You are on page 1of 8

International Journal of Fatigue 114 (2018) 57–64

Contents lists available at ScienceDirect

International Journal of Fatigue


journal homepage: www.elsevier.com/locate/ijfatigue

Failure detection technique under random fatigue loading by machine T


learning and dual sensing on symmetric structure

Dong-Won Jang, Sunghyun Lee, Jong-Won Park, Dong-Cheon Baek
Reliability Assessment Center, Korea Institute of Machinery and Materials, Daejeon 34103, Republic of Korea

A R T I C LE I N FO A B S T R A C T

Keywords: To avoid the sudden failure of mechanical structures under repeated loading and to supplement conventional
Failure detection methods of managing life time, we propose a failure detection technique under random fatigue loading using
Fatigue loading machine learning and dual sensing on a symmetric structure. The state of a shackle, which is used to connect the
Machine learning cargo to the hoist efficiently, under fatigue loading was collected using two strain sensors of a dual system. The
Dual sensing
strains were preprocessed and labeled as normal or abnormal. Logistic regression machine learning was em-
Prognostics and Health Management (PHM)
ployed to determine the decision boundary line. Then, we gathered the decision boundary lines of each ex-
periment for determining the time of failure, and we verified every experiment with the most conservative
decision boundary line. The results indicate that failure was detected before the crack occurred and the time to
notice maintenance could be controlled.

1. Introduction Model-based methods, such as fault tree analysis, particle filter-


based method, and Kalman filter-based method [22–24], are based on
It is very important to avoid the sudden failure of mechanical the fundamental knowledge of physics. It updates the physical model
structures under repeated loading. In general, many structures are de- from the monitoring data and predicts future defects. It can carry out
signed according to a safe-life design concept to avoid sudden failures predictions with a minimal amount of data; however, there is a lack of
[1–4]. This concept requires a large safety factor to account for the established physical models [25].
statistical features of the material properties and the random nature of In this study, we managed the prognosis of a shackle, which is used
the repeated loading. Therefore, even if there is no failure at the end of to connect the cargo to the hoist efficiently, with measured strain data
the specified lifetime, the structure should be replaced. Replacing worn from a dual sensor system. A shackle is an inexpensive component used
out parts always results in loss of resources and time. However, safety within a considerably rough environment. Although the real applica-
and efficiency can be improved if it is possible to consider the statistical tion of prognostic and health management (PHM) is for structures that
features of materials and random loading during prognosis. To consider are expensive and have a small margin of safety, a shackle was used to
these, sensors that can detect the state of materials are used. For ex- validate the algorithm because it has a simple shape and is easy to apply
ample, strain is a time series for a condition monitoring variable [5], fatigue loading to this structure.
and it is measured by a strain sensor. It roughly indicates the loading Using a dual sensor, it is easy to verify the state of the structure by
and the condition of the material. Using such data from the sensor, comparing only two strain data sets [26,27] without the loading history
many researchers have attempted to predict the lifetime of structures by and properties of materials. It can be used under repeated random
utilizing two methods: data-driven method and model-based method loading and the user can verify the abnormality of the sensor.
[5–13]. In this study, we propose a dual sensing system with machine
The data-driven method is based on monitoring data. In this method, learning. The limitation of machine learning is the lack of reliability
the relationship between the lifetime and conditions is based on defect because it depends on data without a physical model. The dual sensing
monitoring data or normal data obtained after processing using computer system guarantees reliability because it does not rely on a single data
techniques such as the Markov model [14,15], support vector machine set; instead, it compares two symmetrical data to diagnose faults under
[16], and neural networks [17–20]; this method does not require a model, fluctuating loads. We combined a dual sensing system and machine
and the application range is relatively wide. However, to improve accuracy, learning to reduce the limitation of machine learning. In particular, two
the amount of data required is large [21]. strains from the dual system were preprocessed and labeled normal or


Corresponding author.
E-mail address: dcbaek@kimm.re.kr (D.-C. Baek).

https://doi.org/10.1016/j.ijfatigue.2018.05.004
Received 26 December 2017; Received in revised form 2 May 2018; Accepted 3 May 2018
Available online 04 May 2018
0142-1123/ © 2018 Elsevier Ltd. All rights reserved.
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

the shackle under repeated loading (Fig. 1). Repeated loading was ap-
plied using a tensile tester (Landmark 100 kN, MTS, USA), and strain
data was collected by a data logger. Repeated loading has sequences
with a frequency of 8 Hz. Data was measured discretely to reduce
computer memory (2, 4, 6, or 10 s in 60 s).
There are standard load spectra for hoists such as ASME HST-4, FEM
9.511, and FEM 9.683 [31–33]. Because shackles are usually used as
connecting components of hoists and a weight, it is better to use the
load spectrum for hoists to evaluate the fatigue life of shackles. How-
ever, the goal of this paper is not to evaluate fatigue life, but to predict
failure using strain signals. Therefore, standard load spectra were only
used as reference, and the load spectra for an accelerated fatigue test
were used.
The details of the loading sequences are presented in Table 1 and
Fig. 2. If the load falls below 0, the shackle and jig can move apart and
cause misalignment and wear. Therefore, the shackle was preloaded
with 980 N and every loads were applied as positive values. Every
loading sequence has the same load valley; however, they have dif-
ferent upper load peaks characterized by the maximum peak load for
each load sequence of 13 kN, 16.7 kN, and 20.6 kN. The load spectra of
three types of loading sequences are presented from Fig. 2(d)–(f). The
Fig. 1. A dual sensing system for a shackle. Two strain gauges were installed on maximum peak loads of 12.40 kN, 13.23 kN, and 13.72 kN are cate-
the outer left and outer right surfaces of the shackle. gorized as Test set I; those of 16.7 kN, Test set II; and those of 20.6 kN,
Test set III. Because it was an accelerated test, the maximum peak load
and some load magnitudes exceed the working load limit of the shackle,
which is 9.8 kN. Furthermore, the maximum peak load in Test set III
exceeds a yield load of 19.6 kN. It included a severe condition to vali-
Table 1
date the algorithm with overloading.
Details of loading sequence.
To distinguish between normal data and abnormal data, the surface
Test set Experiment # Maximum peak load Measuring scheme [s/min] of the shackle was observed with a CCD camera (WISE-MN28V,
[N]
WISECAM, Republic of Korea) to determine whether cracks occurred.
I 1 12,740 2 When a crack was identified in the video image, the time was recorded.
2 4 The crack occurred at a specific location, i.e., the inner curved surface
of the shackle’s leg. Cameras could not observe the inner curved surface
3 13,230 2
of the shackle’s leg perpendicularly owing to geometrical limitations;
4 2
5 2 however, a slightly tilted side view could be observed. The detectable
crack size in the video image was about 1 mm in the titled side view. It
6 13,720 2 occurred randomly in one of the two legs. For the convenience of
processing, we set the strain data from the cracked leg as strain 1, which
II 7 16,660 10
8 10
represents the x-axis in Figs. 4–6. We classified data as normal before
9 6 cracks were detected, and abnormal after cracks were detected. Fig. 3
shows the strain data in one experiment.
III 10 20,580 10 Initially, the two strain plots showed the same trend of variation with
11 10
different magnitudes. Although the shackle is symmetric and the load is
12 10
applied perpendicularly, the two strains have different magnitudes. The
shackle is symmetric in the macro scale; however, it is not perfectly sym-
metric at the micro scale, because material distribution is not uniform and
abnormal. Logistic regression [28–30] machine learning was carried the geometry is also asymmetric. Another reason is that strain gauges were
out to determine the decision boundary line for 12 experimental sets. not perfectly aligned. When the cyclic load was continuously applied, the
Then, we gathered 12 decision boundary lines for determining the time values in one part of the strain plot (black) were reduced, and cracks were
of failure from various experiments and verified every experiment with generated and broken because they originated from the inner side of the
the most conservative decision boundary line. As a result, it was ver- shackle such that the outer surface where the strain gauge is located re-
ified that failure was detected before the crack occurred and the ceived a compressive load. On the other hand, the other strain plot (red) did
prognostic distance could be controlled. not change until the shackle was broken.
If the graph is drawn with one strain data on the x-axis and the other
strain data on the y-axis, the graph is shown in Fig. 4. Since the two
2. Experiments
strain data plots in the initial and intermediate stage showed the same
trend with different sizes, data before 4000 s was plotted on the narrow
We measured the dual strain data of a shackle under repeated
area on the top of the graph. The corresponding color is from red to
loading. Strain gauges were installed on the outer left and outer right of

58
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

Fig. 2. (a–c) Every loading sequence has the same lower load valley; however, they have different upper load peaks that are characterized by the maximum peak load
for which each load sequence is 13, 16.7, and, 20.6 kN. In addition, the load spectra of three types of loading sequence are presented from (d) to (f).

3. Learning and detection process

3.1. Preprocessing and machine learning

As shown in Fig. 4, by using the tendency of data movement as the


cyclic load is applied, machine learning can be used to determine
whether there is a failure; however, before learning, normalization
must be performed. Since the range of the strain depends on the mag-
nitude of the load, learning without normalization will show case-de-
pendent results, and a general model cannot be established.
Therefore, normalization is required for machine learning and for
creating a decision line for all common scenarios. In data preprocessing,
we set the maximum value of strain to 1 and the minimum value of
strain to 0 based on the initial data. In the learning algorithm shown in
Fig. 8, the normalization criteria (Nc) indicates the end of initial data. In
Fig. 5, the initial data is data before 1000 s, and Nc is 1000 s. Further-
more, we carried out data classification: uncracked data (black) and
cracked data (red).
The lifetime before the initiation of crack was 5600 s, which is 80%
of the entire lifetime, and the lifetime after crack initiation was 1450 s,
which is 20% of the entire lifetime. Although the lifetime before
cracking was four times longer, the area of the cracked data was larger
in Fig. 5. This shows that uncracked data converged well near the line
connecting (0,0) and (1,1), and it gradually shifted downwards as cyclic
load was applied. After crack detection, the area of the cracked data
increased in the graph because damage to the shackle increased rapidly.
In the line connecting (0,0) and (1,1), we can observe that the data
formed a curve near the maximum value (1,1) in Fig. 5. When a ductile
Fig. 3. Two strain plots from a dual sensing system in one experiment. material is subjected to a load, the strain is proportional to the load in
the elastic deformation region; however, the strain is not proportional
to the load in the plastic deformation region. In a dual sensing system,
yellow. As the cycle increased, the strain vs. strain data shifted down-
although the structure appears symmetrical, it is locally asymmetric,
ward. This trend was observed in all tests, although each test had dif-
and each part is deformed differently. Even if one part behaves elasti-
ferent loading sequences. It represents the state of the material re-
cally, other symmetric parts behave plastically. Therefore, when plastic
gardless of the magnitude of the load applied. This is an advantage of
deformation occurs on one side, it appears as a curve rather than a
the dual sensing system.

59
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

Fig. 4. Strain vs. strain from the dual sensing system. The color of the data
shows time and its meaning is on the right side of the figure. (For interpretation
of the references to colour in this figure legend, the reader is referred to the web
version of this article.)

straight line near the maximum value. As shown in Fig. 5, the hysteresis
loop is observed. After the maximum load is applied, the data does not
go back along the loading path. However, data does not move along the
linear path as well and its shape is distorted. This implies that there is
plastic deformation, but it is not the main reason for non-linearity on
the top right part of the strain-versus-strain graph. One possible reason
is the non-linear geometrical behavior such as stretching of the legs of
the shackle.
When learning includes the curved section (since the classification
model was characterized by a polynomial), the learning time increased Fig. 6. Graph of normalized and linearized strain from the dual sensing system
and the modification of the decision boundary line becomes compli- (a) and its magnified graph (b).
cated. Therefore, normalization was performed again with the straight-
line section from (0.2, 0.2) to (0.5, 0.5) in Fig. 5 to use only the linear
Data learning, as mentioned earlier in the introduction, was carried
part. When data from the linear part was used, the decision boundary
out using logistic regression.
line obtained through logistic regression was linearly formed such that
Logistic regression is a regression model for classifying data into two
the learning time was shortened and the control of the decision
types [28–30]. With the logistic regression model, the computer
boundary line became simple.

Fig. 5. Graph of the normalized strain from the dual sensing system and one cycle of the magnified graph.

60
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

Fig. 7. Labeled normalized strain 1 (x-axis) versus normalized strain 2 (y-axis) plot and its decision boundary line by logistic regression for 12 experiments.

Fig. 8. Learning algorithm.

attempts to find weights and bias of hypothesis that have the highest typical examples are the mean of squared error and a cross entropy. The
possibility to divide data correctly and this process is called learning. cross entropy is a function to measure the distance of two probability
The hypothesis is a mathematical function. Weights and bias are coef- distributions and is usually used as a cost function in the logistic re-
ficients and the constant of a function. In the learning process, the gression model [34].
computer changes weights and the bias by reducing the cost function; In this paper, a linear function that has one weight and one bias was

61
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

indicated in Fig. 6, because data before cracking was distributed near


the normalized line, it can be observed that even if the decision
boundary lines were distributed over a wide area, the difference in the
estimated lifetime will not be large.
Since data shifted downwards to the normalized data line as the
cycle increased, the decision boundary line closest to the normalized
data line is considered the conservative decision boundary line and the
decision boundary line that is farthest from the normalized data line is
the riskiest decision boundary line, which resulted in wrong decisions
in 11 experiments. Therefore, if the decision boundary line nearest the
normalized line is regarded as the decision criterion, it is possible to
detect an abnormality before cracking in all experiments. Table 2 pre-
sents the prediction of failure for 12 experimental sets, and the sum-
mary of the detection algorithm is shown in Fig. 10. For the 12 ex-
perimental data sets, an average of 84.0 ± 14.2% was detected for the
time the cracks were observed. The most conservative judgment was
54.4%, and the closest judgment was 99.8%.
The 12 experimental data sets used in our algorithm do not cover
Fig. 9. Normalized data line (bold black) and decision boundary lines from every real scenario. Owing to the probability distributions of materials,
each experimental data with machine learning. there may be unobserved experimental data showing a faster fracture
compared to the 12 experimental data sets. Therefore, we propose two
methods to carry out a more conservative decision.
One method is to classify data with an earlier criterion than the
actual crack detection time. In the original algorithm, data were clas-
Table 2 sified into data before crack detection and data after crack detection. In
Ratio of anomaly detection time and crack detection time for each experiment; learning with this classification, the goal of the algorithm is to predict
Δ is the difference between the anomaly detection time and crack detection crack initiation. If data are classified with the criterion which is several
time.
cycles earlier than actual crack detected time, the goal of the algorithm
Experiment # Actual crack Anomaly Anomaly detection Δ [s] is to predict time earlier than crack initiation. For example, when the
detection time detection time time/actual crack classification criterion was set as 500 s subtracted from the crack de-
[s] [s] detection time [%]
tection time, the average warning time was 75.1 ± 21.8% of the crack
1 61,500 55,800 90.7 5700 detection time. It is faster than the results of the original algorithm,
2 44,800 43,900 98.0 900 which is 84.0 ± 14.2%. In addition, when the labeled criterion was set
3 21,200 17,900 84.4 3300 as 1000 s subtracted from the crack detection time, the average warning
4 36,600 31,500 86.1 5100
time was 61.0 ± 25.0%. The details of the values obtained are pre-
5 21,000 17,200 81.9 3800
6 23,900 19,300 80.8 4000 sented in Table 3.
7 6400 6250 97.7 150 Another method for a conservative decision is to modify the deci-
8 5600 4380 78.2 1220 sion boundary line. Because the decision boundary line has two para-
9 8000 7980 99.8 20 meters, a user can modify each parameter to be warned of crack de-
10 3700 2280 61.6 1420
tection faster than the results from the original algorithm. In the
11 3200 3000 93.8 200
12 3200 1740 54.4 1460 original algorithm, the slope of the decision boundary line was 1.086
and the y-axis intercept was −0.427. If the intercept is modified to 90%
and 50% of the original intercept, the average warning times are
82.0 ± 15.2% and 67.8 ± 22.0% of the crack detection time. The
used as the hypothesis. In addition, cross entropy was used as the cost details of the values obtained are presented in Table 4. With these two
function. Further, the hypothesis from machine learning is called a conservative algorithms, users can take a warning and have time to
decision boundary line, because it can classify data based on whether it prepare for maintenance.
is cracked.
As shown in Fig. 6b, the decision boundary line of the straight line
4. Conclusions
distinguished between the cracked and uncracked data very well. In the
same manner, the decision boundary line for the 12 experiments were
For the prediction of failure of a mechanical structure under random
obtained, and the results of the 12 experiments are shown in Fig. 7. The
summary of the learning algorithm is shown in Fig. 8. loading, the failure detection of a shackle with a dual sensing system
was carried out using machine learning under repeated loading. After
preprocessing data for normalization and linearization, it was possible
3.2. Results of detection to classify the cracks before and after crack initiation by linear decision
boundary lines. By selecting the most conservative decision boundary
The decision boundary lines of the 12 learning experiment sets are line, failure prediction was carried out and the result showed
shown in Fig. 9. The decision boundary lines were located below the failure detection at 84.0 ± 14.2% of the time for crack initiation.
normalized data line connecting (0,0) and (1,1). It appears that the Furthermore, we controlled the decision boundary line by two methods.
decision boundary lines were spread over a wide range. However, as In the first method, we classified cracked and uncracked data before the

62
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

Fig. 10. Detection algorithm.

Table 3 Table 4
Ratio of anomaly detection time and crack detection time from modified al- Ratio of anomaly detection time and crack detection time from modified al-
gorithm with the criterion which is several cycles earlier than actual crack gorithm with modified decision boundary; Δ is the difference between the
detected time; Δ is the difference between the anomaly detection time and crack anomaly detection time and crack detection time.
detection time.
Experiment # Conservative method 2–1 Conservative method 2–2 Modified
Experiment # Conservative method 1–1 500 s Conservative method 1–2 1000 s Modified intercept of decision intercept of decision boundary line
earlier classification earlier classification boundary line as 90% of original as 50% of original value
value
Anomaly Anomaly detection Anomaly Anomaly detection
detection time/actual crack detection time/actual crack Anomaly Anomaly detection Anomaly Anomaly detection
time [s] detection time [%] time [s] detection time [%] detection time/actual crack detection time/actual crack
([Δ]) ([Δ]) time [s] detection time [%] time [s] detection time [%]
([Δ]) ([Δ])
1 50,600 82.3 (10900) 33,500 54.5 (28000)
2 42,900 95.8 (1900) 42,100 94.0 (2700) 1 54,900 89.3 (6600) 35,600 57.9 (25900)
3 17,000 80.2 (4200) 16,000 75.5 (5200) 2 43,600 97.3 (1200) 42,100 94.0 (2700)
4 28,200 77.0 (8400) 22,400 61.2 (14200) 3 17,700 83.5 (3500) 16,000 75.5 (5200)
5 15,900 75.7 (5100) 14,500 69.0 (6500) 4 30,900 84.4 (5700) 23,100 63.1 (13500)
6 18,700 78.2 (5200) 17,900 74.9 (6000) 5 16,900 80.5 (4100) 14,700 70.0 (6300)
7 6000 93.8 (400) 5520 86.3 (880) 6 19,100 79.9 (4800) 17,900 74.9 (6000)
8 4080 72.9 (1520) 3780 67.5 (1820) 7 6180 96.6 (220) 5580 87.2 (820)
9 6900 86.3 (1100) 6000 75.0 (2000) 8 4260 76.1 (1340) 3780 67.5 (1820)
10 1870 50.5 (1830) 1560 42.2 (2140) 9 7740 96.8 (260) 6120 76.5 (1880)
11 2940 91.9 (260) 550 17.2 (2650) 10 2160 58.4 (1540) 1560 42.2 (2140)
12 550 17.2 (2650) 480 15.0 (2720) 11 2950 92.2 (250) 2880 90.0 (320)
12 1560 48.8 (1640) 480 15.0 (2720)

actual crack was detected. In the second method, we modified the


equation of the decision boundary line.
The decision boundary lines from multiple experiments have a the strain data. Furthermore, if the distribution of the decision
distribution and the most conservative lines indicate safety; however, boundaries is stochastically predicted, it is expected to be able to make
this also leads to loss of lifetime. Further research is required to make the same decision as B10, which is the life time at which 10% of a
the distribution of the decision boundaries narrower by pre-processing product would have failed.

63
D.-W. Jang et al. International Journal of Fatigue 114 (2018) 57–64

Acknowledgments aerosp conf; 2006. p. 1–19.


[12] Grall A, Bérenguer C, Dieulle L. A condition-based maintenance policy for sto-
chastically deteriorating systems. Reliab Eng Syst Saf 2002;76:167–80.
This study is a part of the research project, “Small and Medium Enterprise [13] Byington C, Roemer M, Kacprzynski G, Galie T. Prognostic enhancements to diag-
Technology Supporting by Reliability Assessment Technology Improvement” nostic systems for improved condition-based maintenance. In: IEEE proc aerosp
(NK208E) and “Development of core machinery technologies for autonomous conf; 2002.
[14] Bellman R. Dynamic programming. Princeton: Princeton University Press; 2010.
operation and manufacturing” (NK210H), which has been supported by a [15] Puterman M. Markov decision processes. New York: Wiley-Interscience; 2005.
grant from National Research Council of Science & Technology under the R& [16] Vapnik V. Statistical learning theory. New York: Wiley; 1998.
D Program of Ministry of Science, ICT and Future Planning. [17] Jackson T, Beale R. Neural computing: an introduction. Bristol: CRC Press; 1990.
[18] Bishop C. Neural networks for pattern recognition. Oxford: Clarendon Press; 1996.
[19] Kasabov N. Foundations of neural networks, fuzzy systems, and knowledge en-
References gineering. Cambridge: MIT Press; 1996.
[20] Haykin S. Neural networks. Upper Saddle River, NJ: Prentice Hall; 1998.
[21] Solomatine D, Ostfeld A. Data-driven modelling: some past experiences and new
[1] Osgood C. Fatigue design. Oxford: Pergamon; 1982.
approaches. J Hydroinf 2008;10:3.
[2] Raithby KD. A comparison of predicted and achieved fatigue lives of aircraft
[22] Fussell JB. Fault tree analysis - state of the art. IEEE Trans Reliab 1974;23(1):51–3.
structures. Technical Report Technical Note Structures-301, Royal Aircraft
[23] Arulampalam M, Maskell S, Gordon N, Clapp T. A tutorial on particle filters for
Establishment, Ministry of Aviation: London; 1961.
online nonlinear/non-Gaussian Bayesian tracking. Proc IEEE Trans Sig
[3] Lundberg B. The quantitative statistical approach to the aircraft fatigue problem. In:
2002;50(2):174–88.
ICAF Symp full scale testing of aircraft structures, Pergamon Press; 1961. p.
[24] Gordon N, Salmond D, Smith A. Novel approach to nonlinear/non-gaussian
393–412.
Bayesian state estimation. IEE Proc Radar Sig Process 1993;140(2):107–13.
[4] Wang H. A survey of maintenance policies of deteriorating systems. Eur J Oper Res
[25] Dragomir O, Gouriveau R, Dragomir F, Minca E, Zerhouni N. Review of prognostic
2002;139:469–89.
problem in condition-based maintenance. In: Proc cont conf ECC; 2009.
[5] Jardine A, Lin D, Banjevic D. A review on machinery diagnostics and prognostics
[26] Baek D, Park J. Development of dual sensor for prognosticating fatigue failure of
implementing condition-based maintenance. Mech Syst Sign Process
mechanical structures. Trans Kor Soc Mech Eng A 2016;40:721–4.
2006;20:1483–510.
[27] Baek D, Bae K, Park J, Choi B. Physical-clone-based prognostics of mechanical
[6] Schwabacher M. A survey of data-driven prognostics. In: Proc. AIAA Infotech@
structures. In: IEEE proc prognostics health manage; 2015.
Aerospace conf; 2005. p. 1–5.
[28] Freedman D. Statistical models. Cambridge: Cambridge University Press; 2009. p.
[7] Vachtsevanos G. Intelligent fault diagnosis and prognosis for engineering systems
128.
methods and case studies. Hoboken: Wiley; 2007.
[29] Walker S, Duncan D. Estimation of the probability of an event as a function of
[8] Luo J, Namburu M, Pattipati K, Qiao L, Kawamoto M, Chigusa S. Model-based
several independent variables. Biometrika 1967;54:167.
prognostic techniques [maintenance applications]. In: Proc IEEE syst readiness
[30] Cox D. The regression analysis of binary sequences. J Roy Stat Soc B
technol conf (AUTOTESTCON); 2003. p. 330–40.
1958;20:215–42.
[9] Smith G, Schroeder J, Navarro S, Haldeman D. Development of a prognostics and
[31] Performance Standard for Overhead Electric Wire Rope Hoists, ASME HST-4; 2016.
health management capability for the Joint Strike Fighter. In: Proc IEEE syst
[32] Series Lifting Equipment – Rules for the Classification of Mechanisms, FEM 9.511;
readiness technol conf (AUTOTESTCON); 1998. p. 676–82.
1986.
[10] Hess A, Calvello G, Frith P. Challenges, issues, and lessons learned chasing the “Big
[33] Series Lifting Equipment - Selection of Lifting and Travel Motors, FEM 9.683; 1995.
P”: real predictive prognostics – part 1. In: IEEE proc aerosp conf; 2005. p. 3610–9.
[34] Murphy KP. Machine learning: a probabilistic perspective. Cambridge: MIT press;
[11] Hess A, Calvello G, Frith P, Engel SJ, Hoitsma D. Challenges, issues, and lessons
2012.
learned chasing the “Big P”: Real predictive prognostics – part 2. In: Proc IEEE

64

You might also like