-Filled Square Cavity and
Heat Transfer, Int. J. Heat Mass Transfer, 71, pp. 236244.
[15] He, L., and Fadl, M., 2017, Multi-Scale Time Integration for Transient Conju-
5 Conclusions gate Heat Transfer, Int. J. Numer. Methods Fluids, 83(12), pp. 887904.
[16] Perelman, T. L., 1961, On Conjugated Problems of Heat Transfer, Int. J. Heat
Impact of the temporal gradient of unsteady flow and its model- Mass Transfer, 3(4), pp. 293303.
ing in relation to numerical stability and modeling accuracy is [17] Jameson, A., 1991, Time Dependent Calculations Using Multigrid, With
assessed for the transient natural convection prediction relevant to Applications to Unsteady Flows Past Airfoil and Wings, AIAA Paper No. 91-
1596.
steam turbine flexible operations. A source term-based dual tim- [18] Arnone, A., Liou, M. S., and Povinelli, L. A., 1995, Integration of Navier-
ing formulation is adopted and implemented in a commercial Stokes Equations Using Dual Time Stepping and a Multigrid Method, AIAA
solver. Furthermore, a loosely coupled procedure has been J., 33(6), pp. 985990.
applied, for the first time, to transient conjugate heat transfer pre- [19] He, L., 2000, 3D Navier-Stokes Analysis of Rotor-Stator Interactions in Axial
Flow Turbines, Proc. Inst. Mech. Eng., Part A, 214(1), pp. 1322.
dictions for natural convection. The results have consistently dem- [20] Giles, M. B., 1997, Stability Analysis of Numerical Interface Conditions in
onstrated that the present source term-based unsteady flow solver FluidStructure Thermal Analysis, Int. J. Numer. Methods Fluids, 25(4), pp.
allows for much larger computational time-steps to be used 421436.