Detection of IGBT degradation in NPC inverter based on infrared thermography

Abstract

Multilevel inverters have been widely applied in high-voltage and high-power applications. Due to the increasing number of the switching devices in multilevel inverter, the probability of fault is increasing simultaneously. The development of reliable systems for fault detection that enable to diagnose a wide range of faults is a motivation of many researchers. In this paper, the theory of IGBT power loss in NPC inverter is introduced and a detecting method of IGBT degradation for diode neutral-point-clamped (NPC) multilevel inverters based on infrared thermography is proposed. Firstly, infrared images of NPC inverter are acquired based on the software FloTHERM and background of the image removed through Otsu’s statistical threshold selection algorithm. Then, the RGB image is turned into a gray image and segmented into small ones. Lastly, hotspots of the image are extracted and an improved detecting method is proposed to diagnose whether the IGBT is in degradation or normal condition. The simulation results prove the feasibility of the proposed method and its advantages in good detecting accuracy.

Keywords

IGBT degradation NPC inverter power loss detection infrared thermography image processing FloTHERM

1. Introduction

As the power electronic technology developed toward high-voltage and high-power applications, diode neutral-point-clamped (NPC) multilevel inverters have been studied extensively. The multilevel inverter could achieve more levels, lower harmonic distortion in the voltage output in addition to lowering the voltage stress of the power devices, as compared with conventional two-level inverters [1, 2, 3, 4]. However, the NPC inverter system is composed of many switching devices, which increases the probability of fault, as a break in any one of these devices will inevitably cause the entire inverter to break down or work under abnormal situation [5]. Therefore the fault diagnosis methods would be necessary to ensure the reliability of the multilevel inverter.

For the reason of high-power, devices of multilevel inverters such as IGBTs and diodes always work under high power loss and high operating temperature condition. Higher operating temperature due to increased power loss and density not only influence and worsen electrical parameters of devices, but also force undesirable physical and chemical processes in component materials and constructions. These factors may finally lead to thermal breakdown, reducing the reliability and even leading to devices and circuits faults [6]. To predict thermal characteristics of semiconductor devices and to provide safe thermal regions of multilevel inverters, temperature detection of circuits or devices is a very important work.

Infrared thermography (IRT) is a very useful method for detecting the conditions of electrical equipment. It can detect infrared energies and convert them into electrical signals by monitoring, recording, and analyzing the subsurface defects characterization [7]. The defective electrical equipment will display abnormal thermal distribution on its surface. Thus the condition of electrical equipment can be estimated depending on the thermal image [8].

Many researchers study the temperature distribution of equipment for detection of element defects. Jung and Lyou [9] proposed a diagnostic system for power distribution lines based on a thermal imaging camera and a color camera. Also aiming at the temperature rising fault, a method based on infrared image for power transmission line detection and fault diagnosis was proposed [10]. Paper [11] concerned with implementation of self-organizing-map (SOM) for intelligent machine fault diagnostics. Infrared images were acquired by thermography camera as data base of machine diagnostics system and feature extraction of images was conducted by calculating area, perimeter and central moment of region of interest (ROI). At last, pattern recognition technique was employed to diagnose the machine conditions by mapping the image features based on SOM. A PCB circuit is another common research object by using IRT method. Contact data of voltage and non-contact data of temperature were acquired from a PCB circuit [12]. Faults of the circuit were diagnosed by contact and non-contact method separately, and then the two results were fused into a final one. In addition, IRT method is widely used in other equipment, such as power capacitors [13], rolling bearing [14, 15], arresters of substation [16], circuit breaker [17] and so on. However, diagnostic methods based on infrared thermography for power electronic converters have not been as thoroughly investigated as for the research objects mentioned above and no published work in this area has been found so far.

This paper will focus on the operation condition of IGBTs in NPC multilevel inverter, infrared image of the inverter will be derived and analyzed to detect whether the devices in degradation or normal condition. The paper is organized as follows: In Section 2, theory of IGBT power loss in NPC inverter is provided. Acquisition and processing of IRT image is given in Section 3. Detection results of the method proposed are shown in Section 4 followed by conclusions in Section 5.

2. Theory of IGBT power loss in NPC inverter

2.1 Brief description of NPC inverter

Figure 1 shows a main circuit of three-level NPC inverter with the load of a motor. The input voltage of the inverter is idealized that the voltages of two capacitors are equal. There are four switching devices S ${}_{x1}$ $\sim$ S ${}_{x4}$ ( $x=a,b,c$ ) connected in series and two clamping diodes D ${}_{x5}$ $\sim$ D ${}_{x6}$ in each phase. Take phase-a as an example, the switching groups (S ${}_{a1}$ , S ${}_{a3}$ ) and (S ${}_{a2}$ , S ${}_{a4}$ ) are operation in a complementary state. That is, when devices in one of the groups are turned on, the devices of the other group are turned off, and vice versa. Here, it is defined that the voltage between points “ $x$ ” ( $x=a,b,c$ ) and “ $o$ ” $U_{xo}$ is named as bridge voltage. The switching command and output bridge voltages of this inverter are listed in Table 1. There are three kinds of switching states: “ $P$ ”, “ $O$ ”, and “ $N$ ” in each phase [18].

Table 1
Switching command and bridge voltage of the NPC inverter

Switching state	Switching command ( $x=a,b,c$ )				Bridge voltage
	S ${}_{x1}$	S ${}_{x2}$	S ${}_{x3}$	S ${}_{x4}$	U ${}_{xo}$
P	On	On	Off	Off	$+$ U ${}_{d}$ /2
O	Off	On	On	Off	0
N	Off	Off	On	On	$-$ U ${}_{d}$ /2

Figure 1.

Main circuit of a three-level NPC inverter.

2.2 Calculation of IGBT power loss in NPC inverter

The power losses of IGBT P ${}_{\textit{IGBT}}$ contain conduction loss P ${}_{\textit{con}}$ and switching loss P ${}_{\textit{SW}}$ as shown in Eq. (1) [19, 20].

$\displaystyle P_{\textit{IGBT}}=P_{\textit{con}}+P_{sw}$ (1)

The conduction loss of IGBT S ${}_{x1}$ and S ${}_{x4}$ in Fig. 1 can be expressed as follows:

$\displaystyle P_{\textit{con},T1}=m\times\left\{\frac{u_{oT}\times\hat{{I}}_{C% }}{4\pi}\times[(\pi-\varphi)\times\cos\varphi+\sin\varphi]+\right.\left.\frac{% r_{oT}\times\hat{{I}}_{C}^{2}}{6\pi}\times(1+2\cos\varphi+\cos^{2}\varphi)\right\}$ (2)

where, $m$ is modulation, $\cos\varphi$ is power factor, $\hat{{I}}_{C}$ is collector current of IGBT, $u_{oT}$ and $r_{oT}$ are device specific constants. The conduction loss of IGBT S ${}_{x2}$ and S ${}_{x3}$ in Fig. 1 can be given as:

$\displaystyle P_{\textit{con},T2}=\left(\frac{u_{oT}\times\hat{{I}}_{C}}{\pi}+% \frac{r_{oT}\times\hat{{I}}_{C}^{2}}{4}\right)+m\times\left(\frac{u_{oT}\times% \hat{{I}}_{C}}{4\pi}\times[\varphi\times\cos\varphi-\sin\varphi]-\frac{r_{oT}% \times\hat{{I}}_{C}^{2}}{6\pi}\right.\left.\times[1-2\cos\varphi+\cos^{2}% \varphi]\right)$ (3)

The switching loss of IGBT S ${}_{x1}$ and S ${}_{x4}$ in Fig. 1 is shown in Eq. (4):

$\displaystyle P_{sw,T1}=f_{c}\times\frac{U_{c}}{U_{\textit{base}}}\left[\frac{% B_{oT}}{2\pi}\times\hat{{I}}_{C}\times(1+\cos\varphi)+\frac{C_{oT}}{4\pi}% \times\hat{{I}}_{C}^{2}\times(\cos\varphi\times\sin\varphi+\pi-\varphi)\right]$ (4)

where, $f_{c}$ is switching frequency. The switching energy is measured at a specific DC voltage $U_{\textit{base}}$ . For a given application, it is linearly scaled with the blocking voltage $U_{c}$ . $B_{oT}$ and $C_{oT}$ are device specific constants. The switching loss of IGBT S ${}_{x2}$ and S ${}_{x3}$ in Fig. 1 is shown in Eq. (5):

$\displaystyle P_{sw,T2}=f_{c}\times\frac{U_{c}}{U_{\textit{base}}}\left[\frac{% B_{oT}}{2\pi}\times\hat{{I}}_{C}\times(1-\cos\varphi)+\frac{C_{oT}}{4\pi}% \times\hat{{I}}_{C}^{2}\times(\varphi-\cos\varphi\times\sin\varphi)\right]$ (5)

It can be seen from Eqs (2) $\sim$ (5) that the power loss of outer IGBTs (S ${}_{x1}$ and S ${}_{x4}$ ) is different from the inner IGBTs (S ${}_{x2}$ and S ${}_{x3}$ ). It can be explained by Table 1 because the switching commands of these IGBTs are not the same.

Figure 2.

Detection steps of the method proposed.

Several studies are focused on criterion of IGBT failure under thermal degradation caused by power loss. The failure criterion is defined differently from each other. In [21], the maximum peak of the collector-emitter ringing at the turn off transient was identified as the degradation variable. In [22], the switching turn-off time was recognized as failure precursor. Collector-emitter on-state voltage was usually identified as a health indicator [23, 24]. When its reliability reduces to a certain level, the IGBT is considered to be failed. The experimental results of paper [23] showed that the collector-emitter on-state voltage significantly increased due to the accelerated aging. This on-voltage is 0.953 V for a new IGBT, and it increased to 0.975 V, 1.1 V and 1.5 V under the duration of aging was 60, 120 and 180 minutes respectively. In addition, the equivalent on-resistance of IGBT is 35.01 m $\Omega$ for a new one, and this resistance increased to 36.79 m $\Omega$ , 41.71 m $\Omega$ and 60.61 m $\Omega$ under the duration of aging was 60, 120 and 180 minutes respectively.

According to the equations for power loss calculation of IGBT and the experimental results of paper [23], the power loss of an IGBT before aging and after aging can be derived.

2.3 Acquisition and processing of IRT image

FloTHERM is 3D software which provides component model of thermal design for electronic equipment. It provides thermal simulation and thus makes possible fault diagnosis of components at the different levels such as IC elements, devices, circuits and so on. The processing of thermal simulation involves construction of equipment, adding the model property, network segmentation, solving and post-processing [19].

After acquisition of the IRT image, the processing of the image in this paper contains four major steps as Fig. 2 illustrates. The first step is to separate objects from background, and then the image is segmented, followed by the hotspots extraction, and the last one is the diagnosis step.

2.4 Separate objects from background

Parameters of thermal simulation of NPC inverter are refer to paper [19], where the simulation result was very close to the real temperature. In the software, the power loss of each IGBT must be determined in the simulation model, and these data are calculated based on the above equations from Eqs (1) $\sim$ (5). The IRT image of NPC derived based on FloTHERM is shown in Fig. 3.

Discrete components are used in simulation and the arrangement of the components refers to Fig. 1 in order to facilitate comparison. It can be easily seen that a clear distinction between the main object and the background of the picture as shown in Fig. 3. Based on the principle of Otsu’s statistical threshold selection algorithm using gray-level histograms [25], the main objects can be separated from the background.

Figure 3.

A simulation infrared thermography.

Figure 4.

Flow of background separation.

The flow of background separation is shown in Fig. 4. As can be seen from this figure that background of the original thermal image is almost in blue color, so the blue component of the picture Fig. 4a is extracted and the main objects are separated based on Otsu’s threshold, as shown in Fig. 4b. The formula of this algorithm is shown as [13]:

$\displaystyle\beta(x,y)=\begin{cases}1&\text{if }\alpha(x,y)>T\\ 0&\text{if }\alpha(x,y)\leqslant T\\ \end{cases}$ (6)

where, $\alpha(x,y)$ represents the original image, $\beta(x,y)$ represents the binary image of the main objects extracted from $\alpha(x,y)$ , and T is the threshold value calculated by Otsu’s method.

Then, multiply the Fig. 4a with inverse value of Fig. 4b, and the result is shown in Fig. 4c. At last, turn the black background of Fig. 4c into white, as shown in Fig. 4d. Equation (7) illustrates the principle of these two steps [13].

$\displaystyle\gamma(x,y)=\begin{cases}\alpha(x,y)&\text{if }\beta(x,y)=1\\ 0&\text{if }\beta(x,y)=0\\ \end{cases}$ (7)

where, $\gamma(x,y)$ is the thermal image of the NPC inverter without background as shown in Fig. 4d.

Figure 5.

RGB and gray thermal images without background.

Figure 6.

Segmentation of thermal image.

Figure 7.

Hotspots extraction of images in Fig. 6.

In addition, the green component of the picture Fig. 4d can be removed too. Based on the Eqs (6) and 7, the green background can be removed with the same principle. The derived thermal image is proposed in Fig. 5a and this RGB image is converted to gray image based on Eq. (8) according to the color sensitivity of human eye [10], as illustrated in Fig. 5b.

$\displaystyle\textit{Gray}=0.2990\times R+0.5870\times G+0.1140\times B$ (8)

2.5 Image segmentation and hotspots extraction

In Fig. 5b, every part of the main object (IGBTs in this paper) is separated from each other. Hence, it is possible to segment the image into small one just include one IGBT. Take the IGBT S ${}_{a1}$ and S ${}_{a2}$ as example, Fig. 6 shows the result of image segmentation.

In order to extract the hotspots, the maximum pixel value of the image Fig. 6 should be determined as shown in Eq. (9).

$\displaystyle T_{\textit{hot}}=\max(\delta(x,y))$ (9)

where, $\delta(x,y)$ presents the image of Fig. 6a or b, T ${}_{\textit{hot}}$ is the hotpots of the image. In a specific image, the number of hotpots of the image may be one or more. Searching hotpots based on the Eq. (9), there are one and five hotspots is found in Fig. 6a and b respectively, as shown in Fig. 7 using the symbol of “ $+$ ”.

3. Detection results of the proposed method

3.1 Proposed detecting method

The temperature information of the hotspots derived above is analyzed to detect the operation condition of IGBTs in NPC inverter. A method based on the threshold of absolute temperature and relative temperature is proposed in this paper. The specific procedure is as follows:

1) 1)
Because thermal images are represented by gray scale which values range is [0, 255], these values are not the real temperature. To obtain real temperature values, the maximum temperature T ${}_{\max}$ and minimum temperature T ${}_{\min}$ in infrared image are needed to be achieved. Then, the real temperature can be calculated by the equation presented below [13]:

$\displaystyle T_{\textit{real}}=T_{\min}+\frac{T_{\textit{gray}}}{255}\times(T% _{\max}-T_{\min})$ (10)

where, T ${}_{\textit{real}}$ represents the real temperature, and T ${}_{\textit{gray}}$ is the gray value of hotspots.
2)
A method includes absolute temperature and relative temperature adopted to detect the operation condition of objects was presented in paper [13]. Absolute temperature represents the real temperature values of objects. Since the accuracy of the real temperature values is often affected by environmental factors, detection method only include this value will reach an inaccurate result. Relative temperature is not a real temperature value of a hotspot. To calculate the relative temperature, a reference temperature of hotspot with similar conditions should be measured, as shown in Eqs (11) and (12).

$\displaystyle\Delta T=T_{\textit{hot}}-T_{\textit{ref}}$ (11)

where, T ${}_{\textit{hot}}$ and T ${}_{\textit{ref}}$ represent the real temperature of a detecting hotspot and a reference hotspot respectively. Both of the values can be derived based on Eq. (10).

$\displaystyle\Delta T_{r}=\frac{\Delta T}{T_{\textit{ref}}}\times 100\%$ (12)

In this paper, three normalized values of $\Delta T$ , $\Delta T_{r}$ and $T_{\textit{hot}}$ are adopted as parameters to detect the condition of IGBT, shown as follow:

$\displaystyle F=f_{1}\times\Delta T_{\textit{norm}}+f_{2}\times\Delta T_{% \textit{rnorm}}+f_{3}\times T_{\textit{hotnorm}}$ (13)

where, $F$ is the result of detection, $f_{1}$ , $f_{2}$ and $f_{3}$ are three constant coefficients, $\Delta T_{\textit{norm}}$ , $\Delta T_{\textit{rnorm}}$ and $T_{\textit{hotnorm}}$ are three normalized value of $\Delta T$ , $\Delta T_{r}$ and $T_{\textit{hot}}$ respectively. If the value of $F$ exceeds a threshold, it means that the IGBT is under abnormal condition (degradation in other word).

Table 2
Temperature of IGBTs in normal or degradation condition

IGBT Temperature [ ${{}^{\circ}}$ C]

Normal S ${}_{a1}$ in degradation S ${}_{a2}$ in degradation

S ${}_{a1}$ 67.4 70.2 67.4

S ${}_{a2}$ 90.9 91.2 96.3

S ${}_{a3}$ 91.8 91.9 91.6

S ${}_{a4}$ 66.6 66.7 66.7

Figure 8.
Temperature curves of detecting IGBTs.

3.2 Simulation results

IGBT	Temperature [ ${{}^{\circ}}$ C]
S ${}_{a1}$	67.4	70.2	67.4
S ${}_{a2}$	90.9	91.2	96.3
S ${}_{a3}$	91.8	91.9	91.6
S ${}_{a4}$	66.6	66.7	66.7

After building the model of NPC inverter in FloTHERM, a curve of temperature versus iteration can be derived, as shown in Fig. 8. Here, X axis “iteration” of this figure means the number of simulation iteration times and Y axis “temperature” means the maximum temperature (temperature of hotspots) of each IGBT. In Fig. 8, temperature of IGBT S ${}_{a1}$ and S ${}_{a2}$ are taken as examples. In the zone “A”, S ${}_{a1}$ is in degradation condition and in zone “B”, S ${}_{a2}$ is in degradation condition. In the left side of zone “A” or the right side of zone “B”, the inverter operates in normal condition. The real temperatures of IGBTs in phase-a under normal and degradation condition are listed in Table 2.

Under different environment temperature and parameters of thermal simulation in FloTHERM, 130 samples of thermal images are derived with 5 groups power losses of IGBT. These total samples include 10 samples of images for each IGBT in degradation and 10 samples of images for normal condition. Based on the image processing and detecting method proposed above, we derived the detecting results of thermal images of 130 samples and the results of 121 samples are right. As a result, the detection accuracy is about 93.1% (121/130).

4. Summary

In the presence of IGBT degradation in NPC inverter, a detecting method based on infrared thermography and image processing technology is proposed. The unique features of the proposed method include non-destructive and high accuracy. And it is especially suitable for the situation that there are lots of detecting hotspots but inaccessible. A simulation test using FloTHERM is performed to validate the effectiveness and accuracy of the method. The results show that the detecting accuracy is about 93.1%. In future research, intelligent methods such as artificial neural network [26, 27] or pattern recognition technology may be used to improve the defects of the detecting results. And an experimental test based on infrared camera or other infrared tools should be performed too.

Footnotes

Acknowledgments

This work is supported by Ningbo Municipal Natural Science Foundation numbered 2016A610222, Educational Commission of Zhejiang Province of China numbered Y201534437, Natural Science Foundation of Zhejiang numbered LY16G020012, Major Research Projects of Humanities and Social Sciences in Colleges and Universities of Zhejiang numbered 2014GH015.

References

Babaei

Laali

and Bayat

, A single-phase cascaded multilevel inverter based on a new basic unit with reduced number of power switches, IEEE Transactions on Industrial Electronics 62(2) (2015), 922–929.

Peng

Liu

Yang

Zhang

Gunasekaran

and Karki

, Transformer-less unified power-flow controller using the cascade multilevel inverter, IEEE Transactions on Power Electronics 31(8) (2016), 5461–5472.

Reznikov

Srndovic

Familiant

Y.L.

Grandi

and Ruderman

, Simple time averaging current quality evaluation of a single-phase multilevel PWM inverter, IEEE Transactions on Industrial Electronics 63(6) (2016), 3605–3615.

Zhu

and Xu

, Control system of speed vector on mineral ventilation motor, Journal of Computational Methods in Sciences and Engineering 15(2) (2015), 277–286.

Ceballos

Pou

Robles

Zaragoza

and Martin

J.L.

, Three-leg fault-tolerant neutral-point-clamped converter, Converter 2007 IEEE International Symposium on Industrial Electronics, Vigo, Spain (2007), 3180–3185.

Petrosyants

K.O.

Kharitonov

I.A.

Ryabov

N.I.

Kozynko

P.A.

and Lvov

B.G.

, Hardware-software subsystem for multilevel thermal fault detection and analysis of electronic components, Components 2016 International Siberian Conference on Control and Communications (SIBCON), Moscow (2016), 1–6.

Andonova

Angelov

and Chernev

, Diagnostics of packaged ICs by infrared thermography, Proceedings of the 2014 37th International Spring Seminar on Electronics Technology, Dresden (2014), 261–266.

Zou

and Huang

, An intelligent fault diagnosis method for electrical equipment using infrared images, Control Conference (CCC) 2015 34th Chinese, Hangzhou (2015), 6372–6376.

Jung

H.H.

and Lyou

, Matching of thermal and color images with application to power distribution line fault detection, Control Automation and Systems (ICCAS) 2015 15th International Conference on, Busan (2015), 1389–1392.

10.

Yang

Xia

and Tang

, Detection and fault diagnosis of power transmission line in infrared image, Cyber Technology in Automation Control and Intelligent Systems (CYBER) 2015 IEEE International Conference on, Shenyang (2015), 431–435.

11.

Widodo

Satrijo

Huda

Lim

G.M.

and Yang

B.S.

, Application of self organizing map for intelligent machine fault diagnostics based on infrared thermography images, Bio-Inspired Computing: Theories and Applications (BIC-TA) 2011 Sixth International Conference on, Penang (2011), 123–128.

12.

Hao

and Wang

, Research on infrared diagnosis of circuit based on data fusion, Fusion 2015 12th IEEE International Conference on Electronic Measurement & Instruments (ICEMI), Qingdao, China (2015), 53–57.

13.

Chou

Y.C.

and Yao

, Automatic diagnostic system of electrical equipment using infrared thermography, Soft Computing and Pattern Recognition 2009 SOCPAR’09 International Conference of, Malacca (2009), 155–160.

14.

Mazioud

Ibos

Khlaif

and Durastant

, Detection of rolling bearing degradation using infrared thermography, Proceeding of the 9th International Conference on Quanitative Infrared Thermography Krakow 2008.

15.

Bai

Zhang

Duan

and Wang

, Coupling fault diagnosis of rotating machinery by information fusion, Fusion 2016 Annual Reliability and Maintainability Symposium (RAMS), Tucson, AZ (2016), 1–6.

16.

Guo

Liu

Xie

and Yang

, Analysis on internal defects of electrical equipments in substation analysis using heating simulation for infrared diagnose, Diagnose 2014 China International Conference on Electricity Distribution (CICED), Shenzhen (2014), 39–42.

17.

Dragomir

Adam

Andrusca

Molodeschi

and Pantelimon

, About thermal stresses monitoring and diagnosis of electrical equipment, Electrical and Power Engineering (EPE) 2014 International Conference and Exposition on, Iasi (2014), 289–294.

18.

Kim

T.J.

Lee

W.C.

and Hyun

D.S.

, Detection method for open-circuit fault in neutral-point-clamped inverter systems, IEEE Transactions on Industrial Electronics 56(7) (2009), 2754–2763.

19.

Huang

Hou

Zhang

Lin

and Yang

, Research on thermal design and temperature increase simulation of converter system for high speed EMU, Journal of Power Supply 9(4) (2011), 68–72.

20.

Dieckerhoff

Bernet

and Krug

, Power loss-oriented evaluation of high voltage IGBTs and multilevel converters in transformerless traction applications, IEEE Transactions on Power Electronics 20(6) (2005), 1328–1336.

21.

Sonnenfeld

Goebel

and Celaya

J.R.

, An agile accelerated aging characterization and scenario simulation system for gate controlled power transistors, IEEE AUTOTESTCON (2008), 208–215.

22.

Brown

Abbas

Ginart

Ali

Kalgren

and Vachtsevanos

, Turn-off time as a precursor for gate bipolar transistor latch-up faults in electric motor drives, Annual Conference of the Prognostics and Health Management Society 2010, OR, Portland (2010).

23.

Nasrin

M.S.

and Khan

F.H.

, Characterization of aging process in power converters using spread spectrum time domain reflectometry, Reflectometry 2012 IEEE Energy Conversion Congress and Exposition (ECCE), Raleigh, NC, USA (2012), 2142–2148.

24.

Patil

Celaya

Das

Goebel

and Pecht

, Precursor parameter identification for insulated gate bipolar transistor (IGBT) prognostics, IEEE Transactions on Reliability 58(2) (2009), 271–276.

25.

Otsu

, A threshold selection method from gray-level histograms, IEEE Transactions on Systems Man and Cybernetics 9(1) (1979), 62–66.

26.

Rafsanjani

M.K.

and Samareh

, Chaotic time series prediction by artificial neural networks, Journal of Computational Methods in Sciences and Engineering 16(3) (2016), 599–615.

27.

Boutalis

Y.S.

and Kosmidou

O.I.

, A feedback linearization technique by using neural networks: Application to bioprocess control, Journal of Computational Methods in Sciences and Engineering 6(5) (2006), 269–282.

IGBT	Temperature [ ${{}^{\circ}}$ C]
	Normal	S ${}_{a1}$ in degradation	S ${}_{a2}$ in degradation
S ${}_{a1}$	67.4	70.2	67.4
S ${}_{a2}$	90.9	91.2	96.3
S ${}_{a3}$	91.8	91.9	91.6
S ${}_{a4}$	66.6	66.7	66.7

Detection of IGBT degradation in NPC inverter based on infrared thermography

Abstract

Keywords

1. Introduction

2. Theory of IGBT power loss in NPC inverter

2.1 Brief description of NPC inverter

Table 1 Switching command and bridge voltage of the NPC inverter

2.4 Separate objects from background

3.1 Proposed detecting method

4. Summary

Footnotes

Acknowledgments

References

Table 1
Switching command and bridge voltage of the NPC inverter