Object-Oriented Fault Modeling and Simulation of DCDC Converter for Electric Vehicle

Object-oriented fault modeling and simulation of DCDC converter for electric vehicle

Zhong Zai-min, Su Heng, Xiong Yun

College of Automotive, Tongji University, Shanghai 201805, China

E-mail of Corresponding Author: xiongyun@fcv-sh.com Telephone: 021-69589230

Abstract—This paper introduces an object-oriented fault modeling and simulation method, which is used to realize the diagnosis of a type of DCDC converter used for EV (electric vehicle). First of all, the necessity of the development of EV and the importance of the DCDC converter used for EV are emphasized. Then this paper describes the background of object- oriented Language and fault diagnosis based on pattern recognition. According to several typical faults of DCDC converter, corresponding fault models are generated based on (object-oriented) technology; the veracity and reliability of these models are proved by simulation and experiment.

Keywords-object-oriented; electric vehicle; DCDC;modeling;

simulation

I. I NTRODUCTION

Considering that the total energy on the earth is limited and automobile is a main consumer of energy, the development of electric vehicle is inevitable. And as an important part of electric vehicle, DCDC converter is always a focus of research. For the reason that the working condition of DCDC converter is usually acuity, the requirement of its reliability is very high. Then how to improve a DCDC converter’s reliability is an urgent problem for most EV researchers now.

The on-board power battery charger can convert an AC power from the public power grid to DC power for charging power battery of EV when it is parking in garage. As a typical DCDC converter used for EV or HEV (hybrid electric vehicle), it is selected as the object of fault diagnosis in this paper [1].

II. Background of Object-oriented Language and Fault Diagnosis Based on Pattern Recognition

A. Object-oriented Language

Modelica is a language for modeling physical systems of multiple domains, such as electrical, mechanics and so on. It is designed to support effective library development and model exchange [2]. Modelica has following three characteristics, which are also advantages:

1) Modelica is primarily based on equations instead of assignment statements. This permits noncausal modeling that

gives better reuse of classes since equations do not specify a certain data flow direction.

2) Modelica has multi-domain modeling capability.

Components from several different domains such as electrical, mechanical, and control applications can be described and connected.

3) Modelica is an object-oriented language with a general class concept that unifies classes. This facilitates reuse of components and evolution of models.

B. Diagnosis based on Pattern Recognition

The diagnosis based on pattern recognition requires us firstly several typical operating modes of a DCDC are defined, including a normal operating mode and several typical fault operating modes [3-5]. After that, several models should be established and each model represents one typical mode. One feature vector should be derived from each of these models.

The feature vector should be carefully selected to well represent the model’s operating mode. To judge a real DCDC’s operating condition, the only thing we need is to collect a feature vector from it and evaluate the gaps between the vector and all the typical modes’ feature vectors. The minimal gap indicates the real DCDC’s operating condition.

In most cases, feature vector can also be regarded as a point in a high-dimension space. In this case, it is difficult to evaluate the gaps between feature vectors. As a result a K-L Transform is necessary to reduce the feature vectors’

dimension without losing feature vectors’ principal information, which is significant contributing to diagnosis.

The process of the diagnosis is shown in Fig. 1.

Fig. 1: Process of Diagnosis Based on Pattern Recognition

Fig. 2: The topological structure of the Charger

The ultimate aim of fault modeling and simulation in this paper is to realize diagnosis to DCDC convertor. After analyzing this diagnosis method, the modeling method based on object-oriented language suits the diagnosis well.

C. Object of the Diagnosis

The DCDC converter, which is also named battery charger, is the object of this diagnosis. In this DCDC converter, AC power is rectified by a rectifier bridge paralleled with a capacitor of 4mF. Through this rectifier bridge, the AC power is rectified approximately to be a constant DC power. The four MOSFETs shown in Fig. 2 are controlled by four PWM waves in a frequency of 100 KHz passed from a DSP. So the voltage transmitted through the transformer is an oscillating voltage in a frequency of 100 KHz. Another rectifier bridge together with an inductor and a capacitor smooth the oscillating voltage to a constant one, which is used to charge the power battery on EV. The topological structure is depicted in Fig. 2.

There is no parasitic components shown in Fig. 2, however, they are well structured in the models shown below.

III. M ODELING , S IMULATION AND E XPERIMENT

A. Normal DCDC Converter Modeling

Dymola is commercial software for object-oriented modeling and simulation; it is derived from the object-oriented language modelica. Besides modelica language’s advantages, it also provides a convenient graphic modeling environment and totally open modelica code of all components.

The DCDC converter model, which is established in Dymola, operating in the normal mode is shown in Fig.3, all components in this model is modeled to simulate the corresponding components’ behavior correctly and precisely.

Also the four MOSFETs are important components in this model and its behavior is determined by its parasite capacitor.

The parasite capacitor is shown in Fig. 4 and MOSFET’s Dymola model is shown in Fig. 5.

Fig. 3: Normal Model of a DCDC converter

Fig. 4: MOSFET’s Parasite Capacitor

Fig. 5: MOSFET’s Dymola Model with Parasite Capacitors Fig. 3 shows how to structure a normal model for a DCDC converter in Dymola.

Fig. 4 shows that there are three parasite capacitors containing in one MOSFET and they are generated in the producing procedure. So although the parasite capacitors can deteriorate the MOSFET’s behavior, we can’t eliminate them from the MOSFET. And because they will bring great influence to the MOSFET’s behavior and can’t be ignored, these parasite capacitors should be carefully evaluated when the MOSFET model is built. In Fig. 5, the MOSFET’s Dymola model is shown.

Fig. 6 shows how to write a nonideal transformer in Dymola’s edit window. In Dymola’s edit window, modelica language is permitted to input and the component’s behavior is defined through the equations, which can reflect the real component’ behavior perfectly.

Fig. 6: Establishment of a Transformer in Modelica Language in Dymola

Fig. 7: The Current in Time Domain on the inductor derived from Normal DCDC Model

B. Simulation

After the DCDC converter is modeled completely in Dymola, the simulation can be started.

By simulating the Normal DCDC Model in Dymola, a series of signals are obtained. Select the inductor’s current signal as the original signal, which is possible to create feature vectors in future diagnosis. The current wave in time domain is shown in Fig. 7.

In consideration of such a fact that, even in a normal real DCDC, when it operates in different conditions, such as different output power, the current wave derived from the same inductor will be very different, so we have to convert the time domain signal to frequency domain. Matlab provides a powerful and convenient FFT analysis tool for us to convert the time domain signal to frequency domain. So the data we obtained from the Dymola DCDC normal model has to be input in Matlab to complete its FFT analysis. The Fourier spectrum of the inductor’s current derived from Matlab FFT tools is shown in Fig. 8.

The figure above in Fig. 8 indicates that the signals in frequency of integral times of 200 KHz and a constant offset are principal significant contributing ingredients of the current signal. Similarly, the figure below shows that, in a frequency area below 2000 Hz, the signals in frequency of integral times of 100Hz can’t be ignored, although the power of these signals are not so great as the signals in frequency of 0 Hz and times of 200 KHz.

Fig.8: Using Matlab FFT Analysis Tool to Obtain Current’s Fourier Spectrum The signals of integral times of 200 KHz result from switching frequency, which is exactly 100 KHz. The rectifier bridge after the transformer doubled the transformer’s output to a wave of 200 KHz, so the current on the inductor has such harmonic wave signals of integral times of 200 KHz.

After preliminary analysis, the signals in frequency of integral times of 100 Hz may result from the rectifier bridge before the transformer. The input of the rectifier bridge is a 50Hz AC power and the output of it has a slight wave of 100Hz. This analysis to the origin of this 100Hz power is strengthened in latter detailed analysis.

The voltage across the capacitor paralleled with the rectifier bridge is shown in Fig. 9. The voltage applied across the capacitor has a distinct wave and the period of this wave is show in Fig. 8. The period of the wave is 0.01 second. Such a period time indicates that the frequency of the wave is 100Hz.

The wave is generated in the process of rectifying the AC power. For the whole system’s input is a sine wave AC power, which has a frequency of 50Hz, every time the AC voltage drops to near zero (at this time, the AC power can only provide little energy, usually not enough to maintain the output power), the capacitor has to provide extra energy to offset the energy gap. The value of this capacitor is 3mF in the DCDC converter.

For the energy stored in capacitor has a direct relation with its voltage, shown below, the capacitor’s energy output will lead a voltage drop.

1 ²

E = 2 CU (1)

Fig. 9: Voltage across the Capacitor

C. Fault Modeling and Simulation

In order to further demonstrate the correctness of the analysis above, it is necessarily to establish a fault model. In order to get a distinct result from this fault model, the value of the corresponding capacitor in it is wrongly set to 100 uF (in normal model the value of it is 3mF).

The fault model is shown in Fig. 10. The capacitor in the figure marked by the red circle is the wrongly set capacitor.

After setting the capacitor’s value to 100uF, energy stored in it will greatly decrease when applied the same voltage across it.

According to the analysis above, this energy decline leads a much more violent wave of the output voltage and the 100Hz power’s proportion will increase greatly in the current Fourier spectrum, shown in Fig. 12 and Fig. 13.

Fig. 13 shows detailed Fourier spectrum in a lower frequency area, 0-2000Hz. In this figure, it is clearly that the proportion of 100Hz’s power increases greatly comparing with normal model. Such a change matches the analysis above.

Fig. 10: Fault Model

With this fault model, it is easy to find the feature of this fault mode. Although in most cases, the fault feature may not be so easily obtained only by observing time domain data or doing FFT analysis, however it is possible to get these fault

feature through more complicated method, for instance, the method introduced in the second section of this paper.

Fig. 11: Inductor’s Current of Fault Model in Time

Fig. 12: Inductor Current’s Fourier Spectrum on Large Frequency Scale

Fig. 13: Detailed Fourier Spectrum in Lower Frequency Area

IV. C ONCLUSION

A newly developed modeling and simulation method is discussed in this paper. A normal model and a typical fault model of a DCDC converter based on the object-oriented language modelica are introduced and how to model such models in Dymola is presented.

The simulated current signals through an inductor in such a DCDC converter, which derived from the normal model and the fault model, are compared and analyzed both in time and frequency domain. In the future, the measured current through the corresponding inductor will be derived from a real DCDC converter. The current signal in time and frequency domain will be measured and analyzed. The signals derived from Dymola models will be compared with measured signals to better prove the models’ accuracy and reliability. Additionally, the simulation and measurement data will be applied to pattern

recognition based fault detection method in order to evaluate such a real DCDC converter’s operating condition.

R ^EFERENCES

[1] J. de Kleer and B. Williams. Diagnosis with behavioral modes. In Proceedings of the IJCAI’89, pages 1324–1330, 1989.

[2] C. Kral, A. Haumer, and F. Pirker, “A modelica library for the simulation of electrical assymetries in multiphase machines - the extended machines library,” IEEE International Symposium on Diagnostics for Electric Machines, Power Electronics and Drives, The 6th, SDEMPED 2007, Cracow, Poland, 2007.

[3] F. Filippetti, G. Franceschini, C. Tassoni, and P. Vas, “Impact of speed riple on rotor fault diagnosis of induction machines,” Proceedings of the International Conference on Electrical Machines, ICEM, pp. 452–457, 1996.

[4] S. Legowski, A. S. Ula, and A. Trzynadlowski, “Instantaneous power as a medium for the signature analysis of induction motors,” IEEE Transactions on Industry Applications, vol. 32, no. 4, pp. 904–909, 1996.

[5] R. Walliser, “The influence of interbar currents on the detection of

broken rotor bars,” ICEM, pp. 1246–1250, 1992.