Large signal inductor modeling for switch mode power supplies
Transcription
Large signal inductor modeling for switch mode power supplies
Large signal inductor modeling for switch mode power supplies Shen Xiao, Silvio Misera, Sven Bönisch, Lutz Göhler Fakultät Verkehrs- und Maschinensysteme Technische Universität Berlin 10623 Berlin Germany Fakultät für Ingenieurwissenschaften und Informatik Hochschule Lausitz 01968 Senftenberg Germany [email protected] [email protected] Kjellberg Finsterwalde GmbH Leipziger Strasse 82, 03238 Finsterwalde Germany [email protected] Abstract— In this paper, a modified Jiles-Atherton model within APLAC is introduced. This model can be used to to simulate inductors and transformers in a switch-mode power supply at large signal input. An optimization method is coupled to an iterative algorithm to identify the Jiles-Atherton parameters for a modified Jiles-Atherton model. The verification of the modified model is done by comparing simulated hysteresis loops to experimental results at different frequencies. The results state that the modified model is valid. Key Words— Hysteresis; Modified JA model; Parameter identification; Genetic algorithm; Jiles-Atherton model I. INTRODUCTION Inductors and transformers which are used in switch-mode power supplies usually work at large signal input. Large signals may cause intensive inrush currents within transformers and inductors and lead to saturation. Saturation, as a phenomenon of lowering inductance with increasing currents in inductors and transformers, is often lowering system efficiency. It should be investigated by inspecting the magnetic hysteresis loop. Therefore, hysteresis models are needed for simulation of inductors and transformers. Among all hysteresis models, the Jiles-Atherton (JA) model is widely used and has been implemented in various system simulation tools. However, if it is compared with the results from experiment, its parameters should be further improved. The purpose of this work is to use a modified JA model which has already been implemented in APLAC, to propose an appropriate identification technique for model parameters and to let the hysteresis loop simulated by a modified JA model fit the measurement results. The simulations are carried out at different frequencies in order to investigate the validity of the modified JA model. Finally, corresponding hysteresis loops are compared to evaluate the determined parameters of this modified JA model. II. MODIFIED JILES-ATHERTON HYSTERESIS MODELING WITHIN APLAC As a simple and accurate modeling technique for the magnetic hysteresis loop, the Jiles-Atherton model has been continuously developed and improved over time. In this section, the principle of JA model is to be elucidated. Besides, a modified JA model within APLAC is then introduced. A. Original Jiles-Atherton Model In the original JA model, total magnetization can be divided into two parts, an irreversible component , which is representing the magnetic domain wall displacement against pinning site of defect in ferromagnetic material, and a reversible component , which accounts for the reversible rotation of domain wall, . The relation between the magnetic effective field and the magnetic field is expressed as: In Eq. (1), is the interdomain coupling coefficient. The anhysteretic magnetization can be represented by using the modified Langevin equation: In Eq. (2), is the saturated magnetization, characterizes the shape of anhysteretic magnetization curve. Then, the differential irreversible susceptibility can be obtained in the simplified form [1]: Finally, the total differential susceptibility which can describe the dynamic behavior of magnetic hysteresis loop is derived as follow: B. Modified Jiles-Atherton Model in APLAC In APLAC, a modified equation has been given, which indicates that the magnetic effective field can be associated with the magnetic field by using the anhysteresis magnetization: By adopting Eq. (5), the differential irreversible susceptibility is obtained as: In Eq. (6), the derivative of depends on both the derivative of the anhysteretic magnetization and the displacement from anhysteretic curve. To fully describe hysteresis, an additional equation of reversible magnetization is needed: After Eq. (7) is differentiated and added to Eq. (6), the modified JA model can be given by: Comparing Eq. (8) to Eq. (4), it can be found that the equation of the modified JA model in APLAC is completely different from that of the original JA model. III. Determination of modified Jiles-Atherton model parameters The procedure of parameter determination is split up into two phases: iterative calculation for preliminary model parameters and further optimization with a genetic algorithm. A. Iterative Method For Parameter Calculation The equations of the total differential susceptibility and its complementary relations are given by: Compared to the parameters of original JA model, those of the modified JA model have different mathematical expressions and are determined using experimental data. In Eq. (10), and are flux density and magnetic field at saturation of the hysteresis loop. Both of them are extracted from experimental results. According to the measured magnetic properties and Eq. (9a), the initial susceptibility is calculated at the initial point ( ) of the initial magnetization curve ( ): Eq. (11b) is derived from reference [3], once is initially known, the parameter can be determined by using Eq. (11a), which is written as: At coercivity ( , previously determined, parameter the following equation: ), once , and are can be calculated using At remanence , parameter can be obtained by the following equation: the Finally, regarding the use of the coordinates of the loop tip ( ), considering the previous determination of parameters , and , the parameter can be obtained by solving: Since there are four equations needed for determining the parameters which are expressed implicitly in terms of these and other parameters, by using Eq. (12), Eq. (13), Eq. (14) and Eq. (15) successively in an iterative algorithm, the value of , , and can be obtained. First, the initial value of and are given. After the first four parameters are calculated as discussed above, the procedure for the calculation is then repeated for several times and the preliminary group of model parameters are finally given. B. The Relation Between Parameters And Hysteresis Loop Because the calculation result of the iterative algorithm is very sensitive to its initial values, it is necessary to elucidate the effect of each parameter for the hysteresis loop. The parameter reflects the coupling between magnetic domains, when increases, the slope of hysteresis loop increases. The parameter a represents the shape of the anhysteretic magnetization curve, the slope of the hysteresis loop decreases with increasing . According to the change trend of the hysteresis loop, two suitable estimated values of and can be given for the iterative calculation. The parameter and relate to the hysteresis loss, the hysteresis loss increases with decreasing parameter or increasing parameter . C. Parameters Optimization with a genetic algorithm After the preliminary parameters are calculated with an iterative algorithm, a further improvement for matching of experimental data and simulation results can be applied to above relationships. First, the hysteresis loop with those five calculated parameters is simulated in APLAC. After, the result of the model simulation is compared to the result of the measurement, five proper variable domains can be generated by introducing another five parameters. The change of each parameter is limited in between the values of the calculated preliminary parameter and simulated parameter. The fitness function f(s) is implemented as follows: While In Eq. (17), is the sampling number of a half period hysteresis loop from experiment. The magnetization is evaluated by Eq. (8) and compared with the jth point of the experimental results under analysis for the same H. In Eq. (16), in order to minimize the differences between measurement and simulation, an optimum group of parameters is determined in terms of evaluation of the highest fitness function. A genetic algorithm (GA) is an ideal method to retrieve the maximum value of the fitness function as well as all optimum parameters. The five preliminary model parameters are input to the GA which gives rise to 50 chromosomes. Each chromosome is codified with a binary code of 50bits, and each parameter of the model corresponds to 10bits. On the basis of the above discussion, the selection of the GA is to find the highest fitness function. Then, the crossover and mutation of GA creates a new optimum of parameters on each generation. This GA proceeds until its 30th generation. Thus, improved parameters are obtained, and the simulated hysteresis loops become more close to experimental results. B-H characteristics at different frequencies 0.6 0.4 B [T] 0.2 0 500Hz 1kHz 2kHz 5kHz 10kHz 20kHz 40kHz PARAM ETERS EXTRA CTED FROM EXPERIM ENT Parameter Measured value Parameter Measured value 1995 56.87 3966 4167 259 0.393 9.53 0.063 TABLE II IDENTIFIED PARAM ETERS OF THE M ODIFIED JA M ODEL Identified parameter Value 342901 22 5e-7 0.75 18 Once the identified parameters are obtained, all hysteresis loops can be calculated and plotted using the Gear-Shichman method for numerical integration in APLAC. The hysteresis loops of measurement as well as the simulation results at 2kHz and 20kHz are respectively shown in Fig. 2. Comparison of B-H characteristics of TR 0.6 0.4 simulation at 2kHz simulation at 20kHz measurement at 2kHz measurement at 20kHz 0.2 0 -0.2 -0.4 -400 -0.2 -200 0 H [A/m] 200 400 Fig. 2 Co mparison between measurement and simulat ion of the transformer at different frequencies -0.4 -400 TABLE I B [T] IV. Results and discussion To verify the applicability of the modified JA model in the simulation of an inductor at large signal input, a transformer which is used in the intermediate circuit of a power supply is selected as test object. To obtain various hysteresis loops, the primary side of the transformer is saturated with an appropriate input current at different frequencies. At the mean time, the secondary side of the transformer is an open circuit during measurements. The measurement results of the primary side are summarized in Fig. 1. All traces of all hysteresis loops overlap once magnetic field reaches saturation. According to these results, it is possible to use one model to simulate the hysteresis loop at different frequencies. To identify the modified JA model, some needed parameters which are extracted from the experiment are presented in TABLE I. These are then utilized for the calculation of the preliminary key parameters of the hysteresis loop. After that, the five obtained parameters are optimized by using GA in Matlab. The identified parameters are finally determined and listed in TABLE II. -200 0 H [A/m] 200 400 Fig. 1 Hysteresis loop of the transformer at different frequencies Two hysteresis loops saturate at 2kHz, however, another two are not saturated at 20kHz. The former shape is called major loop, and the latter shape is referred as minor loop. To verify the validity of the model parameters, all of the simulated results should match with the corresponding measured hysteresis loops. The similarities of these loops can be inspected by using some specific parameters of the hysteresis loop. TABLE III shows the errors between measurement and simulation. TABLE III MODELING ERROR OF HYSTERESIS LOOP Verified Parameter Experiment 2kHz 20kHz 259.7 43.17 0.393 0.192 9.53 8.28 0.063 0.048 4167 4305 3966 4038 17.16 4.92 Simulation 2kHz 20kHz 251.5 41.34 0.401 0.178 9.26 8.63 0.058 0.051 3994 4236 3799 4236 16.17 5.41 Error 2kHz 20kHz 3.2% 4.2% 2% 7.1% 2.8% 4.1% 7.9% 5.9% 4.2% 1.2% 4.2% 4.7% 5.8% 9.1% All parameters shown here extracted from the corresponding hysteresis loops as given by Fig. 2. The upper six parameters describe the size and slope of each loop. They show good match for all loops. As a crucial parameter of criteria among others, the last parameter represents the relative area of the hysteresis loop, which reflects the summation of the dynamic change of the loop during magnetizing. All errors are below 10%, pointing to the validity of the simulated hysteresis loop. To sum up, a good agreement between measurement and simulation is obtained for both frequencies. This result indicates the effectiveness of this technique used to identify model parameters. In this case, the modified JA model can be used for the simulation of the magnetic hysteresis of a transformer at different frequencies. V. Conclusion The modified JA model as well as an identification technique based on an iterative method coupled with a genetic algorithm for parameter optimization has been described. The simulated hysteresis loops are compared to experimental results at different frequencies. The results of comparison indicate that the optimized identified parameters of the modified JA model are valid and accurate. They can be adopted to simulate the hysteresis loop of a transformer which is used in a switch mode power supply. Moreover, as hysteresis loops show in Fig. 2, this model can not only manage saturated loops at sinusoidal excitation, but also can describe unsaturated minor loops. This is beneficial to completely reflect the behavior of magnetic components within system at normal working conditions. ACKNOWLEDEGMENT This paper has been produced in collaboration with Kjellberg Finsterwalde GmbH, the most professional manufacturer of plasma cutting devices in Germany. I would therefore sincerely thank the Kjellberg colleagues who participated in the project “Störarme Stromquellen“ for support in every aspect. I especially thank Dr. Silvio Misera who provide the specific inductors and transformer for testing and carefully answer every question regarding these products. I am heartily thank my supervisor, Prof. Sven Bönisch, who gave me the opportunity to live such experience and guided me during the development phase as well as writing this publication. Last, I offer my regards to all of those who supported and encouraged me during the completion of the project especially Prof. Lutz Göhler and all the co-workers in EMC laboratory of Hochschule Lausitz. REFERENCES [1] D.C. Jiles and D.L. Atherton, Theory of Ferro magnetic Hysteresis, Journal of Magnetism and Magnetic Materials 61 (1986), 48-60 [2] Jarmo Virtanen, Toroid Model in APLAC, November 11, 1997 [3] David C. Jiles, Sen ior Member IEEE, J.B. Thoelke, and M.K. Dev ine, Nu merical Determination of Hysteresis parameters for the modeling of magnetic properties using the theory of ferromagnetic Hysteresis, IEEE Transactions on Magnetics, January 1992 [4] M. Hamimid , M. Feliachi,S. M.Mimoune, Modified Jiles -Atherton model and parameters identification using false position method , Physica B, January, 2010 [5] X. Wang, D.W.P. Tho mas, M. Su mmer, J. Paul, and S.H.L. Cabral, Characteristics of Jiles-Atherton model parameters and their application to transformer inrush current simu lation, IEEE, Trans actions on Magnetics VOL. 44 No. 3 March, 2008