Monte Carlo Simulation

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

Risk Assessment of Worn-out Failure of Transport


Aircraft based on Monte Carlo Simulation
Ziwen Zhang Lei Huang
College of civil aviation Department of conceptual and aerodynamic design
Nanjing University of Aeronautics and Astronautics AVIC Helicopter Research and Development Institute
Nanjing, China Tianjin, China
zhangzw@nuaa.edu.cn huanglei2933@sina.com

Zhong Lu
College of civil aviation
Nanjing University of Aeronautics and Astronautics
Nanjing, China
luzhong@nuaa.edu.cn

Abstract—Risk Assessment is a vital task in the continued set up a quantitative risk assessment model which involves
airworthiness of civil aircraft, whose process is complicated and fuzzy linguistic scale theory, and FMECA [4].W Jiang pro-
involves many parameters. In this paper, the risk assessment poses a modified failure mode and effects analysis (FMEA)
of Worn-out Failures for Transport Aircraft based on Monte
Carlo Simulation is proposed. Risk values of the worn-out cases method based on the fuzzy set theory and fuzzy analytic
is presented based on Monitor Safety/ Analyze Data (MS/AD) hierarchy process [5]. Guo put forward a novel approach
process and Transport Airplane Risk Assessment Methodology to assess the risk of aircraft electrical wiring interconnect
(TARAM) proposed by Federal Aviation Agency (FAA). And a system which accomplishes quantitative assessment through
simulation method has been put forward, which improved the description flowchart [6]. Violette applied the MS/AD process
calculation process of the control-program fleet risk of the worn-
out cases. And a Risk assessment of a simulation fleet targeted on into the risk assessment of crack in the door edge frame at the
the crack in the door edge frame at the main entry door. A control main entry door to monitor single aircraft and the whole fleet
program is proposed and quantitatively evaluated, which realizes [7].Wang et al proposed the adjoin method as a risk assessment
the evaluation of risk level and control program of continued method for life analysis of supersonic ramjet engines [8].
airworthiness. Tamasi proposed a risk assessment method based on RAMS
Index Terms—Continued airworthiness, Risk management,
System safety, Monte Carlo simulation method.Wang et al applied the Monte Carlo method to risk
assessment [10], which involves Fault tree analysis (FTA) and
I. INTRODUCTION simulation process. Wang et al studied the failure time of civil
Risk assessment is viewed as an important task during aircraft during the continued airworthiness phase proposed
continued airworthiness to maintain the safety of the civil air- risk assessment procedures for unsafe incidents [11]. Li et
craft. Nowadays, standards and techniques are put forward to al started further research on fleet risk during the continued
guide the evaluation. ARP5150 proposed in 2003 has provided airworthiness phase and proposed a calculation method of the
a systematic process to help determine the safety priorities fleet risk with failure rate by dividing engine risk levels and
[1].Federal Aviation Agency (FAA) has issued Order 8110.107 determining risk factors [12]. Zhao et al proposed a modified
Monitor Safety/ Analyze Data (MS/AD) [2] alongside with TARAM process to calculate the risk of individual aircraft and
the manual Transport Airplane Risk Assessment Methodology fleet with loss failure using historical fleet operation data [13].
(TARAM) [3]. Reference [3] presents the definition of Fleet
Risk which gives a comprehensive evaluation of the aircraft With the completion of the establishment of the risk as-
safety via the failure mechanism of the components, causal sessment process, the application of estimation methods has
chain analysis, and Risk Level theory. Many types of research also matured. However, there still exist some problems in the
on risk assessment have been done in the past decades. Lee risk assessment process. For example, the studies mentioned
above focused on the failure time estimation of the individual
The authors wish to appreciate the support from National Natural Sci- aircraft and the whole fleet. But the risk model of the state
ence Foundation of China (U1733124), the Aeronautical Science Foundation after demonstrating control measures. This paper proposes a
of China (20180252002),the Special Scientific Research Project for Civil
Aircraft(MJZ-2015-Y-010), and the Foundation of Graduate Innovation Center risk assessment method of worn-out failure of transport aircraft
in NUAA (kfjj20190733). We are grateful to anonymous referees for their based on Monte Carlo simulation and modifies the assessment
constructive comments which have led to improvements of the paper. model during developing control programs to demonstrate a
more accurate assessment.

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

978-1-7281-7102-9/20/$31.00 ©2020 IEEE


978-1-7281-7102-9/20/$31.00 ©2020 IEEE
Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.
2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

II. T RANSPORT A IRCRAFT R ISK A SSESSMENT P ROCESS n


X
A. MS/AD process RF = DA ∗ N D ∗ (CPi · IRi ) (2)
The order, Monitor Safety / Analyze Data (MS/AD) pro- i=1

poses a risk assessment approach that integrates the data- In (2), the definition of N D, CPi and IR are the same
driven method and probabilistic risk estimation. The process as those in (1). Similarly, 90-day fleet risk RF 90 and control-
uses qualitative and quantitative methods to identify and program fleet risk RF C are defined in (3).
analyze security issues in the fleet and determine whether the (
security issue poses a life threat to personnel. The high-level RF 90 = DA90 · N D90 · CP · IR · EO
(3)
view of MS/AD is shown in Fig.1 . RF C = DAC · N DC · CP · IR · EO
In (3), DA90 and N D90 respectively represent the value of
the defect aircraft and the undetected rate for 90 days. DAC ,
N DC is the value of the defect aircraft and the undetected rate
under control-program conditions. EO represents the number
of passengers per flight.
After the calculation of RI and RF , determine whether it
Fig. 1. High-level view of MS/AD is necessary to develop control programs according to table
I. When the aircraft is at risk, a control program is in need.
When aviation data triggers MS/AD, the whole process is In the meanwhile, RF C should be calculated. if RF C < 3 is
as follows: true, the control program is effective. Otherwise, the control
1) Acquire similar incidents from the database; program should be reconsidered and assessed.
2) Set hazard criteria according to the severity and filter
the events with minor effects; TABLE I
3) Perform preliminary risk assessment, study the mech- R ISK G UIDELINES
anism of target accidents and perform causal chain Uncontrolled Uncontrolled Control-program
analysis. fleet riskRF individual risk RI fleet risk RF C
4) Calculate the risk based on the analysis above and > 0.02 > 10−7 >3
identify the outcomes.
5) Set, assess and adjust the control program of the inci- The calculation methods of the variables mentioned in
dents. individual risk RI and fleet risk RF are introduced below.
In the whole process, the first two procedures are qualitative Undetected rate, represented by N D, is the probability of
analysis. The rest is quantitative analysis, which is the key not being detected by the staff during daily maintenance. Usu-
to the whole process. Therefore, this paper will follow the ally, N D is determined by experiment, engineer judgment, etc.
MS/AD process and proposes a risk assessment methodology If the failure is fatigue fracture, N D can be calculated through
to calculate the risk of fleet risk. damage tolerance analysis. In this paper, N D is acquired by
engineer judgment according to [7]. Conditional probability,
B. Transport airplane risk assessment methodology represented by CP , is the probability that the target failure
TARAM is based on 2 kinds of risk defined in MS/AD causes unsafe outcomes. CP is calculated according to the
process, Individual risk and Fleet risk. In the meanwhile, causal chain analysis. A sample causal chain is shown in Fig.2.
TARAM defines 3 risk conditions which are Uncontrolled risk,
90-day risk and Control-program risk.
Individual risk is the probability that a single aircraft
is fatally damaged and causes unsafe outcomes per flight
hour,which is represented by RI in (1).
n
X
RI = λ(t) · N D · (CPi · IRi ) (1) Fig. 2. Sample causal chain analysis
i=1
Fig.2 shows an error propagation process whose original
In (1), N D represents the Undetected rate of the target transition probability P0 . The conditional probability CPA and
failure. CPi is the conditional probability of the unsafe event CPB are calculated as (4)
i due to target failure. IRi is the severity rate of the unsafe
(
outcome i. λ(t) is the Failure rate function of the target failure. CPA = P0 ∗ PA1 · PA2 · PA3
Fleet risk is the probability that the aircraft will be fatally (4)
CPB = P0 ∗ PA1 · PB1 · PB2
damaged during the target interval and causes unsafe out-
comes. Fleet risk is represented by RF . In the condition of The severity of the unsafe outcome, represented by IR, is
uncontrolled risk, the fleet risk is calculated by (2). usually determined by the mortality rate. λ(t) in (1) is the

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.
2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

occurrence rate of the target failure. In passenger aircraft, Assume the life of the i-th aircraft is Ti . When Ti >
faults usually obey Weibull distribution represented in (5). tR , aircraft i will not malfunction until it is retired. When
Ti < t0(i) , A target failure has occurred before aircraft i
β t β−1 is operational which is not possible in this condition. So,
λ(t) = ( · ) (5)
η η simulation should be performed again if Ti < t0(i) is satisfied.
However, β and η should be evaluated in further research. When t0(i) < Ti < tR , the failure has occurred before aircraft
This paper will acquire the parameters via Weibayes method. i is retired.
What’s more, DA in (2) is the number of defect aircraft Ti is acquired according to (8), in which random number
in the fleet which changes over time. This paper proposes rand obeys uniform distribution [0, 1].
a simulation method based on Monte Carlo theory for DA 
1
 
1

targeted on worn-out failure. The calculation of DA also Ti = η · exp ln ln 1 − (8)
β rand
depends on the distribution of the target failure. Therefore,
failure is assumed to obey Weibull distribution. In chapter 3, the simulation process is shown in Fig.3. The whole process
the process to calculate DA will be presented. is divided into 7 procedures.
1) Give the number of the aircraft N , retirement life tR .
III. C ALCULATION METHOD OF FLEET RISK PARAMETERS Simulate the life of the i-th aircraft t0(i) ;
BASED ON M ONTE C ARLO SIMULATION
2) Input the shape parameter β and detected failed parts ra
Assuming there are N aircraft of a certain type in the . Calculate the scale parameter η.
fleet,the operation time of the aircraft i is represented by t0(i) 3) Initialize DA and i;
before which the aircraft has not been failed. the retirement 4) Update i where i = i + 1.
life is tR . The life of a component obeys a Weibull distribution 5) Calculate Ti according to (8).
with shape parameter β and scale parameter η. The following 6) Determine whether t0(i) < Ti < tR is satisfied. If it is
subsections will introduce DA in uncontrolled fleet risk RF satisfied, DA = DA + 1.Otherwise, return to the third
and control-program fleet risk RF C . process.
To implement the simulation of the fleet risk, the distribution 7) When all the simulations are over, output the value of
parameters of the aircraft must be solved . in this paper, the DA, which is the number of the defect aircraft in the
Weibayes method is adopted to calculate the scale parameter fleet in uncontrolled situation.
η in Weibull distribution. When tR = 50000, simulation result is listed in Fig.4, which
A. Weibayes method for parameter estimation is compared to statistical method given in (7). The calculation
error of the simulation method is no more than 1.9%. So the
Life limited parts for aircraft usually have a longer lifetime. process proposed is effective to calculate the defect aircraft,
Therefore, the number of failures of the target component is which gives the possibility to predict DA in control-program
small during cooperation. The regression method and the fit- fleet risk.
ting method generally cannot obtain good results. In weibayes
method, shape parameter β can be determined through past C. Simulation process for DA in control-program fleet risk
experience, historical failure data, and engineering knowledge As is introduced in chapter 2, risk can be reduced through
of failure physics [14]. In this method, the scale parameter η control program methods. Control-program fleet risk is the
is calculated according to (6) evaluation of the aircraft’s entire life cycle after the control
"N #1 program is formulated. Assume the aircraft will be maintained
X tβ β regularly and the maintenance cycle is tM , after which the
i
η= (6) aircraft can be perfectly repaired. the model proposed in the
r
i=1 a previous part does not apply to this situation. Therefore, the
In equation (6), t is the working time of the part. ra repre- model should be modified.
sents the number of the failed parts examined. N represents Firstly, the maximum number of maintenance nmax is
the number of all the failed parts. η is the maximum likelihood acquired referring to (9).
estimation of the characteristic life.
tR
nmax = de (9)
B. Simulation process for DA in uncontrolled fleet risk tM
According to [3], DA is acquired by statistical method, When the fleet is under operational, the aircraft i will be in
which is shown in (7). F (t) in (7) represents the failure the maintenance cycle n0(i) which is referring to (10)
probability function of the component. In this subsection,
t0(i)
a modified Method based on Monte Carlo Simulation is n0(i) = b c (10)
proposed to calculate the value of DA. tM
In (9) and (10), d·e represents rounding up while b·c repre-
X [F (t + ∆t) − F (t)]
DA = (7) sents rounding down. According to the inscription, if Ti > tM ,
1 − F (t) the aircraft i still operates when it is under maintenance. When
f leet

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.
2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

Defect Airplane Comparison (tR=50000FC)


22

20 Statistical Method
Simulation Method

Number of the Defect Aircraft (DA)


18

16

14

12

10

2
1 2 3 4 5 6 7 8 9 10
Number of the failed parts (ra)

Fig. 4. Comparison Result of the Simulation Method

3) Update the aircraft number i = i + 1.


4) Calculate the initial maintenance cycle n0(i) and set the
current maintenance cycle n = n0(i) .
5) Simulate operation life Ti referring to (8).
6) Operate procedure 7) if n = nmax is satisfied. Other-
wise, operate procedure 8).
7) Examine if Ti satisfies t0(i) − (n − 1)tM < Ti < tM .If
so, update DA by DA = DA + 1 and repeat procedure
3). Otherwise, update t0(i) and the number of the main-
tenance cycle n. Then perform the next simulation in
the next cycle of the aircraft i;
8) For the last maintenance cycle, DA = DA + 1 if Ti <
tem is satisfied. Then update i and perform the procedure
3) for the next aircraft;
9) Accomplish the whole simulation and output the value
of DA.

IV. E XAMPLE OF RISK ASSESSMENT BASED ON OPTIMIZED


TARAM
Door frames of the aircraft will crack due to pressurization.
When the cracks occur, the skin of the aircraft will be
Fig. 3. Simulation method of DA in RF damaged, resulting in the cabin pressure loss, Reduction or
loss of aircraft control will eventually lead to unsafe outcomes
such as crashes and disintegration. TARAM process should
t0(i) − (n − 1)tM ≤ Ti ≤ tM , the aircraft is considered to
be taken to control the risk which is divided into 5 parts:
malfunction. And the last working time tem is referring to (11).
Determination of the non-detected rate, causal chain analysis,
Calculation of the failure rate function, the calculation of the
tem = tR − (nmax − 1)tM (11)
uncontrolled fleet risk, and the calculation of the control-
The modified process for the control-program fleet risk is program fleet risk.
introduced in Fig.5.
1) Enter the number of aircraft N , the retirement life of A. Determination of the non-detected rate
the aircraft tR and the operation time of each aircraft. The non-detected rate N D depends on many aspects such as
The operation time of the aircraft i is t0(i) the characteristic of maintenance scenes, failure attribute, and
2) Enter the shape parameter β of the target failure, number detection techniques. In this paper, the expert system method
of the current failure ra , calculate the scale parameter η is adopted to determine the non-detected rate. The process is
referring to (6). listed in Fig.6

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.
2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

Fig. 6. Process to determine non-detected rate

Fig. 5. Simulation method of N D in RF C Fig. 7. Causal Chain analysis

Nowadays, the main techniques to detect fatigue cracks are Therefore, the conditional probability CP of the aircraft
X-ray detection, ultrasonic testing, and fluorescence detection. can be calculated according to the causal chain analysis.Risk
CP · IR is 6.13 × 10−4 .
P
Fatigue cracks are fuzzy in characteristic and difficult to of unsafe outcomes
identify. In the meantime, the cracks do not have multichannel
C. Calculation of the failure rate function
to share the burden. According to Fig.7, N D = 0.3.
Assume there are 507 aircraft in the fleet, retirement life of
B. Causal Chain analysis which are 70000 fly cycles. The operation time of the aircraft
To perform causal chain analysis, error propagation of the is shown in Fig.8.The generation fatigue crack obeys Weibull
cabin frame crack should be evaluated. According to Fig.7, distribution, According to 1 , the shape parameter of cracks
the propagation path is presented. The transition probability caused by stress corrosion β = 2. Assuming the detected failed
of each state is given and the injury ratio IR of each unsafe aircraft ra = 30, according to (6), the scale parameter η =
outcomes is listed in Tab.II. 1.16 × 105 .

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.
2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

TABLE II (1)Simulation value of DA


I NJURY RATIO LIST 120

Defect Airplane(DA)
unsafe outcome mortality unsafe outcome mortality
100
In-flight breakup 100% Individual fatality 0.1%
crash 98% runway departure 3%
80

60
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
160 Number of simulation iterations
(2)Average of DA
90
140

Defect Airplane(DA)
number of the aircraft

88
120

100 86

80 84
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Number of simulation iterations
60

40 Fig. 9. Simulation result of DA in RF

20
(1)Simulation value of DA
40
0

Defect Airplane(DA)
1 2 3 4 5
30
operation time /FC 104
20

Fig. 8. Current fleet operation status 10

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Therefore, the failure rate function λ is presented in (12) Number of simulation iterations
(2)Average of DA
    22
2 t
Defect Airplane(DA)

λ= × (12)
1.16 × 105 1.16 × 105 20

D. Calculation of uncontrolled fleet risk


18
According to (1) and (6), the individual risk RI =
1.37×−9 , RI < 10−7 . Therefore, for every single aircraft, it 16
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
is safe to operate. Risk evaluation for fleet should be taken for Number of simulation iterations
further research.
According to the simulation process introduced in the previ- Fig. 10. Simulation result of DA in RF C
ous chapter, the result of defect aircraft DA in the uncontrolled
fleet risk RF is shown in Fig.9. The average is shown in Fig.9
and DA = 124.84. Therefore, the result of uncontrolled fleet V. C ONCLUSION
risk is 0.0229 according to (1). Compared to the statistical This paper performs a risk assessment on a simulated fleet
method proposed in [7] whose result is 0.0243, the deviation targeted on fatigue cracks of the cabin door. A control program
is 5.6%. has been formulated and verified. The following work has been
E. Calculation of control-program fleet risk complete.
Control programs can be issued according to airworthiness 1) This paper refined the risk assessment methods based on
regulations, Immediately adopted rules (IAR), Emergency the current risk assessment criteria. A control-program
airworthiness directive et al to improve the safety level of fleet risk assessment model is developed.
the failed component. For worn-out failure parts, regular 2) This paper proposed a simulation method based on the
replacement and repair can reduce risk. Monte Carlo method to calculate the key parameter of
For the failure of fatigue cracks, maintenance measures the fleet risk.
are taken to control the fleet risk. According to the simu- However, this article has some limitations. In the mainte-
lation method introduced, if the maintenance interval tm = nance of the fleet, there are usually multiple failure modes.
11000 (F C), the result of DA is shown in Fig.10. DA = The risk assessment of unsafe outcomes cannot meet all the
17.02. If the passenger value of the aircraft EO = 170, scenarios of continued airworthiness and safety. The multi-
RF C = 0.5315 < 3. Therefore, the control program is objective and multi-fault fleet risk will be studied for further
effective. research.

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.
2020 Asia-Pacific International Symposium on Advanced Reliability and Maintenance Modeling (APARM)

R EFERENCES
[1] ARP, SAE. 5150, Safety Assessment of Transport Airplanes in Com-
mercial Service. Washington, SAE International, 2003.
[2] FAA. 8110.107A Monitor Safety/ analyze data. Washington, Federal
Aviation Administration, 2010.
[3] Transport airplane directorate. PS-ANM-25-05 Transport airplane risk
assessment methodology handbook. Washington, Federal Aviation Ad-
ministration, 2011.
[4] Lee Wen-Kuei. “Risk assessment modeling in aviation safety manage-
ment,” Journal of Air Transport Management, vol. 12 , pp. 267–273,
2006.
[5] Jiang Wen, Xie Chunhe, Wei Boya, Zhou Deyun. “A modified method
for risk evaluation in failure modes and effects analysis of aircraft
turbine rotor blades,” Advances in Mechanical Engineering , vol. 8, pp.
1687814016644579, 2016.
[6] Guo Qiang.“ Novel Approach to Risk Assessment of Aircraft Electrical
Wiring Interconnet System,” Journal of aircraft, vol. 48, pp. 1888–1893,
2011.
[7] M. G. Violette, P. Safarian, N. Han, R. Atmur. “ Transport airplane risk
analysis,” Journal of aircraft 52.2, pp. 395–402, 2015.
[8] Wang Qiqi, Karthik Duraisamy, Juan J Alonso, Gianluca Iaccarino.
“Risk assessment of scramjet unstart using adjoint-based sampling
methods,” AIAA journal, vol.50, pp. 581–592, 2012.
[9] Galileo Tamasi, Micaela Demichela. “Risk assessment techniques for
civil aviation security,” Reliability Engineering & System Safety 96.8,
2011, pp. 892–899.
[10] Wang Peng, Jin Tongdan. “Flight safety risk assessment using Monte
Carlo simulation-A real case study,” 2011 International Conference on
Quality, Reliability, Risk, Maintenance, and Safety Engineering. IEEE,
pp. 633–667, 2011.
[11] Wang Guanru, Sun Youchao, Li Longbiao, Xin Huiqiu, Li Yang.“ Occur-
rence risk analysis process and methods during continuing airworthiness
of civil aircraft,” Civil Aircarft Design & Research, vol.3, pp. 29–32,
2011.
[12] Li Longbiao, Chen Hui, Xu Qinghong, Sun Youchao.“ Risk assessment
methods for civil airplane fleet during continued airworthiness,” Aircraft
Design, vol.2, pp. 62–66, 2015.
[13] Guo Yuanyuan, Li Longbiao, Sun Youchao.“ Failure risk assessment
method of civil aircraft based on Monte Carlo method,” Acta Aeronau-
tica et Astronautica Sinica, vol. 38. pp. 150–158.
[14] AQSIQ, SAC. GB/T 34987-2017 Weibull analysis. Beijing, SAC, 2017.

The 9th Asia-Pacific International Symposium on Advanced Reliability and


Maintenance Modeling (APARM 2020 – Vancouver)

Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on November 05,2020 at 20:53:12 UTC from IEEE Xplore. Restrictions apply.

You might also like