IEEE TRANSACTIONS ON COMPONENTS, PACKAGING AND MANUFACTURING TECHNOLOGY, VOL. 12, NO.

5, MAY 2022 769

Intermittent Fault Modeling and RUL Prediction

for Degraded Electrical Connectors
in Vibration Environments
Xianzhe Cheng , Kehong Lv, Guanjun Liu, and Jing Qiu

Abstract— Intermittent faults are very common problems in make it difficult to capture and analyze the intermittent fault
various kinds of electronic devices. This article is focused on the signals. Meanwhile, the intermittent faults caused by perfor-
intermittent fault phenomenon in degraded electrical connectors mance deterioration turn out to be a prelude before failure
in vibration environments. To analyze the intermittent faults
in degraded electrical connectors, a micro-state contact model [5], [6], such as the contact degradation in electric systems.
is constructed to simulate the resistance variation behavior in The condition monitoring and life prediction of electronic
vibration environments. The relationship between the intermit- products is a big challenge in reliability. It is usually difficult
tent fault behavior and the degradation state is studied, and a to find a proper health indicator to characterize the degradation
health indicator parameter is extracted from the signals to act as state of a certain component. The phenomenon of intermittent
a symptom for degradation. A test platform is built to collect the
intermittent fault signals from electrical connectors at different faults in electric systems is highly related to the degradation
degradation states in vibration environments. The degradation process of the devices [5]–[8]. Thus, it is worthwhile to
curve of the health indicator is used to predict the remaining research on the characteristics of intermittent faults within the
useful life (RUL) of the electrical connectors. The method of life cycle of the electrical components.
extreme learning machine (ELM) is adopted to conduct the As typical connection elements in electronic equipment,
prediction of RUL. The genetic algorithm (GA) is used to optimize
the weight and bias parameters in ELM. The results reflect that electrical connectors are very important in transmitting sig-
the intermittent fault characteristic is a proper degradation index nals and building connections among electric components.
in the prediction of RUL for degraded electrical connectors. A statistical analysis shows that the electronic failure caused
Index Terms— Contact resistance, electrical connector, genetic by electrical connectors accounts for about 30%–60% [9].
algorithm-extreme learning machine (GA-ELM), intermittent Intermittent faults start to emerge in electrical connectors
fault, remaining useful life (RUL) prediction. after degradation, caused by environmental stresses such as
temperature, humidity, and vibration. It is necessary to have a
I. I NTRODUCTION deep knowledge about the characteristics of intermittent faults
in electrical connectors.
I NTERMITTENT faults are defined as short-discontinuity
faults that could recover to their normal state within limited
time or under appropriate conditions without repair [1]–[3].
In recent years, many researchers have conducted exper-
iments to explore the degradation regulations in electrical
As a main kind of faults widely distributed in electronic connectors. Ren et al. [10] recorded the electrical contact
components, intermittent faults bring great troubles and chal- resistance of electrical connectors under different vibration
lenges to the use and support of the equipment. According stresses and explored its variation features. Shi et al. [11]
to the experience of aircraft repair and maintenance, nearly proposed a severity evaluation method based on a long short-
half of the faults reported in the built-in test (BIT) system term memory (LSTM) network to assess the severity of inter-
during the mission period cannot be detected through the mittent faults in an electrical connector through a shock test.
subsequent ground maintenance procedure. A large part of Shen et al. [12] investigated the performance of intermittent
such phenomena is due to the intermittent faults caused by faults of the electrical connectors under different vibration
component degradation and environmental stress [4]. conditions. Many other research teams have also conducted
Different from permanent faults that keep a stable state after similar work to research on the fretting mechanism of electri-
emerging, intermittent faults are transient and random, which cal connectors in vibration environments [13]–[17].
However, few articles are focused on the variation char-
Manuscript received December 5, 2021; revised March 10, 2022; accepted acteristics of intermittent faults in the degradation process
April 12, 2022. Date of publication April 18, 2022; date of current ver-
sion May 25, 2022. Recommended for publication by Associate Editor of an electrical connector. In this article, we focus on the
X. Zhou upon evaluation of reviewers’ comments. (Corresponding authors: characteristics of intermittent faults in electrical connectors
Guanjun Liu; Jing Qiu.) during the degradation process. A micro-state contact model
The authors are with the Science and Technology on Integrated Logistics
Support Laboratory, College of Intelligence Science and Technology, National is built to describe the phenomenon of intermittent faults in the
University of Defense Technology, Changsha 410073, China (e-mail: contact resistance of an electrical connector. The health indi-
[email protected]; [email protected]; [email protected]). cator parameters are defined based on the intermittent faults to
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCPMT.2022.3167638. characterize the degradation state of the electrical connector.
Digital Object Identifier 10.1109/TCPMT.2022.3167638 Experiments are conducted to collect intermittent fault signals
2156-3950 © 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
770 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING AND MANUFACTURING TECHNOLOGY, VOL. 12, NO. 5, MAY 2022

Fig. 1. (a) Vibration test platform for electrical connectors. (b) Test method
to collect intermittent fault signals.
Fig. 2. (a) Time-domain curve of the intermittent fault. (b) Frequency
in the electrical connector under different degradation states spectrum of the intermittent fault. (c) Enlarged view of the intermittent fault
and the vibration signals.
in a vibration environment. The health indicator parameters
are extracted from the signals to conduct a prediction of
the insertion force decreases to be around 1 N. Meanwhile,
RUL for degraded electrical connectors. An extreme learning
the static contact resistance keeps almost the same at around
machine (ELM) is adopted as the prediction method for data
1 m after degradation. When vibration stress is excited on
training and test, and a genetic algorithm (GA) is proposed to
a degraded electrical connector, intermittent faults could be
optimize the values of weight and bias in ELM.
obtained from the test system, which are manifested as an
instantaneous increase in the voltage signals.
II. P HENOMENON OF I NTERMITTENT FAULTS IN
Fig. 2(a) and (b) shows the intermittent fault time-domain
E LECTRICAL C ONNECTORS
curve and its frequency spectrum from a degraded electri-
To observe the phenomenon of intermittent faults in an cal connector in a vibration environment. The corresponding
electrical connector, a vibration test platform is established, vibration frequency is 120 Hz and the vibration amplitude
as shown in Fig. 1(a). An electrical connector is fixed on is 10 g. It can be seen from the time-domain diagram that
a vibration table in connection with a fixture. The contact the intermittent faults are distributed throughout the whole
resistance between the plug and the socket is collected through time stage, and the amplitudes of the intermittent faults turn
the four-wire test method, as shown in Fig. 1(b). Through this out to be random. From the frequency spectrum diagram, the
method, the influence of wire resistance would be eliminated main frequency components of the intermittent fault curve
from intermittent fault signals. The current source is set as 1 A. are 120 Hz and its harmonic frequency points, which means
The static contact resistance of the electrical connector is about that the occurrence of intermittent faults is consistent with
1 m, so that the voltage drop between the connector plug the external vibration frequency. From the enlarged diagram
and the socket is around 1 mV. The data acquisition (DAQ) of the intermittent fault curve and the vibration signal curve
card used in the test platform is NI 9232, which has a 24-bit shown in Fig. 2(c), it is found that the phase delay between
resolution and an acquisition voltage range set from–10 to the peak of the intermittent fault signals and the peak of the
10 V. Then the measurement resolution of voltage DAQ is vibration signals keeps almost the same [18]. It reflects that
calculated to be around 1.2 μV, which meets the requirement the intermittent fault signals of the electrical connector appear
of the acquisition resolution. The sample rate of the DAQ in accordance with external vibration signals.
system is set as 100 kHz.
To simulate the degradation of the electrical connectors, III. I NTERMITTENT FAULT M ODELING
the pinhole in the socket is enlarged to imitate the stress W ITH C ONTACT R ESISTANCE
relaxation damage. The original insertion force between the Many researchers have worked on electrical contact model-
plug and the socket is about 3.5 N. In a loose connection, ing. Angadi et al. [19] presented a review of the finite element

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
CHENG et al.: INTERMITTENT FAULT MODELING AND RUL PREDICTION FOR DEGRADED ELECTRICAL CONNECTORS 771

Fig. 3. Empirical evolution curve of the intermittent faults in connection

failures.

model (FEM) involving mechanical and/or electrical and/or

thermal fields of electrical connectors. Fu et al. [20] developed
a 2-D FEM to calculate the motions in connector system.
Xu et al. [21] proposed a theoretical model to describe the
dynamic characteristics of contact resistance for connectors
under vibration conditions. However, a model to explain the
characteristics of intermittent faults in electrical connectors has
not been fully proposed.
The intermittent faults are related to the degradation state
and external environmental stresses simultaneously. When
there is no stress or the stress is not enough to activate
the intermittent faults in a degraded connector, the electrical
performance of the connector seems to be similar as in a
normal state. While under severe environmental stresses, the
intermittent faults start to occur. The empirical evolution curve
of the intermittent faults in connection failures is shown in
Fig. 3 [4]–[6]. In stage 1, the intermittent faults are man-
ifested as small noise fluctuations. Then the amplitude and
duration time start to increase in stage 2, with the variation
becoming severe in stage 3. As for an electrical connector,
the degradation process takes quite a long time from several
years to even several decades. During the degradation process,
the intermittent faults are manifested as an instant increase in
the contact resistance under external stresses. To explore the
variation characteristics of intermittent faults in the electrical
connector, an intermittent fault model is built to describe the
Fig. 4. Schematic of (a) contact surfaces, (b) contact asperities, (c) GW
variation in the contact resistance in a vibration environment. model, and (d) conductive spot.
The model is constructed based on a micro-state contact
analysis. When a plug is inserted into the socket of an elec-
trical connector, the micro asperities on the surface of the the micro asperities as a series of circular tips with the same
pin and pinhole will get into contact and be compressed into curvature and random peak position. In the model, one of the
conductive spots. The schematic of the contact surfaces is contact surfaces is treated as an ideal plat surface. Fig. 4(c)
shown in Fig. 4(a). To calculate the contact resistance, the shows the diagram of the model, in which ϕ(z) is used to
contact surfaces are simplified as two rows of micro asperities, describe the distribution probability of height z of the micro
as shown in Fig. 4(b). While the electrical connector is in asperities. Assume that the distance between the plat surface
a degraded state, relative movement will take place between and the surface with zero-height micro asperities is h. When
the two rows of micro asperities in a vibration environment, the height of the micro asperity is larger than h, the asper-
both in the horizontal and vertical directions. The contact ity would be compressed into conductive spot, as shown in
resistance would vary due to the relative movement between Fig. 4(d). In the diagram, d refers to the press depth of the
the two rows of micro asperities. At a specific moment, the asperity and a is the radius of the conductive spot.
contact surfaces may separate from each other, resulting in The intermittent fault model of the electrical connector is
the disappearance of the conductive spots, which is manifested constructed based on the G-W model. Assume that the two
as an instantaneous increase in the contact resistance within rows of micro asperities follow a normal distribution with
intermittent fault signals. ϕ(z 1 ) and ϕ(z 2 ) separately, and the radius and the number of
A classical model called G-W model was proposed by the micro asperities are r and N, respectively. Then the height
Greenwood and Williamson [22] to analyze the contact state of the asperity in Fig. 4(c) is expressed as z = z 1 +z 2 . ρ refers
between two random rough surfaces, which is used to simplify to the electrical conductivity. According to the Hertz contact

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
772 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING AND MANUFACTURING TECHNOLOGY, VOL. 12, NO. 5, MAY 2022

theory and electrical contact theory [23], [24], the number

of the conductive spots Nc can be calculated, together with
the radius ai and the resistance Ri of each contact spot. The
contact resistance Rc of the electrical connector is formed by
the parallel connection of the conductive spots. The calculation
formula for the contact resistance is shown in the following
equation:
⎧  ∞
⎪
⎪ N = N ϕ(z)dz
⎪
⎪ c
⎪
⎪ h0
⎪
⎪
⎪
⎪ d = z − h0
⎪
⎪
⎪
⎨ √
ai = r d
(1)
⎪
⎪
⎪
⎪ R = ρ/2a
⎪
⎪
i i
⎪
⎪ 
⎪
⎪ Nc
⎪
⎪
⎪
⎩ Rc = 1 1/Ri .
i=1 Fig. 5. (a) Simulation intermittent fault signal and (b) its frequency spectrum.

Under a vibration environment, the contact surfaces in a TABLE I

degraded connector would fluctuate in the vertical direction VALUES OF PARAMETERS IN I NTERMITTENT FAULT M ODEL
and slide in the horizontal direction, following the outside
vibration movement. As a result, conductive spots will disap-
pear and regenerate according to the contact state of the asper-
ities. To simulate this variation process through the model,
it is assumed that the distance between the two rows of micro
asperities would fluctuate in accordance with the external
vibration, between which slide would take place. There will
be a reduction in the distance fluctuation amplitude from
the external vibration amplitude. The reduction coefficient
is supposed to be α. As for the slide process between the
two rows of micro asperities, the number of asperities in the integration of the intermittent fault signals in a certain period
contact surfaces is relatively stable because the contact area of time, which is expressed as S in (2). According to the
keeps almost the same even under sliding conditions. Here, empirical evolution curve of the intermittent faults in con-
the number of the slide asperities Noffset is supposed to be a nection failures shown in Fig. 3, the health indicator would
random integer in the slide process, whose maximum value is increase with the degradation of the electrical connector,
set as one-tenth of the whole number of asperities. Based on so that it is feasible to reflect the degradation state of the
these assumptions in the micro asperities, the intermittent fault connector
model is built up to calculate the variation in the contact resis-  t+t
tance in an electrical connector in a vibration environment. S= Rc (t)dt. (2)
In the calculation process, the external vibration frequency t
is set as 120 Hz, and the vibration amplitude is set as 0.35 mm In the intermittent fault model, the distance between the
with 10 g as the corresponding vibration magnitude. According two rows of micro asperities reflects the degradation state
to the general experience of the electrical contact theory of the electrical connector. The parameter h increases with
[23], [24], the values of parameters of the intermittent fault the degradation process, corresponding to the stress relaxation
model are shown in Table I. Through model simulation, the damage in the connector. In the simulation process, h is
variation in contact resistance under vibration conditions can set as 20 values from 4e−8 to 6e−8 m with a gradient
be calculated, as shown in Fig. 5. The frequency spectrum increase to imitate 20 degradation states. The intermittent
of the intermittent fault signals is also presented in the figure. fault signals under different degradation states are obtained
From the simulation signals, it indicates that the amplitudes of to calculate the value of health indicator. Fig. 6 indicates the
the intermittent faults are random, and the frequency spectrum variation curve of the value of health indicator along with
reflects that the intermittent fault signals are in accordance the degradation process of the electrical connector. The red
with the external vibration. The simulation result is very curves reflect the variation process of the health indicator from
similar to the intermittent fault behavior in Section II. ten simulation results, and the blue curve refers to the mean
With the intermittent fault model, the evolution character- value of the ten simulation results. The figure shows that the
istics of intermittent faults in the degradation process of the evolution of the health indicator follows an exponential growth
electrical connector are analyzed and studied. Considering the trend. It is concluded that the proposed health indicator is
randomness of intermittent faults, the health indicator for appropriate to evaluate the degradation state of the electrical
the degraded electrical connector is proposed as the numerical connector.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
CHENG et al.: INTERMITTENT FAULT MODELING AND RUL PREDICTION FOR DEGRADED ELECTRICAL CONNECTORS 773

Fig. 6. Variation curve of the health indicator values along with the
degradation process of the electrical connector.

Fig. 8. Intermittent fault samples under three different degradation states.

(a) Early-degradation state, (b) medium-degradation state, (c) late-degradation
state.

electrical connector under different degradation states could

be obtained. Fig. 8 shows three intermittent fault samples of
the electrical connector at different degradation states. The
intermittent faults are becoming severer with the degradation
of the connector.
The health indicator parameters are extracted from intermit-
Fig. 7. Degradation platform for the electrical connector.
tent fault samples. Then the degradation curve of the health
indicator along with time can be obtained. Fig. 9 shows
IV. D EGRADATION T EST AND RUL P REDICTION seven degradation curves of seven electrical connectors after
FOR E LECTRICAL C ONNECTORS normalization. The curves are cut off when the health indicator
A. Degradation Test for Electrical Connectors exceeds a uniform failure threshold. The differences among
these seven degradation curves indicate the diversity of the
The degradation rate of an electrical connector is very low electrical connectors, which may be caused by the random
under general vibration conditions, whose performance keeps factors in the connector structure or in the degradation test.
almost the same for years or even decades. To accelerate In addition, all the curves follow an exponential growth trend,
the degradation process, an electrical connector is forced to which is consistent with the simulation results. Therefore,
produce fretting motion between the plug and the receptacle it is appropriate to use the health indicator parameter of
in a vibration environment. As shown in Fig. 7, the receptacle the intermittent fault signal to predict the RUL of electrical
is fixed on a vibration table, and the plug is suspended within connectors.
a clamping fixture. The vibration environment is set to be the
same as the previous condition.
The intermittent fault signals of the electrical connector are B. RUL Prediction for Electrical Connectors Based on
collected in the degradation process. Since the degradation Genetic Algorithm-Extreme Learning Machine (GA-ELM)
process lasts for a long time, continuous sampling will lead to According to the degradation curve of the health indicator
a great amount of data and cause trouble for data process- extracted from the intermittent fault signals of the electrical
ing. Thus, the intermittent sampling method is adopted in connector, the historical data of the intermittent fault signals
the degradation test for electrical connectors. The acquisition could be used to predict the RUL of the electrical connector.
time is set as 10 s to form one sample of the intermittent ELM is a novel machine learning method for time series
fault signals. The time interval between two data samples prediction [25]–[28], which is a special type of single-hidden-
is set as 1000 s. Then the intermittent fault signals of an layer feedforward neural network (SLFN). The weight and

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
774 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING AND MANUFACTURING TECHNOLOGY, VOL. 12, NO. 5, MAY 2022

Fig. 11. Flowchart of GA.

Fig. 9. Degradation curves of the health indicator for electrical connectors.

can be expressed in a matrix form as

Hβ = T (4)

where
⎡ ⎤
g(ωT1 ∗ x1 + b1 ) ... g(ωTL ∗ x1 + b L )
⎢ .. .. ⎥
H=⎣ . ... . ⎦ (5)
g(ω1 ∗ x N + b1 )
T
... g(ω L ∗ x N + b L )
T
N ×L
⎡ ⎤ ⎡ T⎤
β11 . . . β1m t1
⎢ .. ⎥ ⎢ .. ⎥
β = ⎣ ... ... . ⎦ , and T = ⎣ . ⎦ . (6)
β L1 ... β Lm L×m
tTN N ×m

In an ELM, the input weight wi and bias bi are randomly

assigned, and then the output weight β can be determined by
finding the least-square solution of (4), which is expressed as

β = H† T (7)
Fig. 10. Basic structure of an ELM.
where H† represents the Moore–Penrose generalized inverse
bias values between the input layer and the hidden layer of matrix H.
are set as random values in an ELM, and then the weight As the input weight and bias values are randomly chosen
values between the hidden layer and the output layer are deter- in ELM, the performance would vary with different parameter
mined analytically by the training output samples. ELM has settings. GA is an efficient global optimization search algo-
an extremely fast learning speed and superior generalization rithm based on natural selection and genetic theory [29]–[32].
performance. This method is adopted in RUL prediction for GA combines the survival of the fittest rule in biological evo-
electrical connectors. The basic structure of an ELM is shown lution with genetic variation mechanism. The parameters are
in Fig. 10. encoded as chromosomes through the algorithm, and then iter-
Suppose that there are N arbitrary distinct samples ative methods are used to exchange chromosome information
(xi , ti ), where xi = [x i1 , x i2 , . . . , x in ]T ∈ R n , and ti = in the population. The information exchange process includes
[ti1 , ti2 , . . . , tim ]T ∈ R m . An ELM with L hidden nodes and the selection, crossover, and mutation methods. The chromo-
activation function g(x) is mathematically modeled as somes’ population is manipulated through these methods from
generation to generation to obtain the optimal solution, which

L
is determined by the fitness function. The flowchart of GA is
βi g(wiT · x j + bi ) = t j , j = 1, . . . , N (3) shown in Fig. 11.
i=1
GA is used to optimize the input weight and bias values of
where wi = [wi1 , wi2 , . . . , win ]T is the input weight vector ELM to form a genetic algorithm-extreme learning machine
connecting with the i th hidden node, β i = [βi1 , βi2 , . . . , βim ]T (GA-ELM). In GA, the chromosomes correspond to the weight
is the output weight vector connecting with the i th hidden and bias values, which is constructed by binary strings with a
node, and bi is the bias of the i th hidden node. The model fixed length. The main steps of GA-ELM are as follows.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
CHENG et al.: INTERMITTENT FAULT MODELING AND RUL PREDICTION FOR DEGRADED ELECTRICAL CONNECTORS 775

TABLE II
PARAMETER S ETTINGS IN GA-ELM

1) Determine the topology structure of ELM, including the

number of neurons in the input layer, hidden layer, and output Fig. 12. Prediction of the health indicator for sample #1.
layer.
2) Encode the weight and bias values between the input
layer and the output layer to obtain the initial population of
the chromosomes.
3) Decode the weight and bias values from the chromo-
somes, and train ELM using the weight and bias parameters
with the training samples.
4) Calculate the output with the test samples through trained
ELM and define the fitness as the error between the expected
output and the predictive output.
5) If the end condition is not met, go to Step 6. Otherwise,
carry out the selection, crossover, and mutation methods for
the current population of chromosomes and produce a new
generation of chromosomes, and then return to Step 3.
6) Choose the optimal individual chromosome and decode
it as the final weight and bias values.
7) Calculate the output weight β and use GA-ELM for Fig. 13. RUL prediction results for sample #1.
prediction.
The GA-ELM is adopted in RUL prediction for electrical
connectors with the health indicator parameter. Suppose that
the sample data of the health indicator are xt = [x 1 , x 2 , . . . , x t ]
at the current time t. The sample data need to be recon-
structed with the phase space for data training. The embedding
dimension is set as ν, then the reconstructed sample training
input can be expressed as xi = [x i , x i+1 , . . . , x i+v−1 ]T , i =
1, . . . , t − v, and the output is ti = x i+v , i = 1, . . . , t − v.
The future health indicators can be predicted through recursive
multistep forward prediction.

C. Results and Discussion

Considering the fluctuation degree of the fluctuation
processes, samples #1 and #5 are picked out for RUL pre-
diction. The prediction process starts from the last 30 sample Fig. 14. RUL prediction results for sample #5.
points in the degradation curve. The health indicator values
before the prediction point are arranged with phase space
reconstruction to form as training data. GA-ELM is con- In GA-ELM, the parameter settings are listed in Table II.
structed to predict the future trend of the health indicator after Fig. 12 shows the prediction results of the variation trend of
training. When the prediction value of the health indicator the health indicator at the start point of sample #1. It reflects
surpasses the failure threshold, it is judged as the end of useful that the prediction result from GA-ELM is very close to the
life. The distance between the prediction point and the end-of- expectation output. The estimated RUL results of sample #1
life point is the estimated RUL for the electrical connectors. and #5 are shown in Figs. 13 and 14, respectively.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
776 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING AND MANUFACTURING TECHNOLOGY, VOL. 12, NO. 5, MAY 2022

TABLE III R EFERENCES

MAPE OF THE RUL P REDICTION R ESULTS
[1] K. Breitfelder and D. Messina, The Authoritative Dictionary of IEEE
Standards Terms, 7th ed. Piscataway, NJ, USA: IEEE Press, 2000.
[2] H. Qi, S. Ganesan, and M. Pecht, “No-fault-found and intermittent
failures in electronic products,” Microelectron. Rel., vol. 48, no. 5,
pp. 663–674, 2008.
[3] Electronic Test Equipment, Intermittent Fault Detection and Isolation for
Chassis and Backplane Conductive Paths, Standard MIL-PRF-32516,
2015.
In error analysis, mean absolute percentage error (MAPE) is [4] B. Steadman, F. Berghout, N. Olsen, and B. Sorensen, “Intermittent fault
detection and isolation system,” presented at the IEEE AUTOTESTCON,
chosen to judge the performance of the two prediction meth- Sep. 2008.
ods, as shown in (8). The MAPE of the RUL prediction results [5] B. A. Sorensen, G. Kelly, A. Sajecki, and P. W. Sorensen, “An analyzer
is shown in Table III. From the prediction results, it is obvious for detecting intermittent faults in electronic devices,” in Proc. IEEE
Syst. Readiness Technol. Conf. Cost Effective Support Into Next Century
that the prediction performance of GA-ELM is better than (AUTOTESTCON), Sep. 1994, pp. 417–421.
that of ELM. The MAPEs of the prediction results for sample [6] A. Correcher, E. Garcia, F. Morant, E. Quiles, and L. Rodriguez, “Inter-
#1 and #5 with GA-ELM are 7.92% and 9.88%, respectively. mittent failure dynamics characterization,” IEEE Trans. Rel., vol. 61,
no. 3, pp. 649–658, Sep. 2012.
The prediction performance reflects that the proposed RUL [7] A. Correcher, E. García, F. Morant, and E. Quiles, “Intermittent failure
prediction method based on the health indicator and GA-ELM diagnosis based on discrete event models,” IFAC Proc. Volumes, vol. 37,
is appropriate for electrical connectors. It reflects that the no. 18, pp. 147–152, Sep. 2004.
intermittent fault can be treated as a degradation index for [8] Q. Li, K. Lv, J. Qiu, and G. Liu, “Research on residual life prediction
for electrical connectors based on intermittent failure and hidden semi-
prognostic and health management of electrical connectors Markov model,” Appl. Sci., vol. 8, no. 8, p. 1373, Aug. 2018.
n   [9] J. Swingler, J. W. McBride, and C. Maul, “Degradation of road tested
1  tˆi − ti  automotive connectors,” IEEE Trans. Compon. Packag. Technol., vol. 23,
MAPE = ∗ 100%. (8) no. 1, pp. 157–164, Mar. 2000.
n i=1 ti [10] W. Ren, D. Du, and Y. Du, “Electrical contact resistance of connector
response to mechanical vibration environment,” IEEE Trans. Compon.,
In other degraded samples, there are remarkable fluctuations Packag., Manuf. Technol., vol. 10, no. 2, pp. 212–219, Feb. 2020.
[11] J. Shi, Q. He, and Z. Wang, “An LSTM-based severity evaluation method
in the degradation curves of the health indicator, which will for intermittent open faults of an electrical connector under a shock test,”
bring significant errors to the prediction results. This phe- Measurement, vol. 173, Mar. 2021, Art. no. 108653.
nomenon is caused by the random factors in the degradation [12] Q. Shen, K. Lv, G. Liu, and J. Qiu, “Dynamic performance of elec-
trical connector contact resistance and intermittent fault under vibra-
test, which come from the environment and individual diver- tion,” IEEE Trans. Compon., Packag., Manuf. Technol., vol. 8, no. 2,
gence of the test samples. For further research, the mecha- pp. 216–225, Feb. 2018.
nism of these random factors could be studied and explored [13] S. Ling, L. Xu, D. Li, and G. Zhai, “Fretting wear reliability assessment
to improve the RUL prediction results of the electrical methodology of gold-plated electrical connectors considering manufac-
ture parameters distribution,” Microelectron. Rel., vol. 114, Nov. 2020,
connectors. Art. no. 113787.
[14] B.-H. Wu et al., “Study on the contact performance of electronic
EON connectors under axial vibration,” IEEE Trans. Compon., Packag.,
V. C ONCLUSION Manuf. Technol., vol. 8, no. 12, pp. 2090–2097, Dec. 2018.
[15] J. W. McBride, “Electrical contact bounce in medium-duty contacts,”
The characteristics of intermittent faults in electrical con- IEEE Trans. Compon., Hybrids, Manuf. Technol., vol. 12, no. 1,
nectors in vibration environments have been explored in this pp. 82–90, Mar. 1989.
article. The main conclusions are as follows. [16] H. Yang and G. Flowers, “Fretting in electrical connectors induced by
axial vibration,” IEEE Trans. Compon., Packag., Manuf. Technol., vol. 5,
1) A micro-state contact model is constructed to simulate the no. 3, pp. 328–336, Mar. 2015.
intermittent fault behavior of electrical connectors. The sim- [17] C. Maul, J. W. McBride, and J. Swingler, “Intermittency phenomena in
ulation results of the intermittent fault signals are consistent electrical connectors,” IEEE Trans. Compon. Packag. Technol., vol. 24,
no. 3, pp. 370–377, Sep. 2001.
with the experiment results, which have random amplitudes [18] X. Cheng, K. Lv, G. Liu, and J. Qiu, “Characteristics of intermittent
at different times and occur in accordance with the external fault in electrical connectors under vibration environment,” IEEE Trans.
vibration stress. Compon., Packag., Manuf. Technol., vol. 10, no. 9, pp. 1575–1578,
Sep. 2020.
2) The evolution characteristics of the intermittent faults at [19] S. V. Angadi, R. L. Jackson, V. Pujar, and M. R. Tushar, “A compre-
different degradation stages are studied with simulation and hensive review of the finite element modeling of electrical connectors
experiment analysis. The health indicator parameter is defined including their contacts,” IEEE Trans. Compon., Packag., Manuf. Tech-
nol., vol. 10, no. 5, pp. 836–844, May 2020.
as the numerical integration of the intermittent fault signals in
[20] R. Fu, S.-Y. Choe, R. L. Jackson, G. T. Flowers, and D. Kim, “Modeling
a certain period of time. Both the simulation and experimental and analysis of vibration-induced changes in connector resistance of
results reflect that the degradation curve of the health indicator high power electrical connectors for hybrid vehicles,” Mech. Based Des.
follows an exponential growth trend. Struct. Mach., vol. 40, no. 3, pp. 349–365, Jul. 2012.
[21] L. Xu, S. Ling, D. Li, and G. Zhai, “Vibration-induced dynamic
3) The GA-ELM method is proposed to conduct RUL pre- characteristics modeling of electrical contact resistance for connectors,”
diction for electrical connectors through degradation curves. Microelectron. Rel., vol. 114, Nov. 2020, Art. no. 113868.
The GA-ELM has better performance than ELM in prediction [22] J. A. Greenwood and J. B. P. Williamson, “Contact of nominally flat
surfaces,” Proc. Roy. Soc. London A, Math., Phys. Eng. Sci., vol. 295,
results, with MAPEs of within 10%. It indicates that the no. 1442, pp. 300–319, 1966.
intermittent faults can be treated as a degradation index for [23] P. G. Slade, Electrical Contacts: Principles and Applications.
prognostic and health management of electrical connectors. Boca Raton, FL, USA: CRC Press, 2017.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.
CHENG et al.: INTERMITTENT FAULT MODELING AND RUL PREDICTION FOR DEGRADED ELECTRICAL CONNECTORS 777

[24] M. Braunovic, V. V. Konchits, and N. K. Myshkin, Electrical Contacts: Kehong Lv received the B.S. degree from Xi’anjiao
Fundamentals, Applications and Technology. Boca Raton, FL, USA: Tong University, Xi’an, China, in 2001, and the
CRC Press, 2017. M.S. and Ph.D. degrees in mechatronic engineering
[25] G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, “Extreme learning from NUDT, Changsha, China, in 2003 and 2008,
machine: Theory and applications,” Neurocomputing, vol. 70, nos. 1–3, respectively.
pp. 489–501, 2006. Since 2008, he has been carrying out teaching and
[26] S. Ding, H. Zhao, Y. Zhang, X. Xu, and R. Nie, “Extreme learning research with NUDT. He is currently an Associate
machine: Algorithm, theory and applications,” Artif. Intell. Rev., vol. 44, Professor since 2014. His current research interests
no. 1, pp. 103–115, Jun. 2015. include prognostics and health management (PHM)
[27] M. A. A. Albadra and S. Tiuna, “Extreme learning machine: A review,” technology, design for testability (DFT), mechanical
Int. J. Appl. Eng. Res., vol. 12, no. 14, pp. 4610–4623, 2017. signal processing, condition monitoring, and fault
[28] G. Huang, G.-B. Huang, S. Song, and K. You, “Trends in extreme learn- diagnosis.
ing machines: A review,” Neural Netw., vol. 61, pp. 32–48, Jun. 2015.
[29] W. Martin, Foundations of Genetic Algorithms (FOGA 6). Amsterdam,
The Netherlands: Elsevier, 2001.
[30] A. Diker, D. Avci, M. Gedikpinar, and E. Avci, “A new technique
for ECG signal classification genetic algorithm wavelet Kernel extreme Guanjun Liu received the B.S. and Ph.D. degrees
learning machine,” Optik, vol. 180, pp. 46–55, Feb. 2019. from NUDT, Changsha, China, in 1994 and 2000,
[31] Y. Zhou, N. Zhou, L. Gong, and M. Jiang, “Prediction of photo- respectively, all in mechanics engineering.
voltaic power output based on similar day analysis, genetic algo- Since 2000, he has been carrying out teaching and
rithm and extreme learning machine,” Energy, vol. 204, Aug. 2020, research with NUDT. He is a Professor since 2011.
Art. no. 117894. His research interests include design for testability,
[32] K. Chen, S. Laghrouche, and A. Djerdir, “Proton exchange membrane testability demonstration, fault diagnosis, and fault
fuel cell prognostics using genetic algorithm and extreme learning prognostics.
machine,” Fuel Cells, vol. 20, no. 3, pp. 263–271, Jun. 2020.

Jing Qiu received the B.S. degree in solid mechan-

Xianzhe Cheng received the M.S. degree from ics from the Beijing University of Aeronautics and
the National University of Defense Technology Astronautics, Beijing, China, in 1985, and the M.S.
(NUDT), Changsha, China, in 2017, where he is degree in solid mechanics and the Ph.D. degree
currently pursuing the Ph.D. degree with the College in mechanics engineering from NUDT, Changsha,
of Intelligence Science and Technology. China, in 1988 and 1998, respectively.
His current research interests include intermittent Since 1988, he has been carrying out teaching and
fault modeling and estimation. research with NUDT, where he became a Teaching
Assistant in 1988, a Lecturer in 1991, an Asso-
ciate Professor in 1994, and a Professor in 1998.
His research interests include design for testability,
testability demonstration, fault diagnosis, and fault prognostics.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:07:51 UTC from IEEE Xplore. Restrictions apply.