Professional Documents
Culture Documents
http : //www.cigre.org
(to be centred)
SUMMARY
Accurate prediction of winding hotspot temperature is critical for reliability of power transformer
design. Thermal analysis of a transformer disc winding has been performed using CFD analysis in
three and two dimensions as well as thermohydraulic networks. If the losses and mass flow rate are
scaled to account for the existence of spacers, the results show good correspondence between 2D and
3D. The average winding temperature is the least sensitive to the modelling approach, whereas the
exact value of the hot spot temperature depends on local flow conditions and therefore shows larger
variation. Due to the considerably lower computational effort compared to 3D modelling, 2D CFD
simulation is the preferred approach for transformer thermal validation, whereas thermohydraulic
network modelling is suitable for design and optimization.
KEYWORDS
Transformer thermal modelling, CFD, thermohydraulic networks, hot spot
tor.laneryd@se.abb.com
1. INTRODUCTION
Trustworthy prediction and measurement of the winding hot spot temperature are vital in
establishing the thermal design quality of a power transformer, guaranteeing that the ageing of
insulation materials does not exceed the guarantee limits to fulfill the requirements on reliability.
[1] In the past, hot spot factors have played an important role in predicting the hot spot
temperature, but this approach has limitations. The hot spot factor is often taken as a default
value of 1.3 (as suggested by the standards [2] [3] in case no further information on the actual
design is known) or estimated by loss calculations. In recent years, the hydrodynamic effects
inside the winding have been recognized to have a large effect on the local cooling performance.
This affects both the magnitude and the location of the hot spot, which may therefore differ from
the position of the highest losses. For this reason, fiber-optic measurements of hot spot
temperature are only dependable if they are installed in the right location. [4]
Today direct calculation of the temperature distribution in the entire transformer winding
using detailed thermal simulations is entirely feasible. The choice of modelling approach is a
balance between computational effort and ensuring that geometrical aspects and detailed loss
distribution are taken into account. Large power transformers are typically customized
according to the specific needs of each installation, and are normally manufactured in small
series with a unique electrical, mechanical and thermal design. Fast computational speed is
therefore essential.
Thermo-hydraulic networks are a computationally efficient approach to model the
combination of oil and heat transport to predict the position and level of the hot spot
temperature, suitable for the generally well-defined and confined nature of a winding. Of
course, fast computational speed cannot be permitted to cause significant loss of accuracy.
More details of the flow field can be considered by using Computational Fluid Dynamic
(CFD) modelling. The level of model complexity can be optimized based on the
characteristics of the winding geometry and cooling type. [5]
The heat transfer problem can be described as conjugate heat transfer with natural or mixed
convection of a high Prandtl number fluid in the laminar regime through a series of parallel
cooling ducts. Because of the axisymmetric geometry that is obtained if the spacers in the
transformer winding are neglected, it is tempting to use a two-dimensional CFD approach to
reduce computational cost, and this is indeed the strategy employed in most of the existing
publications. It’s important to validate that this simplification does not cause erroneous results.
Incorrect temperature levels in the solution process will adversely affect the accuracy if the flow
is dominated by natural convection.
It should be mentioned that there are many reasons for discrepancies between CFD modelling
results in comparison to measured results. There are errors involved in the formulation of the
numerical method and its application, but also errors and uncertainty in how measurements are
performed. In this investigation, our target is to estimate the modelling errors due to reduced
geometrical complexity when a three-dimensional problem is approximated by a two-
dimensional model or a thermohydraulic network.
The topic is not new. Comparisons between 3D CFD modelling and other models have been
presented in [6], [7] and [8], with the general conclusion that the more inexpensive modelling
approaches do not provide sufficient accuracy. The novelty of the present investigation is
therefore that we propose a 2D method that seems to offer a good accuracy compared to 3D
modelling without any need for post-processing correction coefficients.
2
2. WINDING TEMPERATURE CALCULATION METHODS
3
3. CASE STUDY
Calculations have been performed on a geometry corresponding to the low voltage winding of
a 160 MVA transformer with ONAF cooling. It is a disc winding with 106 discs and mineral
oil as the coolant. All CFD modelling has been performed in Ansys Fluent 14.
For a well-designed winding, the hot spot can be expected to occur in the top part of the
winding at rated load. To reduce the computational effort, only the two top sections of the
winding is modelled in 3D. Because the solution is sensitive to the boundary conditions at the
inlet, a CFD analysis of the full winding has been performed using the 2D axisymmetric
formulation with rescaled properties, and the velocity profile and temperature profile have
been used as inlet boundary conditions for the 3D model. For the given design, the covered
disc surface by spacer corresponds to 27% of the total disc surface and the volume loss and
mass flow rate had to be rescaled correspondingly. The 3D model was later used as a
reference and the results were compared to three different models (2D with rescaling, 2D
without rescaling and THNM). The calculation cases are presented in Table 1.
The disc losses in the winding are considered non-uniform due to eddy losses and are
presented in Figure 1. In the bottom disc, only the heat losses that are cooled from above (as
concluded from the 2D full winding model) are included.
Figure 1 : Winding losses per disc in the CFD model. Additional eddy losses are introduced at the top of
the winding. The bottom disc has reduced losses to compensate that the cooling duct below is not
modelled.
Detailed spacer heights have been considered throughout the whole model to correctly capture
the hydraulic effects and their influence of the temperature distribution. The accuracy of the
models is demonstrated by comparing radial oil duct flow velocities, and winding disc maximum
4
and average temperatures between the thermohydraulic network model and the two-dimensional
CFD models with the three-dimensional CFD model as a reference.
Contour plots of temperature rise over top oil for the different CFD models are displayed in
Figure 2. The result for the 3D CFD model is a slicing plane in the center of the cooling duct.
Figure 2: Contour plot of temperature rise over top oil for CFD model in 3D (left), 2D without rescaling
(center) and 2D with rescaling (right).
Tabulated results for the different CFD approaches as well as for the thermohydraulic network
model are presented in Table 2. The maximum temperature rise over top oil per disc is shown in
Figure 3, and the relative mass flow rate in the horizontal cooling ducts is shown in Figure 4.
Table 2: Calculated temperature rise above top oil for different models
Max ∆T in Top Average ∆T in Average ∆T in All
Disc Top Disc Discs
3D 14.0 K 10.8 K 3.9 K
2D without
17.4 K 12.7 K 2.3 K
rescaling
2D with
15.1 K 11.6 K 4.0 K
rescaling
Thermohydraulic
12.5 K 9.9 K 4.0 K
network model
5
T over top oil (K)
max
Figure 3: Maximum temperature rise over top oil per disc for 3D and 2D CFD models.
mass flow ratio (%)
Figure 4: Mass flow rate in cooling ducts between disc for 3D and 2D CFD models.
6
The results show that 2D CFD without rescaling gives a large discrepancy compared to the
3D model. For this specific design, the winding average temperature is underestimated
whereas the hotspot temperature is overestimated. This should not be considered a general
conclusion but a clear indication that the 2D CFD results without proper rescaling are not
dependable. Since the flow behavior is fundamentally different, it would be difficult to
compensate for it through post-processing.
On the other hand, the results from the 2D CFD analysis with rescaling as well as the THNM
show good correspondence with the 3D CFD model. The average winding temperature is the
least sensitive to the modelling approach, whereas the exact value of the hot spot temperature
depends on local flow conditions and therefore shows larger variation.
As seen in Figure 5, the circumferential variation in temperature is small, which demonstrates
the validity of the assumption that heat conduction effectively distributes the generated losses
to the cooling surface.
Figure 5 : Contour plot of temperature rise over top oil in a cross section of top disc in 3D CFD model.
The calculation time for models of various complexity is shown in Figure 6. The exact values
of calculation time will depend on model parameters such as mesh size, discretization scheme
and convergence criteria.
BIBLIOGRAPHY
[1] Working Group A2.38 CIGRE, “Transformer Thermal Modelling” (June 2016)
[2] IEC 60076-7 (Edition 1.0 2005-12) Power transformers – Part 7: Loading guide for oil-immersed
power transformers (Annex E), IEC, 2005
[3] IEEE Std 1538-2000 (R2011) IEEE Guide for Determination of Maximum Winding Temperature
Rise in Liquid-Filled Transformers (Annex C), IEEE, 2011
[4] T. Laneryd, A. Gustafsson, J. Kranenborg, P. Duarte, W. Calil, J. Zacharias, J.C. Mendes, “Hot
spot determination in transformer windings through CFD analysis,” VII Workspot – International
workshop on power transformers, equipment substations and materials (2014)
[5] A. Gustafsson, Y. Jiao, J. Kranenborg, T. Laneryd, B. Samuelsson, Transformer winding oil flow
rate and hot spot temperature: A straightforward relationship?,” Conference of The Electric
Power Supply Industry CEPSI (2016)
[6] P. Picher, F. Torriano, M. Chaaban, S. Gravel, C. Rajotte, B. Girard, “Optimization of transformer
overload using advanced thermal modelling”, (Cigre (2010), A2-305)
[7] F. Torriano, P. Picher, M. Chaaban, “Numerical investigation of 3D flow and thermal effects in a
disc-type transformer winding” (Applied Thermal Engineering, 40 (2012), pages 121-131)
[8] S. Tenbohlen, C. Breuer, F. Devaux, R. Lebreton, N. Schmidt, T. Stirl, “Evaluation of the thermal
performance of transformer windings by numerical investigations and measurements” (Cigre
Science & Engineering, 7 (2017), pages 16-22)
[9] F. P. Incropera, D. P. Dewitt, T. L. Bergman, A. S. Lavine, ”Fundamentals of Heat and Mass
Transfer”, 6th edition, John Wiley & Sons (2007)