Sunteți pe pagina 1din 4

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO.

8, AUGUST 2010

3197

A General Cuboidal Element for Three-Dimensional Thermal Modelling


R. Wrobel and P. H. Mellor
Department of Electrical and Electronic Engineering, University of Bristol, Bristol, BS8 1UB, UK
This paper presents a lumped parameter approach for three-dimensional (3D) thermal modelling based upon the use of general 3D elements which are formulated to accurately cater for internal heat generation. Commonly used thermal networks tend not to account for an internal heat generation that is essential for accurate temperature predictions. In contrast, the general element proposed in this paper includes the internal heat generation as well as a material thermal anisotropy. To validate the technique, thermal models of an inductor using the equivalent circuit method based around a general cuboidal element and a full 3D nite element analysis was constructed and analyzed. The calculated results from the cuboidal element method show good agreement with the FEM predictions. Index TermsFinite element analysis (FEA), general cuboidal element, inductor, thermal modelling.

I. INTRODUCTION HE thermal performance of electrical machines and wound components is a major limiting factor in their design. Accurate prediction of the temperature within a device is highly desirable. The calculation of temperatures distribution can be classied into two principal methods: analytical methods involving lumped-parameter thermal networks and numerical techniques involving nite element based analyses (FEA) catering for heat conduction and computational uid dynamics [1]. Since numerical techniques involve a lengthy model setup and large computation times particularly in the case of transient behavior, the analytical lumped-parameter network remains a valuable tool. The resistive element commonly used in thermal equivalent networks is derived from a one-dimensional solution to the steady-state heat diffusion equation for zero internal heat generation [2]. This solution, however does not include internal heat generation essential for accurate temperature predictions. To include the internal heat generation a combination of solutions for two heat diffusion equations with zero internal heat generation and zero surface temperatures is required [3]. This approach has proven its usefulness resulting in development of general elements that have been used in thermal analysis of electrical machines [4], [5]. Despite this earlier work the change in temperature distribution within an element that results from internal heat generation, compared no heat generation, is often overlooked in thermal networks. This paper extends the earlier work by demonstrating the additional accuracy achieved by accounting for internal heat generation in the structure of the thermal network. Using an inductor as an exemplar, the temperature predictions of a standard and the more accurate formulation of the lumped parameter network are compared against a full 3D nite element solution. The lumped network equivalent of a general cuboidal element in Cartesian coordinates with internal heat generation and material thermal anisotropy is reviewed.
Manuscript received November 18, 2009; accepted February 10, 2010. Current version published July 21, 2010. Corresponding author: R. Wrobel (e-mail: r.wrobel@bristol.ac.uk). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TMAG.2010.2043928

Fig. 1. Outline of the general cuboidal element.

Fig. 2. Equivalent thermal network for the general cuboidal element.

II. MATHEMATICAL DESCRIPTION To dene the general cuboidal element an assumption of independent heat transfer within axes of the Cartesian coordinate system has been made. That simplies the cuboidal element derivation to a one-dimensional problem as the all axes of the coordinate system can be analyzed individually. Fig. 1 shows a geometrical representation of the general and are the surface tempercuboidal element, where is the thermal conductivity in -axis, is the atures, is the surface area normal to conduction heat rate in -axis, , and is dimension of the cuboid in the -axis. The equivalent thermal network in the -axis for the cuboid is shown in Fig. 2. The mean temperature and the total heat generation within the cuboid are given by and respectively, where (1) To determine the thermal resistances ( and ), the two cases of the one-dimensional, steady-state heat diffusion

0018-9464/$26.00 2010 IEEE

3198

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 8, AUGUST 2010

equation need to be analyzed for: . zero internal heat generation and . zero surface temperatures and then applying superposition. A. Zero Internal Heat Generation The heat equation for one-dimensional steady-state example with zero internal heat generation and constant thermal conducis given in the following form: tivity (2) Assuming that thermal conductivity is constant, the (2) may be integrated twice to obtain a general solution (3) Applying the boundary conditions, the integration constants are found and Substituting (11) into the general solution, the temperature distribution is then

Fig. 3. Temperature variation for one-dimensional heat equation with internal heat generation.

(4) Substituting (4) into the general solution, the temperature distribution is then (5) The conductive heat rate within the cuboid is given as (6) The equivalent thermal resistance for conduction is then (7) (14) In the network given in Fig. 2 this resistance is split into two equal thermal resistances and , where (8) The negative value for (14) can be explained by considering the temperature prole with heat generation, as illustrated in Fig. 3. The standard two resistor network without would overpredict the average temperature in the element, thereby leading to inaccuracies particularly when considering the change in heat storage within the element during a thermal transient. Fig. 4 presents the complete 3D equivalent network for the general cuboidal element. The network is made of individual thermal circuits representing each axis of the Cartesian coordinate system. The common node is at the point of average temperature for the element, Fig. 3. The total heat generation within the cuboid is . A resistive thermal network yields only the steady state temperature distribution within an element. To cater for thermal transients, thermal capacitance needs to be included in the equivalent network. The capacitance together with the heat generation source is connected to the node of average temperature as shown in Fig. 4. Thermal capacitance for the cuboidal element is (15) The third thermal resistance of the network is determined from mean temperature , the total heat generation and parallel combination of the resistance and The average temperature within the cuboid is then (12)

(13)

B. Zero Surface Temperatures The heat equation for one-dimensional steady-state example with internal heat generation is given in the following form: (9) For constant thermal conductivity twice to obtain general solution (11) may be integrated

(10) Applying the rst kind boundary conditions, the integration constants can be found and

(11)

WROBEL AND MELLOR: A GENERAL CUBOIDAL ELEMENT FOR THREE-DIMENSIONAL THERMAL MODELLING

3199

Fig. 4. Three-dimensional network representation of the general cuboidal element.

Fig. 5. Outline of the inductors model with region labels.

Formal equation (16)

where is the specic heat capacity, is the density and is the volume of cuboidal element. Assuming a homogeneous density and thermal capacity alongside independent heat ow in each Cartesian axis (16) reduces to (13). As the thermal conductivity has been dened independently for axes of the Cartesian coordinate system the cuboidal element is able to account for material anisotropy. This feature is particularly useful when composite materials are analyzed. However, a tensor of thermal conductivity is required (17) As such data is not commonly available the authors have developed an experimental setup to measure the thermal conductivity tensor. This is described in the next section of the paper. III. FINITE ELEMENT THERMAL MODEL A. 3D Thermal Model To demonstrate the applicability of the cuboidal element a thermal analysis of a power inductor has been carried out. The inductor includes a laminated core pack and type 8 Litz wire wound coil. Fig. 5 presents an outline of the inductor together with region labels assumed in the equivalent network model. Due to symmetry only a quarter of the inductor is modelled. The equivalent thermal network model includes only 13 cuboidal elements in contrast to 70 629 elements in the case of the FEA. It has been assumed that the inductor is mounted on a cold-plate whose temperature is xed at 60 C. Surfaces that are not in contact with the cold-plate are adiabatically insulated. The Litz wire wound coil is represented as a lumped region

Fig. 6. Sample of impregnated Litz wire prior to testing.

with anisotropic properties that were obtained experimentally. Calculations were carried out for a constant power loss of 40 W within the inductors winding. B. Anisotropic Material Properties The principles of the experimental setup for the thermal conductivity measurements are based on solution of the one-dimensional heat transfer problem, Fig. 1. The conduction (energy diffusion) rate equation for this problem is given by (6). From (6) the thermal conductivity in axis is (18) Therefore by measuring the temperature gradient between the wall surfaces of a material sample and power transfer across the thermal conductivity can be derived. Fig. 6 the wall presents a sample of impregnated Liz wire representing lumped region of the inductor winding. The sample has been formed approximately as a cube. This shape ensures the same measuring conditions in all three axes of the sample, Fig. 6. To emulate the one-dimensional heat transfer the sample is mounted between a liquid-cooled cold plate at xed temperature and power resistor energized from a DC source, Fig. 7. This represents the main heat path. The remaining outer surfaces are thermally insulated. Outer surface of the power resistor shown in Fig. 7 is also thermally insulated during a test. The sample is

3200

IEEE TRANSACTIONS ON MAGNETICS, VOL. 46, NO. 8, AUGUST 2010

eraged over elements of discretization mesh assigned for the subregions. The results from FEM and cuboidal element network are showing good agreement. If the effect of the internal heat generation is neglected the thermal network indicates a slower transient response and overpredicts the steady state temperature. V. CONCLUSION The cuboidal element technique presented allows for fast three-dimensional temperature prediction using the equivalent circuit method. This is essential when a transient thermal analysis is required. The element includes the internal heat generation and material anisotropy, in contrast to the commonly used resistive element. This enables accurate model representation when composite material are analyzed. Due to its fast solving time the method is a desirable tool in a coupled electromagnetic and thermal optimization. Model construction using the cuboidal element does not require initial information about the heat ux and only geometrical and material data is used. This new technique has been validated with a 3D FEM model showing good agreement. The proposed T network aims to improve the accuracy of the mean temperature estimate within an element. Inevitably compared to a full numerical model lumped network approach will introduce approximations, for example temperature dependence of thermal conductivity is not catered for. However in identifying the mean temperature as a separate node of the network a more accurate model for the thermal transient response is obtained. Further based on the dening equations the thermal resistance values could be adjusted to identify the peak temperature if required, an approach which has been used to identify end-winding hot spots in [3]. ACKNOWLEDGMENT
Fig. 8. Thermal transient within the inductors winding.

Fig. 7. Experimental setup for the thermal conductivity measurements. TABLE I MATERIAL THERMAL DATA

instrumented with a number of type-K thermocouples to measure the temperature differential. Table I includes material data assumed in the thermal analysis. It is shown that the measured thermal conductivity for the winding and core regions is axis dependant. The thermal conductivity is temperature independent for the analyzed temperature range. IV. RESULTS Fig. 8 presents temperature predictions from the 3D FEM model, the equivalent network method with cuboidal element and the classical thermal network approach which neglects in. The calculated results show ternal heat generation the variation of averaged temperature within inductors winding Fig. 8. In the case of the FEM model the temperature was av-

The authors thank R. Davies from the Department of Electrical and Electronic Engineering at the University of Bristol for his assistance with manufacture of the experimental setup together with the material samples. REFERENCES
[1] A. Boglietti, A. Cavagnino, and D. A. Staton, TEFC induction motors thermal models: A parametric sensitivity analysis, IEEE Trans. Ind. Applicat., vol. 41, no. 3, pp. 756763, 2005. [2] F. P. Incropera and D. P. DeWitt, Introduction to Heat Transfer. Hoboken, NJ: Wiley, 1996. [3] P. H. Mellor, Improvement in the efciency and thermal ageing of single and parallel machine drives, Ph.D. dissertation, University of Liverpool, U.K., 1983. [4] P. H. Mellor, D. Roberts, and D. R. Turner, Lumped parameter thermal model for electrical machines of TEFC design, Proc. Inst. Elect. Eng.-B, vol. 138, no. 5, pp. 205218, 1991. [5] I. J. Perez and J. G. Kassakian, A stationary thermal model for smooth air-gap rotating electric machines, Elect. Power Compon. Syst., vol. 3, no. 3, pp. 285303, 1979.

S-ar putea să vă placă și