Академический Документы
Профессиональный Документы
Культура Документы
Carlos Recalde
2
Maintenance Strategies
Maintenance
Strategies
CBM, RCM
Equipment
Flowcharts
Prevent
Failure
Mode
3
Action Plan - Strategy
Strategy Equipment
Game Prevent
Failure
Equipment
Mode
Evaluation
Maintenance Failure Keep
Function
Behavior
Action plan -
maintenance
activities
Selection Random
Application Failure
4
Failure Evaluation
Asset classes
Equipment, etc.
Prevent
Failure Human life
Evaluation
Mode Human
Failure development
Keep
Function
5
Failure Evaluation
Evaluation
Failure
Characteristics: Reliability
Failure cause Analysis
Failure
interval
6
Selection
8
Selection - Bathtub Curve
Early Failure
9
Selection - Bathtub Curve
Degradation failure
• Wear-out factors.
10
Selection - Bathtub Curve
Random failure
• External shocks.
11
Random Failure Premise
Function Obtained =
50%
12
Random Failure Premise
13
Random Failure Premise
• Reliability Patterns
4% 7%
2% 14%
5% 68%
14
Random Failure Premise
• F + E = 82 %
16
Application
17
Reliability Analysis - Evaluation of Failure
No Probability
Reliability values
Approaches
Evaluation Qualitative Classify failure
Failure FMEA
Failures
Quantitative
Characteristics:
Failure cause Probability
or period - values
What?
Failure interval
- When? Non parametric analysis
Parametric or Distribution
based method
18
Parametric or Distribution Based Method
• Best results.
• More insightful for prediction.
• More information and important contribution to
performance improvement.
• Based on knowledge of the reliability
distribution.
• Common probability distributions: normal,
exponential, Binomial, Poisson, Rayleigh and
Weibull.
19
Weibull Distribution
20
Weibull Distribution
21
Weibull Parameters
22
Random Failure Influence
Maintenance
activities
Failure cause or
Random period - What?
Failure
Failure interval
- When?
MTTF and
MTBF
23
Weibull Parameters Sequence
Iterative evaluation
Maintenance
activities Failure Mode
Failure cause or Weibull parameters
period behavior sequence
Beta behavior
Failure interval
behavior
Eta behavior
24
METHODS
25
Failure Definition
26
Weibull Parameters Evaluation
27
Standard Method
28
Failure Mode
Failure Mode
Event that causes
FMEA Component
functional failure
EN ISO
14224, Effect of failure Equipment
2016
29
Failure Mode
30
Life Data
31
Life Data
Aging units
32
Time to Failure
33
Median Ranks
34
Iterative Evaluation
Correlation Method
Equipment
Adjusted Weibull Save
Failure Mode parameters
Ranks
t1- x1 TTF1 Beta
t2- x2 TTF2 Benard´s Eta
Median
. . r⌃2
. . Ranks
tn- xn TTFn
35
Weibull Parameter Sequence
36
RESULTS
37
38
Eta
50.00
0.00
100.00
150.00
200.00
250.00
300.00
350.00
9/14/2017 7:15
9/19/2017 16:30
9/22/2017 10:40
9/27/2017 15:00
9/27/2017 19:45
10/2/2017 14:30
10/3/2017 8:30
Early Failure
10/5/2017 8:00
10/18/2017 18:30
10/23/2017 17:10
10/23/2017 18:00
10/24/2017 7:30
10/25/2017 7:00
11/6/2017 16:00
11/7/2017 13:30
11/7/2017 16:00
11/7/2017 18:30
11/10/2017 18:00
11/14/2017 8:00
11/15/2017 5:30
11/15/2017 14:15
11/16/2017 19:00
11/21/2017 10:00
11/22/2017 20:15
11/23/2017 14:00
11/27/2017 17:45
11/30/2017 17:00
Failure Mode F3 - Equipment C
12/4/2017 14:00
12/12/2017 14:00
12/14/2017 15:10
12/21/2017 8:00
1/9/2018 5:00
1/10/2018 14:15
1/16/2018 6:45
1/18/2018 9:10
β<1. Task frequency may be according to eta.
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
Beta
Eta
Beta
• Average r⌃2 = 97%. Steady behavior of beta, always
39
Eta
0.00
2.00
4.00
6.00
8.00
10.00
12.00
9/26/2017 8:30
10/3/2017 11:00
10/4/2017 7:00
10/20/2017
10/20/2017 15:00
10/24/2017 11:00
10/24/2017 14:00
10/29/2017 14:00
11/10/2017 9:00
Degrading Failure
11/16/2017 15:15
11/20/2017 13:30
11/24/2017 11:00
11/27/2017 11:00
12/1/2017 9:00
12/5/2017 8:02
12/7/2017 10:00
12/12/2017 13:00
12/20/2017 12:00
1/16/2018 7:30
1/22/2018 15:00
Failure Mode F1 - Equipment A
1/29/2018 15:30
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
4.50
5.00
Beta
β>1. Task frequency may be according to eta.
Eta
Beta
• Mean r⌃2 = 89%. Changing behavior of beta, always
40
Eta
0.00
1.00
2.00
3.00
4.00
5.00
6.00
10/23/2017 9:00
10/25/2017 10:30
11/6/2017 18:30
11/7/2017 5:00
11/7/2017 16:00
11/8/2017 5:00
11/10/2017 7:30
11/15/2017 10:40
11/16/2017 9:00
11/20/2017 9:10
11/20/2017 18:00
11/22/2017 16:00
11/23/2017 8:40
Random Failure
11/30/2017 9:10
11/30/2017 14:00
12/6/2017 10:40
12/11/2017 16:10
12/12/2017 6:00
12/18/2017 9:00
12/19/2017 14:20
12/20/2017 19:20
12/21/2017 12:30
12/26/2017 6:53
1/2/2018 13:30
1/8/2018 6:20
1/10/2018 14:00
1/11/2018 9:45
1/11/2018 11:40
Failure Mode F2 - Equipment B
1/12/2018 16:00
1/16/2018 5:30
1/22/2018 13:30
1/26/2018 20:00
0.00
0.20
0.40
0.60
0.80
1.00
1.20
Beta
Eta
Beta
tends to 1 . Task frequency may be according to eta.
• Mean r⌃2 = 94%. Increasing behavior of beta, which
CONCLUSIONS
41
Weibull Parameter Sequence
100
80
Frequency
60
40
20
0
0.3 0.5 0.8 0.9 1 1.1 1.2 1.5 2 3 4 5
Beta
42
Beta Data Histogram
• 17% 0.9<=beta<=1.1
43
Weibull Parameter Sequence
44
QUESTIONS?
45
Contact Information
Carlos Recalde
46
Page 1 of 10
Abstract
This work evaluates the behavior of Weibull parameter sequences of failure modes, in order to give some
insights to improve selection and application of maintenance activities. These activities are developed by
most of the maintenance strategies, which sometimes assume random failures. This premise is inferred
because of the easy applying of basic reliability terms like Mean Time Between Failures (MTBF) and
Mean Time to Failure (MTTF), and because of the difficult evaluation of failure causes. To avoid this
issue, it is proposed the application of a distribution-based method like the two-parameter Weibull
distribution, which is the best for modeling each period of the bathtub curve. These periods are related to
failure causes like early, degradation and random failure. The Weibull parameter sequence was
developed with an iterative evaluation automated in a spreadsheet application. Understanding failure
behavior according to the sequence of Weibull parameters may help to improve decision making and
focus all maintenance efforts.
Page 2 of 10
INTRODUCTION
Maintenance Strategies
Most maintenance strategies develop action plans or maintenance activities, which work to keep on
demand the function of equipment and prevent failure modes. Maintenance strategies like Reliability
Centered Maintenance (RCM) and Condition Based Maintenance (CBM) evaluate failure mode applying
process flowcharts or logic trees, which help to define the best action plan. This action plan relates to
strategy in the context of the interaction of two non-cooperative entities. These entities can be described
1
as the equipment and the maintenance organization. According to Maschler et al , strategy is a plan of
action for an interdependence situation of two entities, which is also known as a game, where the first
entity profits by breaking any rule and the second monitors the behavior of the first. Given this, the
equipment may win by “failing” and the maintenance organization may monitor the equipment behavior.
Then, maintenance organization may develop maintenance activities according to some evaluation of the
equipment failure behavior. However, sometimes the selection and the application of the action plans or
maintenance activities are influenced by random failures.
Failure Evaluation
Failure evaluation has become a focus of attention because of the failure effect in human life. Since
human development is directly depended on the function of systems, equipments and components, which
may relate to assets classes like: real estate and facilities, plant and production, mobile assets,
2
infrastructure and information technology . Then, when failure happens, it may affect the customer needs
and the operation requirements. And sometimes catastrophic failures may affect safety and the
environment causing human and financial loss.
Failure evaluation requires some reliability analysis. In addition, it may be developed according to some
characteristics of failure like the failure cause and the failure interval. Respecting to failure cause, bathtub
curve may give some ideas.
Selection
The selection of maintenance activities or the insights of “what” to do may relate to the failure cause,
3
which is important to understand to implement action plans . The cause of failure may relate to the
4
periods of bathtub curve, which represents the whole life of a population of products . It comprises three
failure periods: early, degradation, and random. First, early failure period refers to errors at some
5
equipment life cycle stages like: design, manufacture, installation, operation or maintenance. Abernethy
states that it leads to suspect of inadequate burn in, production problems, mis assembly, quality control
4
problems, and overhaul problems. Then, Yang suggests it is caused by mayor latent defects as
manufacturing process variations, material flaws or customer misuse. Second, degradation failure period
Page 3 of 10
refers to wear-out factors. It leads to suspect of low cycle fatigue, corrosion, erosion, stress corrosion or
5
material properties . To minimize this failure effect, scheduled replacement or Preventive Maintenance
4
(PM) is often required . And third, random failure period refers to external shocks, which leads to suspect
5
of human errors, abusive events, failure due to nature or foreign object damage . Therefore, maintenance
activities selection or the insights of “what” to do may require the evaluation of failure cause according to
bathtub curve periods.
Figure #1
9
Reliability Patterns
Page 4 of 10
Application
The application of maintenance activities or the task frequency may relate to the failure interval. Defining
“when” to develop the action plan may be a difficult job. However, sometimes random failure also
influences this decision. Given this, the common used failure interval is the fixed interval of age, which
12
works best to avoid fixed or constant failure rate. Benbow and Broome point out that when the
underlying failure causes are constant failure rates then the easy approach to evaluate these failures are
9
basic reliability terms like MTTF or MTBF. However, Smith and Hinchcliffe suggest that this approach is
9
not really a valued and useful technique for selecting the failure interval. Moreover, Smith and Hinchcliffe
suggest that the lack of statistical data may give operation and maintenance experience the chance to
9
guess the task frequency . Given this, a fix interval is not the best approach because it only leads to
avoid random failure.
Reliability Analysis
The evaluation of failure requires some reliability analysis. Reliability analysis has a quantitative and a
5
qualitative approach . Qualitative analysis like Failure Mode and Effect Analysis (FMEA) is important and
5
recommended . It works to classify failures. It does not require the assignment of probability values and it
13
is the first step for quantitative analysis, which does require probability values . Then, according to
14
Regattieri , quantitative approach is categorized in nonparametric analysis and parametric or
distribution-based method. The first requires methods like Product Limit Estimator or Kaplan-Meier and
the second requires methods like Maximum Likelihood Estimator (MLE) or Least Square. Given this,
failure evaluation may apply both reliability approaches.
Distribution-based Method
Regarding to parametric or distribution-based method, it may give best results. Parametric approach is
4
more insightful when it refers to prediction . It gives more information and can provide important
14
contribution to performance improvement of industrial and non-industrial complex systems . Moreover, it
is based on the knowledge of the reliability or probability distribution, which refers to some of the common
distributions like: Normal, Exponential, Poisson, Binomial, Rayleigh and Weibull. Respect to Weibull
5
distribution, it includes exactly or approximately all the reliability distributions stated before . In addition, it
4
best models each period of the bathtub curve . Its defining parameters are beta (β) and eta (η). Eta
relates to the interval of failure and to the age at which 63% of equipment or components will fail. Beta
relates to the failure cause or the failure period. When β<1, it describes early failures; when β>1, it
describes degradation failures; and when β=1, it describes random failures and Weibull changes to
Exponential distribution, moreover, then eta relates to MTTF or MTBF. Therefore, Weibull parameters
beta may relate to the failure period or the failure cause and eta to the task frequency or the failure
interval.
Page 5 of 10
Finally, random failure sometimes influences some decision like the selection and the application of
maintenance activities. This influence provides a bias selection of maintenance activities because it
rejects the evaluation of a different failure period like early or degradation. Random failure provides some
insights to the question of “what” to do, because it is usually caused by external shocks. Moreover, the
easy approach to evaluate these failures are basic reliability terms like MTBF or MTTF, which provide
some answers to the question of “when” to do. According to this, the improvement of the selection and
the application of maintenance activities require an understanding of the failure cause and the failure
14
interval behavior. Regattieri states that failure modeling must be an iterative procedure applied at
different instants of the system service. Therefore, it is proposed an iterative evaluation automated in a
spreadsheet application to develop the Weibull parameter sequences of failure modes.
METHODS
Failure Definition
Failure evaluation requires a clear definition of failure. Then, failure is the inability of equipment to
3 6
perform a required function . Moreover, failure is described as a state variable . Therefore, for this work it
is used a binary state like: function obtained (FO) and function not obtained (FNO).
Failure Mode
Failure Mode is important for failure classification. According to its definition it may be categorized in the
FMEA and the ISO 14224 approach. Given this, FMEA approach suggests failure mode is an event that
7
causes a functional failure, which may apply at component level . However, ISO 14224 approach
suggests that failure mode is an effect of failure, which applies at equipment level and as alternative at
3
component level . Both approaches may work good for failure classification. But when failure mode
5
classification is good the correlation coefficient r⌃2 shows a good fit . In addition, Weibull should evaluate
Page 6 of 10
5
a single failure mode at a time . Therefore, the more possible likely failure modes should be described
and applied. For this work it was used the ISO 14224 approach.
Life Data
Censored life data or suspensions may relate to failures of different failure modes. If the evaluation does
14
not have concern about censor data, then a significant error is introduced . Moreover, suspensions
cannot be ignored and must be included in the evaluation because suspensions increase eta and have
5
little effect on beta . For this work, all failure data is evaluated in sequence according to the equipment
and the failure mode.
Life data also may relate to aging units like hours, miles, cycles, etc., which sometimes are not available.
When the best aging parameter is not available and only calendar time data is available then a measure
5
of goodness of fit r⌃2 will determine if the Weibull evaluation is useful .
Figure #2
TTF Evaluation
Median Ranks
5
Adjusted Rank and Benard’s Median Rank are evaluated according to Abernethy . In Table #1, both
formulas require the total number of failures (N). When a new failure joins to the evaluation then the total
number of failure increases and adjusted and median ranks need to be re-evaluated.
Page 7 of 10
Table #1
5
Median Ranks Formulas
Table #2
Sequence of Weibull Parameters
RESULTS
Early Failure
According to Graphic #1, the evaluation of failure mode F3 of equipment C shows a steady behavior of
beta, which is always less to 1. Therefore, maintenance activities should aim to avoid errors at some
equipment stage like design, manufacture, installation, operation or maintenance, and these activities
should apply according to failure interval eta. The average of r⌃2 = 97%.
Eta Eta
10.00
12.00
0.00
2.00
4.00
6.00
8.00
50.00
100.00
150.00
200.00
250.00
300.00
350.00
0.00
9/26/2017 8:30 9/14/2017 7:15
Degrading Failure
9/19/2017 16:30
10/3/2017 11:00 9/22/2017 10:40
10/4/2017 7:00 9/27/2017 15:00
9/27/2017 19:45
10/20/2017 10/2/2017 14:30
10/3/2017 8:30
10/20/2017 15:00 10/5/2017 8:00
10/18/2017 18:30
10/24/2017 11:00 10/23/2017 17:10
10/24/2017 14:00 10/23/2017 18:00
10/24/2017 7:30
10/29/2017 14:00 10/25/2017 7:00
11/6/2017 16:00
Graphic #2
Graphic #1
11/30/2017 17:00
12/4/2017 14:00
12/12/2017 13:00 12/12/2017 14:00
12/20/2017 12:00 12/14/2017 15:10
12/21/2017 8:00
1/16/2018 7:30 Eta 1/9/2018 5:00
Beta
Eta
1/10/2018 14:15
Beta
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
4.50
5.00
Beta
Beta
factors. A further analysis should identify the degrading part and the replace interval should be related to
According to Graphic #2, the evaluation of failure mode F1 of equipment A shows a changeable behavior
of beta, which is always greater to 1. Therefore, maintenance activities should aim to avoid wear-out
Page 8 of 10
Page 9 of 10
Random Failure
According to Graphic #3, the evaluation of failure mode F2 of equipment B shows an increasing behavior
of beta, which tends to 1. Therefore, maintenance activities should aim to avoid external shocks and
these activities should apply according to failure interval eta. The average of r⌃2 = 94%.
Eta
Failure Mode F2 - Equipment B
Beta
6.00 1.20
5.00 1.00
4.00 0.80
Beta
Eta
3.00 0.60
2.00 0.40
1.00 0.20
0.00 0.00
11/7/2017 5:00
11/8/2017 5:00
1/2/2018 13:30
1/8/2018 6:20
1/11/2018 9:45
1/16/2018 5:30
10/23/2017 9:00
10/25/2017 10:30
11/6/2017 18:30
11/7/2017 16:00
11/10/2017 7:30
11/15/2017 10:40
11/16/2017 9:00
11/20/2017 9:10
11/20/2017 18:00
11/22/2017 16:00
11/23/2017 8:40
11/30/2017 9:10
11/30/2017 14:00
12/6/2017 10:40
12/11/2017 16:10
12/12/2017 6:00
12/18/2017 9:00
12/19/2017 14:20
12/20/2017 19:20
12/21/2017 12:30
12/26/2017 6:53
1/10/2018 14:00
1/11/2018 11:40
1/12/2018 16:00
1/22/2018 13:30
1/26/2018 20:00
Graphic #3
Weibull Parameters Sequence
CONCLUSIONS
The evaluation of failure according to the Weibull parameter sequences of different equipment and failure
modes shows a changing behavior of the failure cause beta and the failure interval eta. In the case of
beta, it does not show a steady behavior with tendency to 1 for all events. By contrast, according to a
histogram of 260 beta data results, see Graphic #4, It suggests that only a 17% of failures are close to 1
or 0.9<=beta<=1.1, and 65% of failures with beta <=0.8 and 18% of failures with beta >=1.2. The first is
close to pattern E, and de second is close to pattern F. Both are the same than the sum of patterns E and
F, 82% of failures. Given this, the influence of random failure or exponential distribution avoids the
evaluation of a different failure cause or period like early or degradation and influences the application of
fix intervals. Therefore, an iterative evaluation of failure to develop the sequence of Weibull parameters of
failure modes allows understanding failure behavior according to the behavior of failure cause and failure
interval providing insights to the selection and application of maintenance activities. Finally, the Weibull
parameter sequences may provide more information, which may help to improve decision making and
focus all maintenance efforts.
Page 10 of 10
150
Frequency
100
50
0
0.3 0.5 0.8 0.9 1 1.1 1.2 1.5 2 3 4 5
Beta
Graphic #4
Beta Data Histogram
REFERENCES
1. Maschler M, Solan E, Zamir S. Game Theory. Cambridge: Cambridge University Press. 2013.
2. Campbell J, Jardine A, McGlynn J. Assets Management Excellence: Optimizing Equipment Life-
Cycle Decisions. Boca Raton, Florida: Taylor & Francis Group. 2011.
3. ISO 14224. Petroleum, Petrochemical and Gas Industries. Collection and Exchange of Reliability
and Maintenance Data for Equipment. BSI Standard Limited. 2016.
4. Yang G. Life Cycle Reliability Engineering. New Jersey: John Wiley & Sons. 2007
5. Abernethy R. The New Weibull Handbook. Fifth edition. Florida: Robert Abernethy. 2006.
6. Rausand M, Høyland A. System Reliability Theory: Models, Statistical Methods, and Applications.
New Jersey: John Wiley & Sons. 2004.
7. Moubray J. RCM II-Reliability Centered Maintenance. Oxford: Butterworth-Heinemann. 1997.
8. Nowlan FS, Heap HF. Reliability-Centered Maintenance. Washington, DC: National Technical
Information Service, Report No. AD/A066-579. December 29, 1978.
9. Smith AM, Hinchcliffe GR. RCM-Gateway to World Class Maintenance. Burlington, MA: Elsevier
Butterworth-Heinemann. 2004.
10. MIL-HDBK-217. Military Handbook. Reliability Prediction of Electronic Equipment. 1991.
11. OREDA. Offshore Reliability Data. Norway: OREDA Participants. 2002.
12. Benbow DW, Broome HW. The Certified Reliability Engineering Handbook. Milwaukee: ASQ
Quality Press. 2009.
13. Karimi R, Rasmussen N, Wolf L. Qualitative and Quantitative Reliability Analysis of Safety
Systems. PhD diss. Massachusetts Institute of Technology. Massachusetts, US. 1980.
14. Regattieri A. Reliability Evaluation of Manufacturing Systems: Methods and Applications,
Manufacturing System, Dr. Fayez Abdul Aziz (Ed.), ISBN: 978-953-51-0530-5. 2012. InTech,
Available from: http://www.intechopen.com/books/manufacturing-system/reliability-evaluation-of-
manufacturing-systems-methods-and-applications
Keywords: Maintenance strategy, failure evaluation, bathtub curve, random failure, reliability analysis,
Distribution-based method, Weibull distribution, life data, failure mode, Weibull parameter sequences.