A Parametrization Scheme for High Performance Thermal Models of Electric Machines using Modelica Anton Haumer ∗ Christian Kral ∗∗ Vladimir Vukovic ∗∗ Alexander David ∗∗∗ Christian Hetteisch ∗∗∗ Attila Huzsvar ∗∗∗ Austrian Institute of Technology, Gienggasse 2, A-1210 Vienna, Austria (e-mail: Anton.Haumer@ait.ac.at). ∗∗ Austrian Institute of Technology, Gienggasse 2, A-1210 Vienna, Austria ∗∗∗ University of Applied Sciences Technikum Vienna, Höchstädtplatz 5, A-1200 Vienna, Austria ∗ Abstract: Thermal models oer great advantages for enhancement of design, protection and control of electric machines. Detailed thermal models take a great number of time constants into account and provide accurate prediction of the temperatures. However, to parameterize such models detailed geometric data are needed. Whenever such detailed information is not available, or the performance of the detailed models is not satisfying, simplied thermal models as described in this paper are advantageous. The calculation of parameters is described in detail, in order to achieve best accordance with temperatures obtained from measurements or from simulations with detailed thermal models. Thermal resistances are calculated from end temperatures of a test run with constant load (and known losses). Thermal capacitances are obtained using optimization to minimize deviation of simulated and measured temperatures during the whole test run. The thermal model of an asynchronous induction machine with squirrel cage is coupled with an electrical model of the drive. For validation, simulation results of an optimally parameterized simplied model are compared with temperatures obtained by simulation of a detailed thermal model, which in turn has been validated against measurement results, both for continuous duty S1 and intermittent duty S6 (6 minutes no-load followed by 4 minutes of 140% nominal load). The deviations are not more than 4 K which is quite satisfying. Keywords: Electric machines, Induction machines, Thermal models, Model reduction, Parameter identication, Parameter optimization. 1. INTRODUCTION For many applications of electric drives it is desired to have a thermal model of the electric machine. Being able to simulate the relevant temperatures allows to take advantage of the thermal inertia during non-steady duty cycles, cooling down the machine during a period of low load condition after overload operation. The possibility to enhance the machine's design in an early stage of the engineering process has been shown in Haumer et al. (2010). However, such detailed thermal models also described in Kral et al. (2005) require detailed geometric data of the machine to calculate the parameters of the thermal model. In many cases, such a detailed thermal machine model is not performant enough, or detailed geometric data is not available from the manufacturer. In both cases a simplied thermal model is desired. An application of a simplied thermal model is checking whether a chosen machine is sucient for a given duty cycle with varying speed and load during an early design stage of the drive. Another application is the thermal protection of the machine during such load cycles, without using too many thermal sensors. A third case is represented by high precision inverter control of an asynchronous induction machine with squirrel cage rotor: Measurement of the rotor temperature is nearly impossible for practical applications, but knowing the rotor temperature allows to adjust the controller parameters. The thermal model suggested in Section 2 covers three relevant temperatures of an asynchronous induction machine with squirrel cage rotor even during non-steady operation: temperatures of the stator winding, the rotor cage and the stator core. Parametrization can be done from measurement results, as described in Section 3. For this work, simulation results obtained with a detailed thermal model which has been validated in Haumer et al. (2010) instead of measurement results have been used. • friction losses • stray load losses However, rotor core losses are nearly zero during normal operation with small slip of an asynchronous induction machine. A big part of the friction losses are used to transport the medium. For simplicity reasons, these two losses as well as the stray load losses are dissipated to thermal sinks with constant temperatures, whereas the other three losses are fed to the simplied thermal equivalence circuit. The DAE system can be obtained from gure 1 as equations: dTsw = Psw − Rsw sc (Tsw − Tsc ) dt dTrc Crc = Prc − Rrc sc (Trc − Tsc ) dt dTsc Csc = Psc + Rsw sc (Tsw − Tsc ) dt +Rrc sc (Trc − Tsc ) Csw Figure 1. Simplied thermal model For both the thermal model and the drive model as described in Section 4, the modeling language Modelica is used. Modelica is a non-proprietary, object-oriented, equation based language to conveniently model complex physical systems containing, e.g., mechanical, electrical, electronic, hydraulic, thermal, control, electric power or process-oriented subcomponents. ® ® Section 5 compares simulation results achieved with the suggested simplied thermal model with the results obtained with a detailed thermal model. 2. THERMAL MODEL A simplied thermal model is only capable of describing a limited number of relevant temperatures. In this case, it is suggested to model the following ones as shown in gure 1: • sw: temperature of the stator winding, averaged over slot and end winding region • sc : temperature of the stator core, averaged over the whole cross-section of the stator core • rc : temperature of the rotor cage, averaged over slot and end ring region The three regions are represented by thermal capacitors, they are connected with thermal resistors respectively conductors. Additionally, the coolant ow is modeled by means of a sub-library of the Modelica Standard Library the FluidHeatFlow library as described in Kral et al. (2005). A varying coolant inlet temperature is enabled by the signal input inletTemperature. The parameter record in the upper left summarizes the losses and the end temperatures of the machine, as used in Section 3 for calculation of the thermal resistances. The thermal port as described in Haumer et al. (2009); Kral and Haumer (2011) contains a thermal connector for each loss source: • • • • ohmic losses of the three stator phases stator core losses ohmic losses of the rotor cage rotor core losses −Rsc c (Tsc − Tc ) cp ṁ (TcOut − TcIn ) = Rsc c (Tsc − Tc ) (1) (2) (3) (4) Tc = TcIn + tapT (TcOut − TcIn ) (5) The coolant temperature (index c) that is relevant for the heatow from the stator core to the coolant is interpolated by parameter tapT between inlet and outlet temperature of the coolant. 3. PARAMETRIZATION Knowing the losses and the temperatures at the end of a test run (or a simulation with a detailed thermal model), from equations 1 - 5 the thermal resistances as well as the coolant's mass ow rate can be obtained since in steady state the derivatives of temperatures with respect to time are zero, i.e. all losses have to be dissipated to the coolant. Psw = Rsw sc (Tsw − Tsc ) (6) Prc = Rrc sc (Trc − Tsc ) (7) Psc + Prc + Psw = Rsc c (Tsc − Tc ) (8) cp ṁ (TcOut − TcIn ) = Psc + Prc + Psw (9) After determination of the thermal resistances which are summarized in table 1, the thermal capacitances have to be calculated. Having temperature characteristics versus time during the above mentioned test run, it is possible to vary the thermal capacitances and compare simulated with measured temperatures (or results obtained with a detailed thermal model). During motor type test for obtaining rotor temperature measurements, either a device as described in Ganchev et al. (2010) or an infrared optical sensor has to be used. Using an aggregating criterion, e.g. integrating the absolutes of the deviations with respect to time and summing up the integrals, it is possible to utilize optimization methods to determine best-t thermal capacitances. For optimization, the freely available Java-based application GenOpt http://gundog.lbl.gov/GO/ has been used. Psw [W ] Prc [W ] Psc [W ] 779.2 483.4 409.7 Tsw [°C] Trc [°C] Tsc [°C] TcIn [°C] TcOut [°C] 96.0 132.9 78.1 20.0 29.6 Table 1. Losses and temperatures Csw [J/K] Crc [J/K] Csc [J/K] 5150 13672 40469 Table 2. Thermal capacitances GenOpt writes the tuning parameters in this case the three thermal capacitances to an input text le, starts the simulation model which reads the parameter le and writes the criterion to a result le. GenOpt reads the result and varies the tuning parameters to minimize the criterion which is dened as the integral of the absolutes of the deviations between reference and simulated temperatures with respect to time. Estimating start values for the thermal capacitances from masses, the optimization with algorithm GPSCoordinateSearch unveils the optimal thermal capacitances summarized in table 2. 4. DRIVE MODEL The model of the asynchronous induction machine with squirrel cage see gure 2 is explained in detail in Kral and Haumer (2005); Haumer et al. (2009); Kral and Haumer (2011) and models the transient electromagnetic behavior as well as the generation of losses. The ohmic resistances depend on the simulated temperatures as shown in the thermal port. Figure 2. Model of the asynchronous induction machine with squirrel cage The complete model coupling the electro-mechanical model with the simplied thermal model is shown in gure 3. The characteristic of the load torque can easily be changed from a continuous load S1 to an intermittent load S6: 6 minutes no-load followed by 4 minutes 140% nominal load. The block combiTimeTable reads the temperatures obtained by measurement or simulation of a detailed model from a text le; the coolant inlet temperature is fed to the thermal model. 5. VALIDATION To ease the handling of reference results, instead of measurements a detailed thermal model as shown in gure 4 has been used. This model has been validated using measurements in Haumer et al. (2010). The temperatures for comparison are dened as: • stator winding: averaged over slot and end windings • stator core: averaged over the whole cross-section • rotor cage: averaged over slots and end rings Temperature sensors require very careful mounting to ensure optimal contact for measurements. Additonally, a sufcient number of temperature sensors has to be distributed over the cross section of interest to enable averaging which is quite challenging without destroying the iron core. For obtaining rotor temperature measurements, a device as described in Ganchev et al. (2010) or special brushes Figure 3. Model of the complete drive and sliprings have to be used to transfer measurements from the rotating part. A feasible solution was to validate local temperatures of the detailed model compared with measuements, and using these simualtion results for averaging reference temperatures. Figures 5, 6 and 7 compare reference temperatures with simulation results of stator core, stator winding and rotor cage for continuous duty S1. The maximum deviations are for stator core 3.4 K, for stator winding 0.6 K and for rotor cage 1.8 K. These deviations are satisfactory low and can be explained by the fact that compared with a detailed temperature [degC] 140 Figure 4. Detailed thermal model 120 100 80 60 rotorCage reference rotorCage simulated 40 20 0 0 0.5 1 1.5 time [s] Figure 7. Comparison of reference and simulated rotor cage temperatures for continuous duty S1 120 100 140 80 60 statorCore reference statorCore simulated 40 20 0 0 0.5 1 1.5 time [s] 2 4 x 10 Figure 5. Comparison of reference and simulated stator core temperatures for continuous duty S1 temperature [degC] temperature [degC] 140 120 100 80 60 120 statorCore reference statorCore simulated 40 20 0 0 140 5000 10000 time [s] 15000 Figure 8. Comparison of reference and simulated stator core temperatures for intermittent duty S6 100 80 140 60 statorWinding reference statorWinding simulated 40 20 0 0 0.5 1 1.5 time [s] 2 4 x 10 Figure 6. Comparison of reference and simulated stator winding temperatures for continuous duty S1 thermal model time constants are omitted that inuence the temperature versus time characteristic. Figures 8, 9 and 10 compare reference temperatures with simulation results of stator core, stator winding and rotor cage for intermittent duty S6 (6 minutes no-load followed by 4 minutes 140% nominal load). The maximum devia- temperature [degC] temperature [degC] 2 4 x 10 120 100 80 60 40 20 0 0 statorWinding reference statorWinding simulated 5000 10000 time [s] 15000 Figure 9. Comparison of reference and simulated stator winding temperatures for intermittent duty S6 models for electric machines. Proceedings of the 7th Modelica Conference, 847854. Kral, C. and Haumer, A. (2005). Modelica libraries for DC machines, three phase and polyphase machines. International Modelica Conference, 4th, Hamburg, Germany, 549558. Kral, C., Haumer, A., and Plainer, M. (2005). Simulation of a thermal model of a surface cooled squirrel cage induction machine by means of the SimpleFlow-library. temperature [degC] 140 120 100 80 60 rotorCage reference rotorCage simulated 40 20 0 0 5000 10000 time [s] 15000 Figure 10. Comparison of reference and simulated rotor cage temperatures for intermittent duty S6 tions are for stator core 3.8 K, for stator winding 2.1 K and for rotor cage 2.8 K. Although these deviations are a little bit higher than those for continuous duty S1, the results are suitable for the applications mentioned in Section 1. 6. CONCLUSION A simplied thermal model comprising three relevant temperatures has been suggested. The parameters of this model have been determined using reference data either from measurements or simulations with a detailed thermal model: • Thermal conductances have been calculated from losses and temperatures in a thermal equilibrium. • Thermal capacitances have been derived from minimizing deviations of simulated and reference temperatures versus time. The detailed thermal model used to obtain reference temperatures for validation takes into account approximately 50 time constants (with an axial discretization of 5 regions). Reducing the order of the model to 3 time constants increases performace substantially, although the loss in precision a deviation of not more than 4 K is satisfactorily low. It is planned to adapt and validate the model as well as the parameterization algorithm for permanent magnet synchronous machines, which are of interest for applications in electric vehicles. Furthermore, investigations shall me made to nd a mathematical method for model reduction. REFERENCES Ganchev, M., Kubicek, B., and Kapeller, H. (2010). Rotor temperature monitoring system. Proceeding of the XIX International Conference on Electrical Machines. Haumer, A., Bäuml, T., and Kral, C. (2010). Multiphysical simulation improves engineering of electric drives. 7th EUROSIM Congress on Modelling and Simulation. Haumer, A., Kral, C., Kapeller, H., Bäuml, T., and Gragger, J.V. (2009). The AdvancedMachines library: Loss International Modelica Conference, 4th, Hamburg, Germany, 213218. Kral, C. and Haumer, A. (2011). Object Oriented Modeling of Rotating Electrical Machines. INTECH.