Motor-CAD reduced node model tutorial (February 2015)

Transcription

1 Motor-CAD reduced node model tutorial (February 2015) Description Motor-CAD allows the machine performance, losses and temperatures to be calculated for a BPM machine. In this tutorial we will describe how to use the reduced node model routines built into Motor-CAD to produce small equivalent thermal networks. This tutorial is aimed as a guide for creating your own reduced thermal networks. Model Definition Motor-CAD has both electromagnetic and thermal models. These models can be used separately or as a coupled electromagnetic + thermal model. In this case we will start by using the thermal model by selecting the option shown below. Page 1

2 Machine Geometry The default machine geometry for a BPM machine is used in this tutorial as shown below: We will make some minor changes to the model as detailed below: This begins with setting a duty cycle transient calculation, with 10 cycles, and running from steady state temperatures: Page 2

3 Once we have done this, we set up a 10 period duty cycle as below: Not that the Duty Cycle Data Definition is set to Loss Speed, and the Duty Cycle pu or Value Input is set to Value. Furthermore we have the duty cycle itself set up with 10 periods. The values in the non-empty columns are transcribed below (note that the Ambient Temp [End] Column always remains at 40): Period Time Points Stator Copper Stator Back Iron Stator Tooth Speed [Start] Speed [End] Page 3

4 Due to the way the reduced model works, by calculating the steady state equivalent circuit, and the fact that we are fixing the resistances in the model before comparing the transient performance, it is important to use an appropriate loss model for the steady state model. Because of this, we transcribe the average duty cycle losses into the steady state loss model: We also disable speed dependent losses for this model, and then copy the losses into the input data losses. This should allow us to have an appropriately reduced steady state model, which should retain accuracy when further considering a transient calculation as well. We show this on the following page. Page 4

5 Changing of losses to match the average for the transient is shown below: Our first measure of the quality of our model will be comparison with steady state temperatures, so we begin by copying our full model temperatures to the validation table: Page 5

6 Reduced Node Model Next, to begin reduced node modelling we select reduced node model analysis from the tools menu bar, as shown below: Once we click the menu option the reduced node selection window appears: Page 6

7 Node selection In this tutorial we attempt to reduce the default model so that we can have a steady state equivalent thermal circuit, while keeping nodes we consider thermally important. This often entails keeping the nodes with losses, along with some others. In this case we keep the following nodes: 1 Housing [Active] 335 EWdg_F (Average) (C1) 9 Stator Back Iron 341 EWdg_R (Average) (C1) 11 Stator Surface 350 Wdg (Average) (C1) 13 Magnet 351 Stator Tooth (C1) 15 Rotor Back Iron 358 EWdg_F (Average) (C2) 26 End Space [F] 364 EWdg_R (Average) (C2) 27 End Space [R] 373 Wdg (Average) (C2) 65 Bearing [Front] 374 Stator Tooth (C2) 66 Bearing [Rear] As an explanation, we have kept the average winding nodes, as the temperatures of these nodes are important to us, and the average nodes are at the centre of the cuboids in the cuboidal model, and so should give us a good idea of the behaviour of the different sections of the Winding. Furthermore we have kept the Stator nodes, since because they contain a contain a considerable amount of the losses in the model; they can have a great effect on the temperature. For the same reason as the winding we have kept the Magnet and Rotor Back Iron nodes, as we are interested in the temperatures of these. Also, it may seem surprising to keep the End space and Bearing nodes, but we have decided to keep them for the reason that they serve somewhat as a buffer between the Ambient node and the Winding nodes. This is important because of the cuboidal model introducing negative resistances in its implementation. To copy the model used in this example, all that is needed is to simply ensure the only nodes checked are those given in the previous table. We also set our maximum resistance as This simplifies the model by removing resistance connections we do not want. We can set this value as low as we want if we want to simplify the model as much as possible. Page 7

8 Subsequently we run the model: Page 8

9 We get a circuit somewhat like the one shown below: Due to the way the model compensates for missing resistances by creating new ones, we can see that this model can appear a little random in its layout, but the nodes remain in their original position by default, with the display grid automatically compressing itself. However, if we check the validation graph now, we can see that there are only very small differences between this model and the full node model: Page 9

10 With the maximum absolute error of 0.045, we are happy with this result in comparison to the steady state result. Since this model could be used in a full system simulation, however, it is a good idea to check transient performance compared to the full model. We begin by saving our result: This allows us to use the result in a transient comparison. When we save or load a reduced node model, it is added as the transient model reduced node list. Page 10

11 This should allow us to quickly compare the reduced node model in the future if we desire it. We click the comparison button. While observing the calculations, we estimate a speed increase in calculation of around 10 times the calculation speed. We select the difference graph: If we look at the difference temperature graph, we have something like the following: Page 11

12 This seems like a lot of oscillation, but is in fact only +/-2 degrees, which is an almost neglible difference. Reduced Model: Full Model: Overlaying the two models it appears as though the reduced model is less susceptible to changes that the full model. We can change this, however. There is an option in the reduced model to skew the capacitance of Page 12

13 the model. This allows alterations to the transient performance of the model without affecting the steady state performance. We can use this to try and improve the model. Bear in mind that nodes will have increased capacitance as the model has attempted to reduce the model to an equivalent one. Testing the model with 0.55 times the capacitance yields an image like the one below: Here the model varies more, and because of this it fits more closely to the full model. Page 13

14 Matrix Export Once we are happy with our model, we may want to export the associated matrices in order to integrate them into a full system simulation. Once we choose the file location we get 3 files: The files correspond to:.cmf; capacitance..pmf; power (losses)..rmf; resistance. The capacitance and losses files will simply be two column text files listing the nodes and their corresponding values. Part of the resistance matrix is shown below: Page 14

15 We can now manipulate and subsequently transfer these text files into a system simulation program of our choice, such as Matlab/Simulink. Conclusion We can see that using reduced node modelling is a powerful tool for modelling equivalent steady state models, and can also give us accurate results in the transient case also. Furthermore the ability to export the model for use in other programs after sound model order reduction is also extremely powerful, and can enable accurate reduction of full system simulation models while retaining good accuracy. Page 15