Fault Diagnosis of SEPIC Converters Based on
PSO-DBN and Wavelet Packet Energy Spectrum
Quan Sun
1
, Youren Wang
1
, Yuanyuan Jiang
1,2
, Liwei Shao
1
, Donglei Chen
1,3
1
College of Automation Engineering
Nanjing University of Aeronautics and Astronautics
Nanjing, China
2
College of Electrical Engineering and Information
Anhui University of Science and Technology
Huainan, China
3
School of Hydraulic, Energy and Power Engineering
Yangzhou University
Yangzhou, China
sequel2005@163.com
Abstract—Effective fault diagnosis for mission-critical and
safety-critical systems has been an essential and mandatory
technique to reduce failure rate and to prevent unscheduled
shutdown. A novel optimization deep belief network (DBN) in
this study has been proposed for a closed-loop single-ended
primary-inductance converter (SEPIC) fault diagnosis. Firstly,
the wavelet packet decomposition technique is used to extract the
energy values from voltage signals of four circuit nodes, as the
fault feature vectors. Then, a four-layer DBN architecture
including input and output layers is developed. Meanwhile,
neuron numbers of the two hidden layers are selected by particle
swarm optimization (PSO) algorithm and training data. Finally,
eleven fault modes such as power MOSFET, inductance, diode
and capacitor open circuit faults (OCFs) and short circuit faults
(SCFs) are isolated by PSO-DBN. Compared with other
intelligent diagnostic method such as back propagation neural
network (BPNN) and support vector machine (SVM), experiment
results show that the proposed method has a higher classification
accuracy rate which means its effectiveness and superiority.
Keywords—SEPIC converters; Faults Diagnosis; Fault
Feature; Particle Swarm Optimization(PSO); Deep Belief
Network(DBN)
I. INTRODUCTION
Power electronic converters (PECs) have been witnessing
remarkable progress in various fields such as aerospace,
photovoltaic/wind power generation, smart grid and electric
vehicles [1-2]. With the increasing of switching frequency and
power-level of PECs, power electronic systems are toward
more complex and prone to suffer performance degradation,
even critical failures when they are exposed to harsh operating
and environmental conditions, i.e., high temperature, over
current, over voltage, mechanical vibration, electromagnetic
stress and radiation. Additionally, it’s often not possible and
cost effective for the faulty PECs to entirely apply redundancy
strategies due to the constraints of their volume and the ability
of fault tolerance. Therefore, effective fault diagnosis technique
has emerged as an essential advanced technology to prevent
PECs from a malfunction in many mission-critical and safety-
critical systems, which can improve the reliability and
availability as well as reduce the downtime and maintenance
cost [3].
In recent years, fault diagnosis and condition estimation of
power switches and capacitors has been the focus of PHM for
PECs[4-7]. Buiatti et al. evaluated the capacitor parameters by
collecting the input current and output voltage in different
discrete state with the Boost topology[8]. The capacitance of
electrolytic capacitors could be obtained when the power
switch is under on-state condition, while the equivalent series
resistance (ESR) can be calculated during the off-state,
employing a polynomial fitting approach based on least mean
squares algorithm. Amaral et al. proposed an on-line fault
detection strategy of electrolytic capacitors using the relational
expression of input current and output voltage ripple[9]. Ma et
al. established the hybrid model of a Buck circuit and presented
a parameter identification-based method to obtain the ESR and
capacitance[10]. However, this method needs a higher data
sampling frequency more than the power switch operating
frequency. These strategies above mainly refer to the circuit
model-based of the converters by extracting the value of ESR
and/or capacitance from output voltage or current ripple. Yao
et al. deduced the ESR and capacitance computational formulas
from analyzing the capacitor voltage ripple and designed a
measurement system based on DSP without additional current
sensor[11]. This method has an advantage that capacitor
voltage values during a switching cycle need to be collected
twice at different time point, in that it can on-line monitor the
ESR and capacitance variation at different working conditions.
Sun et al. presented a parameter estimation approach using
intelligent optimization algorithm to complete multi-
component (including inductor, capacitor and power MOSFET)
soft fault diagnosis[12]. Twenty-one methods for OCF and ten
methods for SCF of IGBTs in three-phase power inverters are
evaluated and summarized by Lu et al. in [13]. Oh et al. [14]
analyzed the present condition monitoring techniques for IGBT
modules, including on-state collector emitter voltage-based,
gate emitter threshold voltage-based, switching time-based and
This work is supported by National Natural Science Foundation of China
(No. 61371041), the Fundamental Research Funds for the Central Universities
and Funding of Jiangsu Innovation Program for Graduate Education (No.
KYLX_0250).
978-1-5386-0370-3/17/$31.00 ©2017 IEEE