
The bear concerned with wound components at BI has developed a smoother interpolation method for core loss data than our prior method – Carlsberg don’t make core loss maps but, if they did, they’d be the smoothest core loss maps in the world… We gave the bear some honey; not too much though, just a little and we didn’t give him any Carlsberg, it makes him snoozy. We’ve promised him a large salmon if he can incorporate the effects of temperature to form an interpolation represented by a 3D hypersurface.
The surface shown in the figure is a representation of some data which takes the form of a function, P(B,f), where P is the core loss density, B is the flux density and f is the frequency. Such an implementation is desirable when simulating power converters that are not readily amenable to purely analytical analysis. These include for example, resonant converters in which turn off currents through or turn on voltages across switches can only be accurately obtained by solving the time domain differential equations for the converter, see for example [1].
When simulating power converters working directly with the differential equations of the converter, rather than virtually wiring up circuits, can lead to greatly reduced simulation time / computational requirements compared to explicitly modeling the switching behaviour. Some simplifications usually have to be accepted, however. This premise is the basis of simulators such as PLECS and PSIM compared with SPICE and Simscape/Simulink – although the modelling detail in both SPICE and Simulink can be adjusted with effort and some knowledge of how the models and solvers work. Reduced simulation times are generally desirable when developing efficiency maps which require many runs – just ask any of the bears that design electrical machines!
One of the disadvantages of working with the differential equations of a converter is that the dead-time is often ignored. Another is that the switch is assumed to complete the switching action instantly. This is troublesome because the power lost in the switch is, mostly, lost during the switching event – unless the frequency is very low and conduction losses dominate. Systems like PLECS tend to ascribe an analytically calculated switching loss to each switching event. The same approach can be taken in Matlab/Simulink or Python/OpenModelica with a little effort.
The analytical loss calculations for the converter can be performed using a script called from the Matlab workspace (or called using Python). The script calls Simulink (or OpenModelica) and passes the required parameters for the particular simulation run. The differential equations are solved and a set of data is passed back into the Matlab Workspace (or to another Python Script – one opens and reads a CSV file which contains the output data from Modelica). Having obtained the resonant tank currents and voltages, which are difficult to calculate analytically, the losses of the various components in the converter are computed using closed form analytical equations (e.g. Ohm’s Law) which is fast.
For the copper loss of a wound component only the RMS current and the winding resistance as a function of temperature are required assuming there is no significant AC resistive losses above the DC value. If there are significant AC resistive loss there are a couple of ways to proceed 1) do some off line FEA and make a winding resistance vs frequency vs temperature graph to be read by the efficiency mapping simulation or 2) use Dowell equation or its derivatives or 3) Use harmonic analysis and treat each frequency of the converter waveforms as seeing a different value of resistance. Anyway… for the simple case of frequency independent winding resistance we only need to worry about the temperature of the winding to obtain its resistance, and the RMS of the current flowing through the component over one steady state cycle of the converter operation, which we take from the (P)DE solver (Simulink / Modelica).
For the core loss we have numerous options as well. The fact that there are a number of ways of attacking the problem is indicative that none of them are particularly ‘good’. If there was one sure-fire method… all the engineering bears (and all the engineers who are not bears) would be using it.
First option, do some off-line FEA to establish the core loss in advance of the main efficiency mapping simulation and make a loss map. Problems are that you’re dependent on the FEA tool’s model (Jiles-Atherton Model, Preisach Model, Separation methods (Altair Hyperworks / FLUX), Imaginary Permeability, Steinmetz Equation or its derivatives (ANSYS Maxwell) – if you can find a tool that uses i2GSE [2] you’re probably on a winner as it’s a beary good model even when the duty ratio is at the extremes. One major difficulty is that extra pre-simulation activities make more work and if you decide to change the design of the wound component it means running this activity again. The other problem with all the sexy models is that one needs some good data to input to them to find all the coefficients and this data needs collecting; often using taxing experiments. Alternatively one can rely on the data supplied by the FEA tool vendor e.g. ANSYS has a good selection of MnZn Ferrites but, we have little knowledge of how the data that drives the models was acquired and the limits of its applicability. In the particular case of Maxwell iGSE vs i2GSE at low and high duty is an ongoing debate as is the effect on AC induced losses when there is a DC flux biasing the core.
Second option, use the manufacturers’ data fitted to some equations (e.g. Steinmetz). One problem is that it’s hard to get a good fit over a wide range of several parameters. Another more general problem is that although we know the provenance of the data we still don’t know how it was acquired and the limits of its applicability are also somewhat open to debate. The bears used to like to fit equations but, often it only works well for one parameter e.g. loss as a function of flux density which temperature and frequency held constant. If you’ve got your snout in a resonant converter then having the ability to calculate at different frequencies easily is a big deal because the gain of the system is (often but, not always) controlled by the frequency. Some companies, like Micrometals, provide the core loss equations on the datasheet but, we have no knowledge of the accuracy of the equations over any particular range of flux density or frequency – we have to take it on trust. Bears are notoriously beady eyed, suspicious, creatures.
Third option, use the manufacturers’ data interpolated and posed as a function in Matlab/Python/Octave. All the difficulties around how the data was obtained and the limits of applicability still stand but, now we know that, if the interpolation algorithm is good then the interpolated data will fit the measured data we have, everywhere. If we don’t stray too far outside the region over which the data was supplied then we’re probably reasonably safe. In recent times this has been the preferred method of the bears. It’s simple, easy to implement and effective. No hunting for endless coefficients, no experiments and we know at least some of the limitations of the data and can qualitatively contemplate the effect that inaccuracies in the data might have on our simulations. The main inaccuracy other than the general loss data measurement methods that the manufacturer has used is that the waveform shape is often square or a portion of a sinusoid rather than a nearly perfect sinusoid which was probably used to gather the data.
The estimation of the efficiency map of a power electronic converter as part of the system engineering design is always subject to some errors, the objective is to obtain a yes/no on if the system loss budget will be missed and to provide some benchmark data to validate the measured system efficiency against when all the hardware, basic software and application software have been worked though. Another job of the efficiency map is to allow the end customer to run performance cycles in their own simulations to estimate energy use/loss over a particular activity for example some possible activities could include: a robot arm in a production line making a series of moves repeatedly, a set of flight surfaces making a series of moves intermittently, an electronic gear select system actuating intermittently in a vehicle powertrain or an entire electric vehicle drive cycle. The end customer will have these drive cycles in mind and is most likely to benchmark several competing suppliers converters/inverters based on the efficiency maps the supplier provides. The end customer will usually do this before nominating a supplier. Having a quick and defensible way to develop efficiency map under your fez – bears sometimes wear a fez… and a cummerbund too – is highly desirable for the modern system engineering bear working in power electronics.
A good efficiency map is fairly flat right to the edges of the permissible operating area – not always easy to obtain though while also keeping cost, weight and volume in mind!
References
[1] Z. U. Zahid et al., “Modeling and Control of Series–Series Compensated Inductive Power Transfer System,” in IEEE Journal of Emerging and Selected Topics in Power Electronics, vol. 3, no. 1, pp. 111-123, March 2015, doi: 10.1109/JESTPE.2014.2327959.
[2]J. Muhlethaler, J. Biela, J. W. Kolar and A. Ecklebe, “Improved Core-Loss Calculation for Magnetic Components Employed in Power Electronic Systems,” in IEEE Transactions on Power Electronics, vol. 27, no. 2, pp. 964-973, Feb. 2012, doi: 10.1109/TPEL.2011.2162252.
