Professional Documents
Culture Documents
https://doi.org/10.1007/s42835-019-00272-0
ORIGINAL ARTICLE
Abstract
This paper presents the prognostics of in-circuit power MOSFETs in power electronic systems using artificial neural network
(ANN). Recent industrial surveys on reliability of power electronic systems shows that the switching transistors are one of
most life-limiting component and thermal stress is major cause for their parametric degradation. The on-state drain-source
resistance (Rds(on)) is an important fault signature parameter of power MOSFET which increases with its junction tempera-
ture rise. In this work, ANN is used to estimate variation in R
ds(on) of target MOSFETs at different operating conditions. The
data set for training and testing of proposed back-propagation trained artificial neural network are experimentally obtained
from the developed test bed. Using test bed, the electrical stressed based accelerated aging test is performed on target power
MOSFETs by subjecting them to repetitive unclamped inductive switching at different operating frequency and temperature.
After off-line training, the proposed ANN is implemented using National Instruments LabVIEW software to estimate real-
time Rds(on) values of in-circuit MOSFETs. For this purpose, voltage fed half bridge inverter circuit is designed and target
MOSFETs are taken at output section of the circuit. A low cost microcontroller is programmed for acquiring and serially
transmitting the real-time data set of target MOSFETs to the LabVIEW software. The performance of the proposed method
is evaluated in real time by comparing the ANN estimated R ds(on) value with the experimental obtained value for in-circuit
target MOSFETs at test condition.
Keywords Accelerated aging · Artificial neural network · Back-propagation · On-state drain-source resistance · Prognostics
13
Vol.:(0123456789)
Journal of Electrical Engineering & Technology
the dataset for junction temperature, drain-source voltage, locations and can be extended for health monitoring of other
and drain current is experimentally monitored for the target critical circuit components in power electronic systems with
MOSFET. The dataset is used in Saber simulation tool to suitable training.
obtain the electro-thermal model of MOSFET. The rela-
tion between R ds(on) and junction temperature is obtained
which shows that R ds(on) value is directly proportional to 2 Parametric Degradation Failure in Power
junction temperature. To enhance the reliability of power MOSFET
MOSFETs in power electronic converters, their in-circuit
condition needs to be continuously monitored. This paper The understanding of the failure mechanisms in the MOS-
presents ANN model to determine the R ds(on) value of in-cir- FETs will aid in determining of the critical fault signature
cuit power MOSFETs at real-time operating conditions from parameter and based on its parametric degradation, their in-
both onsite and remote location. In this paper artificial neu- circuit condition based maintenance can be performed. As
ral network (ANN) approach is presented for determining stated earlier, thermal stress is the most important factor
the variation in the on-state drain-source resistance ( Rds(on)) that causes the failure of power MOSFETs. The junction
of target power MOSFETs at different operating frequency temperature rise leads to the degradation of power MOSFET
and junction temperature. R ds(on) value is an important fault due to the die attachment degradation. It is reported in [20]
signature parameter in power MOSFET that increases with that the die attachment degradation results in the increase
the increase in its junction temperature. The input and out- of the Rds(on) value. The increase of the drain source resist-
put datasets for implementing the ANN estimated Rds(on) ance increases the conduction losses that produce localized
value at test conditions are experimentally obtained from hot spots in MOSFETs causing its failure due to secondary
the developed test bed. Using the test bed, target power breakdown. Thermal stress based degradation due to con-
MOSFETs are subjected to the repetitive unclamped induc- duction losses in power MOSFET causing the localized hot
tive switching (UIS) based accelerated aging test at differ- spots is observed using the scanning-electron microscope
ent operating frequencies. From the resulting input data sets (SEM) [21] and the damaged chip of MOSFET is shown
obtained at different test conditions, ANN is trained using in Fig. 1,
back-propagation training algorithm to estimate the R ds(on) Rds(on) is therefore the critical fault signature in power
values for target MOSFETs. Back-propagation is an exten- MOSFET that increases with the increase in the junction
sion of the least mean squares (LMS) algorithm for multi- temperature of transistor. At wear out condition of power
layer perceptron. Back-propagation training of three layers MOSFET, its R ds(on) value becomes equal to maximum rated
ANN has been done by selecting different parameters like value due to increase in conduction losses and hence the
learning rate, momentum term and number of hidden layer junction temperature. Therefore, the variation in the on-state
neurons and their performance is evaluated to find the best drain-source resistance value from initial (pristine condi-
fit of ANN model for the R ds(on) estimation of target MOS- tion) value to the maximum rated value (weak condition)
FETs. Subsequently, weights of trained ANN are used to of the MOSFET corresponds to its wear-out or degradation
implement it for estimating the R ds(on) of target MOSFETs in failure. The experimental measurement of in-circuit R ds(on)
the real-time using LabVIEW software and data acquisition
system. For this purpose a voltage fed half bridge inverter
circuit is designed [19] with overload protection and iso-
lated gate drive switching circuit. A low cost microcontroller
is programmed for acquiring and serially transmitting the
real-time operating data set of target MOSFETs at the out-
put section of half bridge inverter circuit to the LabVIEW
software installed at host computer. In order to facilitate the
web based condition monitoring of MOSFETs, the control
of running state front panel virtual instrument (VI) file hav-
ing the ANN estimated output R ds(on) values is continuously
transferred from local host to the client over the internet as a
HTML file using web publishing tool of LabVIEW. Finally,
the implementation is validated in real time by comparing
the ANN estimated R ds(on) with its practical in-circuit value
at test condition. The proposed ANN based scheme is very
useful for in-circuit health monitoring of power MOSFETs
in power electronics circuits from both onsite and remote Fig. 1 SEM photograph showing thermal damage
13
Journal of Electrical Engineering & Technology
value for the target MOSFETs is presented at test conditions Using the test bed, two samples of target power MOS-
in Sect. 3. FETs (D2 package of STP 7NK60Z) from ST Microelec-
tronics are subjected to the repetitive UIS at different fre-
quencies. The isolated gate drive pulses for the MOSFET
3 Experimental Measurement of In‑circuit is supplied from microcontroller based variable frequency
Rds(on) For Target MOSFETs pulse generator circuit. The variable frequency square pulses
of 2 kHz, 4 kHz, 8 kHz, 10 kHz and 15 kHz are obtained
In this section, real-time in-circuit condition monitoring of by programming the timer of the microcontroller to obtain
target MOSFETs due to electrical stress based repetitive desired switching frequency in accordance to the on state
unclamped inductive switching test is presented. The electri- of the respective frequency selector switch. The generated
cal stress based accelerated aging test of power MOSFETs is switching frequency pulses are amplified using power tran-
performed using the microcontroller based test bed in which sistor BC547 and are isolated by the opto-isolator, MCT2E.
the target MOSFETs are subjected to repetitive UIS [22]. These pulses are finally provided to the gate drive of MOS-
Schematic diagram of developed test bed used to perform FET of the repetitive UIS circuit. The hardware assembly of
the aging of power MOSFETs due to the repetitive UIS is test bed is shown in Fig. 3.
shown in the Fig. 2. UIS causes avalanche energy dissipation in MOSFET that
introduces thermal cycle in it resulting in the increase of its
junction temperature. The test parameters data values for
case temperature ( Tcase), drain-to-source voltage ( VDS) and
drain-to-source current (IDS) are monitored after half hour of
the aging test at different test frequencies. The thermal data
for the case temperature is acquired precisely using linear
integrated-circuit based analog temperature sensor; LM35
which is glued on metal case of target MOSFETs. LM35
is interfaced with microcontroller based data acquisition
(DAQ) card that is programmed to acquire the case tem-
perature at different frequencies after the test interval. Case
temperature value are displayed on liquid crystal display
(LCD) interfaced with DAQ card and are also stored as text
file by serially transmitting them from microcontroller to the
HyperTerminal application program of Windows installed
on host computer. The voltage and current waveforms for
the drain-source current ( IDS) and the drain-source voltage
Fig. 2 Schematic diagram of Test bed (VDS) for target MOSFETs are captured at test frequencies
Fig. 3 Hardware Assembly of
Test Bed
13
Journal of Electrical Engineering & Technology
Fig. 4 Drain-source current through MOSFET for repetitive UIS test Fig. 6 Drain-source current through MOSFET for repetitive UIS test
at 4 kHz at 8 kHz
13
Journal of Electrical Engineering & Technology
Table 1 Variations in R
ds(on) Repetitive UIS Monitored parameters for STP 7NK60Z Monitored parameters for STP 7NK60Z
values due to repetitive UIS at test frequency (Sample-1) (Sample-2)
test conditions
F(in kHz) Junction tempera- On-state drain-source Junction tempera- On-state drain-source
ture TJ (in °C) resistance R
ds(on)(Ω) ture TJ (in °C) resistance R
ds(on)(Ω)
Stop
13
Journal of Electrical Engineering & Technology
[K ]
∑
P
∑( )2 ( )
F(w) = dkp − okp = ET E (4)
p=1 k=1
where,
E = Cumulative error vector (for all input patterns);
dkp = Desired value of kth output and pth pattern;
okp = Actual value of kth output and pth pattern;
ekp= (dkp − okp); k = 1,2…K; p = 1,2…P; and,
Fig. 9 Architecture for ANN
F(w) = Sum of squares of errors.
From relation (4), the Jacobian matrix is obtained that is
further used to update the weights. Jacobian matrix is given
in (5) as,
α = Temperature coefficient calculated from sets of (TJ1,
Rds(on1)) and (TJ2, Rds(on2)) from relationship curves given in
⎡ 𝜕e11 𝜕e11
⋯ ⎤
𝜕e11
datasheet of STP 7NK60Z. ⎢ 𝜕e
𝜕w1 𝜕w2
⎥
𝜕wN
𝜕e21 𝜕e21
From the junction temperature, the R ds(on) values for target ⎢ 𝜕w21 𝜕w2
⋯ ⎥
𝜕wN
MOSFETs at test conditions are empirically obtained by using ⎢ 1 ⎥
⎢ 𝜕e⋮ ⋮ ⋯ ⋮ ⎥
relation (2). The value of temperature coefficient (α) for STP ⎢ K1 𝜕eK1
⋯ 𝜕w ⎥
𝜕eK1
7NK60Z is calculated from its datasheet and is obtained as 0.7 ⎢ 𝜕w1 𝜕w2 N ⎥
tal aging test results obtained for variations in R ds(on) values of ⎢ 2P 𝜕e2P
⋯
𝜕e2P
⎥
⎢ 𝜕w1 𝜕w2 𝜕wN ⎥
two samples of target power MOSFETs at different switching ⎢ ⋮ ⋮ ⋯ ⋮ ⎥
frequency and the junction temperature are shown in Table 1. ⎢ 𝜕eKP 𝜕eKP
⋯ 𝜕wKP ⎥⎦
𝜕e
⎣ 𝜕w1 𝜕w2 N
In this section, the application of Levenberg–Marquardt (LM) The weights update from nth to (n + 1)th iteration is given
based back-propagation trained artificial neural network by relation (6) as,
(BPNN) model [25] for estimating the on-state drain-source [( )−1 ]
resistance of target MOSFETs at different operating frequency Wn+1 = Wn − JTn Jn + μn I JTn En (6)
and junction temperature is presented. LM algorithm mini-
mizes the sum of squares of errors with respect to weights where,
of multilayer BPNN having N hidden layer neurons, P input I = Identity matrix; and,
patterns and K outputs as per relation (3–6),
Table 2 Normalized datasets Switching frequency Normalized data for STP 7NK60Z (Sam- Normalized data for STP 7NK60Z
for Implementing BPNN to ple-1) (Sample-2)
estimate Rds(on) of MOSFETs
[F]i {F}i [T]i {T}i {Rds(on)}o [T]i {T}i {Rds(on)}o
13
Journal of Electrical Engineering & Technology
Fig. 10 BPNN model for Rds(on) estimation of target MOSFET at α = 0 and η = 0.9
13
Journal of Electrical Engineering & Technology
Fig. 11 BPNN model for Rds(on) estimation of target MOSFET at α = 0.1 and η = 0.9
13
Journal of Electrical Engineering & Technology
Serial Communication
Host Computer
(Internet Connected)
Client Computer
Real-time data
Web based Front panel array acquisition
control of Rds(on) in Front
Condition
transfer to panel at Test
monitoring Client Conditions
of MOSFET
13
Journal of Electrical Engineering & Technology
6 Results and Discussions
Fig. 17 Block Diagram VI
13
Journal of Electrical Engineering & Technology
Fig. 18 Front panel VI
for in-circuit health monitoring of MOSFETs in power elec- in-circuit monitored Rds(on) at the test conditions. The result
tronics circuits. obtained shows a low percentage of relative error which
verifies validity of proposed method. The data values for
ANN estimated Rds(on) is also stored on hard disk of host
7 Conclusions computer as MS Excel file and the front panel information
is continuously transferred to the remote user as an html file
This paper presents online condition monitoring of power using web publishing tool of LabVIEW for further analysis.
MOSFETs using ANN approach. Proposed ANN model is Therefore, using the proposed method the forecast on aging
successfully trained off-line with experimental dataset using condition of in-circuit target MOSFETs can be done in real
LM based back propagation learning algorithm. Further, time from both onsite and remote location. Proposed ANN
the ANN model is implemented using National Instruments based scheme is useful for in-circuit health monitoring of
LabVIEW software to determine real-time Rds(on) value MOSFETs in power electronics circuits.
of target MOSFETs at the operating conditions. To evalu-
ate the performance of proposed method, ANN estimated
Rds(on) values of target MOSFETs are compared with their
13
Journal of Electrical Engineering & Technology
Fig. 19 Snapshot of current and voltage waveforms for upper and lower MOSFET
13
Journal of Electrical Engineering & Technology
In: IEEE transactions on device and materials reliability 14(1): 24. Graovac D, Pűrschel M, Kiep A (2006) MOSFET power
220–228 losses calculation using the data-sheet parameter, Infineon
12. Habchi R, Salame C, Mialhe P, Khoury A (2007) Switching times Technologies, pp 1–22. http://appli c atio n -notes . digch
variation of power MOSFET devices after electrical stress. Micro- ip.com/070/70-41484.pdf. Accessed 22 July 2018
electron Reliab 47:1296–1299 25. Hagan M, Menhaj M (1994) Training feedforward networks with
13. Ye X, Chen C, Wang Y, Zhai G, Vachtsevanos G (2017) Online the Marquardt algorithm. IEEE Trans Neural Netw 5(6):989–993
condition monitoring of power MOSFET gate oxide degradation 26. Huang S, Huang Y (1991) Bounds on the number of hidden
based on miller platform voltage. IEEE Trans Power Electron neurons in multilayer perceptrons. IEEE Trans Neural Netw
32(6):4776–4784 2(1):47–55
14. Dusmez S, Bhardwaj M, Sun L, Akin B (2016) In situ condition 27. Khera N, Khan S (2018) Prognostics of aluminum electrolytic
monitoring of high-voltage discrete power MOSFET in boost con- capacitors using artificial neural network approach. J Microelec-
verter through software frequency response analysis. IEEE Trans tron Reliab 81:328–336
Ind Electron. 63(12):7693–7702
15. Dusmez S, Duran H, Akin B (2016) Remaining useful lifetime Publisher’s Note Springer Nature remains neutral with regard to
estimation for thermally stressed power MOSFETs based on on- jurisdictional claims in published maps and institutional affiliations.
state resistance variation. IEEE Trans Ind Appl. 53(3):2554–2563
16. Zhang H, Kang R, Luo M, Pecht M (2009) Precursor parameter
identification for power supply prognostics and health manage-
ment. In: Proceedings of IEEE conference on reliability, main- Dr. Neeraj Khera has received B.Tech. in Electrical Engineering from
tainability and safety. pp 883–887 Kurukshetra University in 2000 and Masters of Engg. in Electronics
17. Erturk F, Akin, B (2017) A method for online ageing detection Product Design and Technology from P.E.C Chandigarh in 2005. He
in SiC MOSFETs. In: Proceedings of applied power electronics received the Ph.D. degree in Electrical Engineering from the Jamia
conference and exposition. pp 3576–3581 Millia Islamia, Delhi, India, in 2015. He is currently Associate Profes-
18. Zhou W, Zhong X, Sheng K (2014) High temperature stability sor in Department of Electronics & Communication Amity University,
and the performance degradation of SiC MOSFETs. IEEE Trans NOIDA. His research and teaching interests include Intelligent Instru-
Power Electron 29(5):2329–2337 mentation, Embedded System Design, Fault diagnosis and condition-
19. Rashid M (2004) Power electronics circuits, devices, and appli- based maintenance and Power Electronics.
cations, 3rd edn. Prentice-Hall, New Delhi
20. Celaya J, Saxena A, Wysocki P, Saha S, and Goebel K (2010) Dr. Shakeb A. Khan has received B.Sc. Engg. degree in Electrical
Towards prognostics of power MOSFETs: accelerated aging and Engineering and M.Sc. Engg. Degree in Instrumentation and Control
precursors of failure. In: Proceedings of annual conference of from Aligarh Muslim University, Aligarh, India, in 1992 and 1994,
the prognostics and health management society. pp 1–10 respectively. He received the Ph.D. degree in Instrumentation from
21. Kibushi R, Hatakeyama T, Nakagawa S, Ishizuka M (2013) the Indian Institute of Technology, Delhi, India, in 2005. He has been
Analysis of hot spot temperature in power Si MOSFET with with Jamia Millia Islamia (A Central University), New Delhi, India,
electro-thermal analysis. In: Proceedings of IEEE conference since 1995, where he is currently Professor in the Department of Elec-
on microsystems, packaging, assembly and circuits technology. trical Engineering. He has received grants from DST (Department of
pp 211–213 Science and Technology), New Delhi, and KACST (King Abdul Aziz
22. Khera N, Tiwari S (2016) Prognostics of power MOSFET due City for Science & Technology), Riyadh, KSA to pursue research on
to unclamped inductive switching. In: Proceedings of IEEE “Thin Film Sensor & Signal Conditioning”. His research and teaching
conference on power electronics, intelligent control and energy interests include Electronics Instrumentation, Thin Film Sensors, ANN
systems. pp 1–4 applications, and Sensor Signal Conditioning.
23. Sattar A (2010) Power MOSFET basics, IXYS Corporation,
IXAN0061. http://www.ixys.com/Documents/AppNotes/IXAN0
061.pdf. Accessed 15 June 2018
13