Professional Documents
Culture Documents
Replacement Decisions
Andrew K S Jardine
CBM Laboratory
Department of Mechanical & Industrial Engineering
University of Toronto
Canada
jardine@mie.utoronto.ca
August 2006
www.ipamc.org
Making Systems more Reliable through
Component Replacement
www.ipamc.org
Andrew Jardine, CBM Lab 2
“Inspection Decisions including Condition-Based Maintenance”
Weibull Distribution
β -1
β ⎛t−γ ⎞ ⎛ ⎛ t − γ ⎞ β⎞
f (t ) = ⎜ ⎟ exp⎜⎜ − ⎜ ⎟ ⎟⎟
η⎝ η ⎠ ⎝ ⎝ η ⎠ ⎠
⎛ ⎛ t − γ ⎞ β⎞
R(t ) = exp⎜⎜ − ⎜ ⎟ ⎟⎟
⎝ ⎝ η ⎠ ⎠
β: Shape parameter
η: Scale parameter
γ: Location parameter
www.ipamc.org
Andrew Jardine, CBM Lab 3
“Inspection Decisions including Condition-Based Maintenance”
Shape Parameter, β
ß = 0.5
ß = 5.0 γ=0
ß = 3.44
ß = 2.5
f(t)
ß = 1.0
0
0 www.ipamc.org
t
Andrew Jardine, CBM Lab 4
“Inspection Decisions including Condition-Based Maintenance”
Shape Parameter, β
Hazard rate, h (t)
β = 2.5
β = 1.0
β = 0.5
? t
www.ipamc.org
Andrew Jardine, CBM Lab 5
“Inspection Decisions including Condition-Based Maintenance”
Commercially Available Software
for Weibull Analysis
• OREST (www.banak-inc.com)
• Weibull ++ (weibull.reliasoft.com)
• RelCode (www.albanyint.com.au))
• M-Analyst (www.m-tech.co.za)
Etc.
www.ipamc.org
Andrew Jardine, CBM Lab 6
“Inspection Decisions including Condition-Based Maintenance”
System Hazard Function
Equipment Life Periods
Infant Useful Life Wearout
Mortality
Failure
Rate,
Overall Life
Risk, or Characteristic Curve
Hazard
function
Stress Related
Failures
Quality Wearout
Failures Failures
Time
www.ipamc.org
Andrew Jardine, CBM Lab 7
“Inspection Decisions including Condition-Based Maintenance”
For Preventive Replacement
Cost minimization
Check two conditions:
1. Total cost of a failure replacement is
greater than total cost of a preventive
replacement
2. Wear-out effect occurring
www.ipamc.org
Andrew Jardine, CBM Lab 8
“Inspection Decisions including Condition-Based Maintenance”
For Preventive Replacement
Availability maximization/downtime
minimization
Check two conditions:
1. Total outage (downtime) of a failure
replacement is greater than total outage
(downtime) of a preventive replacement
2. Wear-out effect occurring
www.ipamc.org
Andrew Jardine, CBM Lab 9
“Inspection Decisions including Condition-Based Maintenance”
RCM Methodology Logic
www.ipamc.org
Andrew Jardine, CBM Lab 10
“Inspection Decisions including Condition-Based Maintenance”
Optimize the plan
Time-Based Discard
• Operating hours
• Calendar time
• Cycles
– Operating
– Launch
www.ipamc.org
Andrew Jardine, CBM Lab 11
“Inspection Decisions including Condition-Based Maintenance”
Preventive Replacement Cost Conflicts
Optimal Replacement Time
Total Cost/Week
C(tp)
$/Week
Failure Replacement
Cost/Week
Preventive Replacement
Cost/Week
tp
Optimal t p
www.ipamc.org
Andrew Jardine, CBM Lab 12
“Inspection Decisions including Condition-Based Maintenance”
Constant Interval Replacement Policy
Cf Cf
New
Item Cp Cp Cp Cp
tp tp tp tp t
www.ipamc.org
Andrew Jardine, CBM Lab 13
“Inspection Decisions including Condition-Based Maintenance”
Construction of Model
Cp: total cost of a preventive replacement
[labour, parts, outage cost, etc.]
tp
One cycle
Cp Cf Cf Cp Cp
New
Item
tp tp tp
www.ipamc.org
Andrew Jardine, CBM Lab 16
“Inspection Decisions including Condition-Based Maintenance”
Optimal Preventive Replacement Age of
Equipment Subject to Breakdown
Failure Failure
replacement replacement
Preventive Preventive
replacement replacement
tp tp
0 Time
www.ipamc.org
Andrew Jardine, CBM Lab 17
“Inspection Decisions including Condition-Based Maintenance”
Determination of Optimal Preventive
Replacement Age
Construction of model:
0 tp 0 M(tp)
GOOD FAILED
CYCLE CYCLE
www.ipamc.org
Andrew Jardine, CBM Lab 18
“Inspection Decisions including Condition-Based Maintenance”
Determination of Optimal Preventive
Replacement Age
Construction of model:
C(tp) = total cost / unit time
when preventive replacement occur at
age tp.
Cp * R(tp) + Cf * [ 1 – R(tp) ]
C(tp) =
tp * R(tp) + M(tp) * [1 – R(tp)] www.ipamc.org
Andrew Jardine, CBM Lab 19
“Inspection Decisions including Condition-Based Maintenance”
Note:
Age-based policy is cheaper than the interval
policy but disadvantage is:
- need to reschedule.
www.ipamc.org
Andrew Jardine, CBM Lab 20
“Inspection Decisions including Condition-Based Maintenance”
Sugar Refinery Centrifuge Case
Wet Sugar
Sugar Refinery
Centrifuge
Dry Sugar
36 Problems Top
6 Analyzed 5
Months Data
www.ipamc.org
Andrew Jardine, CBM Lab 21
“Inspection Decisions including Condition-Based Maintenance”
Failure frequency: cloth interval
CLASS CUMULATIVE CLASS CUMULATIVE
INTERVAL FREQUENCY RELATIVE INTERVAL FREQUENCY RELATIVE
(weeks) FREQUENCY (%) (weeks) FREQUENCY (%)
0 < 2 24 10.5 26 < 28 4 86.9
2 < 4 36 26.2 28 < 30 1 87.3
4 < 6 27 38.0 30 < 32 4 89.1
6 < 8 23 48.0 32 < 34 4 90.8
8 < 10 15 54.6 34 < 36 5 93.1
10 < 12 9 58.5 36 < 38 2 93.9
12 < 14 12 63.8 38 < 40 2 94.8
14 < 16 11 68.6 40 < 42 2 95.6
16 < 18 13 74.2 42 < 44 2 96.5
18 < 20 4 76.0 44 < 46 2 97.4
20 < 22 12 81.2 50 < 52 4 99.1
22 < 24 5 83.4 56 < 58 1 99.6
24 < 26 4 85.2 76 < 78 1 100.0
TOTAL: 229
www.ipamc.org
Andrew Jardine, CBM Lab 22
“Inspection Decisions including Condition-Based Maintenance”
Parameter Estimation for
Cloth Replacement
Estimation point Cloth Replacement 229
1
13 13
0
̭
β =1
η estimator
Perpendicular
̭
η=13
13
www.ipamc.org
Andrew Jardine, CBM Lab 23
“Inspection Decisions including Condition-Based Maintenance”
Bearing Replacement
Historical Data
12 25 9 13 19
www.ipamc.org
Andrew Jardine, CBM Lab 24
“Inspection Decisions including Condition-Based Maintenance”
Median Ranks
When only a few failure observations are available
(say ≤ 20) use is made of median rank tables.
1 2 3 4 5 6 7 8 9 10 11 12
1 50 29.289 20.630 15.910 12.945 10.910 9.428 8.300 7.412 6.697 6.107 5.613
2 70.711 50.000 38.573 31.381 26.445 22.849 20.113 17.962 16.226 14.796 13.598
3 79.370 61.427 50.000 42.141 36.412 32.052 28.624 25.857 23.578 21.669
4 84.090 68.619 57.859 50.000 44.015 39.308 35.510 32.390 29.758
5 87.055 73.555 63.588 55.984 50.000 45.169 41.189 37.853
6 89.090 77.151 67.948 60.691 54.811 50.000 45.951
7 90.572 79.887 71.376 64.490 58.811 54.049
8 91.700 82.018 74.142 67.620 62.147
9 92.587 83.774 76.421 70.242
10 93.303 85.204 78.331
11 93.893 86.402
12 94.387
www.ipamc.org
Andrew Jardine, CBM Lab 25
“Inspection Decisions including Condition-Based Maintenance”
Median Ranks
Example: Bearing failures times (in weeks):
12, 25, 9, 13, 19.
From median rank tables:
1st failure time 13.0% 9 weeks
2nd failure time 31.5% 12 weeks
3rd failure time 50.0% 13 weeks
4th failure time 68.6% 19 weeks
5th failure time 87.1% 25 weeks
By using the median rank for the second ordered observation where we are
estimating that 31.5% of the population will have failed. There is a 50%
probability that the true % of failures will be below and above 31.5%. That is
half the time we’ll be underestimating it and half the time we’ll be
overestimating it.
www.ipamc.org
Andrew Jardine, CBM Lab 26
“Inspection Decisions including Condition-Based Maintenance”
Failure Distribution
Bearing Failure Distribution
0.06
f(t) 0.05
0.04
0.03
0.02
0.01
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Time
www.ipamc.org
Andrew Jardine, CBM Lab 27
“Inspection Decisions including Condition-Based Maintenance”
The Best Time
• Risk curve
• Economics
(Cf & Cp)
• Blend to
establish the
optimal tp
www.ipamc.org
Andrew Jardine, CBM Lab 28
“Inspection Decisions including Condition-Based Maintenance”
PUMP FAILURE DATA
RUNNING SUSPENSION
TIME TO OR
FAILURE CENSORED
(MONTHS) TIME
3
6 6
9
s S S F F S S F S
× × × × × × × × ×
3F + 6S
Failure Suspension (or
censored
observation)
www.ipamc.org
Andrew Jardine, CBM Lab 31
“Inspection Decisions including Condition-Based Maintenance”
Suspensions (Sometimes termed censored data)
45
f(t) 40
F(t) 1.2 h(t)
1.2
35
1 1
30
0.8 0.8
25
0.6 0.6
20
15 0.4 0.4
10
0.2 0.2
5
0 0 0
1 2 3
3 2 1
∫0
t
− r (t ) dt h(t)
f(t) = h(t)[1-F(t)] F (t ) = 1 − e
When dealing with grouped multiply censored data we proceed as above.
www.ipamc.org
Andrew Jardine, CBM Lab 32
“Inspection Decisions including Condition-Based Maintenance”
WATER PUMP FAILURE
TIME TO FAILURE NUMBER OF NUMBER OF
NOTE: Suspension (MILES X 103) FAILURES SUSPENSIONS
means that when the 0 < 5 0 1
data was collected, 5 < 10 2 4
the water pump was 10 < 15 3 3
still operational. For 15 < 20 2 3
example, the 4 20 < 25 1 1
suspensions in the 25 < 30 3 1
class 5000 - 10,000 30 < 35 3 0
35 < 40 1 3
miles means that 4
40 < 45 1 7
pumps had not failed 45 < 50 0 2
and had been in 50 < 55 1 4
operation for 55 < 60 1 7
between 5000 and 60 < 65 0 6
10,000 miles. 65 < 70 0 3
70 < 75 2 1
75 < 80 0 1
80 < 85 0 1
www.ipamc.org
Andrew Jardine, CBM Lab 33
“Inspection Decisions including Condition-Based Maintenance”
Water Pump 68
Estimation point 1.5
85000 94000
̭
β =1.5
η estimator
Perpendicular
̭
η=94000 miles
www.ipamc.org
13
Andrew Jardine, CBM Lab 34
“Inspection Decisions including Condition-Based Maintenance”
Life-time distribution of water pumps
MEAN TIME
TO FAILURE
MILES x 103
0 50 100 150 200 250
MEAN = 85,000 MILES www.ipamc.org
Andrew Jardine, CBM Lab 35
“Inspection Decisions including Condition-Based Maintenance”
Analysis of Censored Data
When censoring takes place then the value of F(t) which is required for Weibull
plotting of the failure data is obtained via the cumulative failure rate as
illustrated in the following table:
Sugar Feed Failures and Censorings
Class S t +δt + S t −δt Instantaneous Failure rate F (T ) = 1 − e − Σr ( t )
Weeks F C S 2 Observed h(t) Cumulative
0< 1 9 5 89 82.0 0.110 0.110 0.104
1< 2 16 1 75 66.5 0.241 0.351 0.296
2< 3 9 2 58 52.5 0.171 0.522 0.407 F = Frequency of Failure
3< 4 7 2 47 42.5 0.165 0.687 0.497 C = Censoring Frequency
4< 5 2 5 38 34.5 0.058 0.745 0.525 S = Survivors at
5< 6 2 12 31 24.0 0.083 0.828 0.563 Commencement of Interval
r(t) = f/|(St-δt + St+δt)/2|
6< 7 3 0 17 15.5 0.194 1.022 0.640
7< 8 2 1 14 12.5 0.160 1.182 0.693
8< 9 2 0 11 10.0 0.200 1.382 0.749
9 < 10 0 2 9 8.0 0.000 1.382 0.749
10 < 11 0 0 7 7.0 0.000 1.382 0.749
11 < 12 1 1 7 6.0 0.167 1.549 0.788
12 < 13 0 0 5 5.0 0.000 1.549 0.788
13 < 14 1 1 5 4.0 0.250 1.799 0.835
14 < 15 1 2 3 1.5 0.667 2.466 0.915
Σ=55 Σ=34
ΣΣ=89 www.ipamc.org
Andrew Jardine, CBM Lab 36
“Inspection Decisions including Condition-Based Maintenance”
Sugar Feed 89
Estimation point Censored Data 0.80
7.0 6.60
̭
β =0.8
η estimator
Perpendicular
̭
η=6.60 weeks
www.ipamc.org
13
Andrew Jardine, CBM Lab 37
“Inspection Decisions including Condition-Based Maintenance”
T-33 Silver Star Aircraft
www.ipamc.org
Andrew Jardine, CBM Lab 38
“Inspection Decisions including Condition-Based Maintenance”
T-33 Silver Star Aircraft
The T33 aircraft engine is supplied with fuel provided by two fuel pumps (upper and lower).
The fuel system design is such that either pump can provide the necessary fuel pressure and
quantity to operate the engine satisfactorily. That is, the system is redundant and the failure of
a pump is not a catastrophic event.
The decision to be arrived at is: Should the pump be removed after “x” hours and overhauled
and relifed, or should we repair/overhaul it after failure only?
Failure Data
Collected over a 2-year period. Censored items represent a “snapshot” of all pumps still
operating successfully on one specific day.
Interval Failures Censored Items
(Hours) Upper Lower Upper Lower
0 – 200 1 2 7 5
200 – 400 5 1 6 5
400 – 600 10 1 5 1
600 – 800 4 1 4 10
800 – 1000 1 1 6 3
1000 – 1200 6 1 9 3
1200 – 1400 2 1 10 6
1400 – 1600 2 1 0 4
1600 – 1800 4 2 0 4 www.ipamc.org
Andrew Jardine, CBM Lab 39
“Inspection Decisions including Condition-Based Maintenance”
T-33 Silver Star Aircraft
Pump Failure Data
Class Interval Failures Censored
(Hours) Observations
0 200 1 7
200 400 5 6
400 600 10 5
600 800 4 4
800 1000 1 6
1000 1200 6 9
1200 1400 2 10
1400 1600 2 0
1600 1800 4 0 Analysis of Pump Failure Data
Class F C h(t) Σh(t) F(t)
0 200 1 7 .01282 .0128 .0127
200 400 5 6 .07288 .0858 .0822
400 600 10 5 .18018 .2660 .2336
600 800 4 4 .0908 .3569 .3002
800 1000 1 6 .0274 .3843 .3191
1000 1200 6 9 .2353 .6196 .4618
1200 1400 2 10 .0833 .7863 .5445
1400 1600 2 0 .4000 1.1863 .6947
1600 1800 4 0 1.000 2.1863 .8877
̭
0
β =2.25
Perpendicular
η estimator
̭
η=1320 hours
www.ipamc.org
13
Andrew Jardine, CBM Lab 41
“Inspection Decisions including Condition-Based Maintenance”
Order Number (A 2.7)
Time Failure /
(hours) Suspension
544 F
663 F
802 S
827 S Sample
897 F
size n = 10
914 F
939 S
1084 F Source: Handling Ungrouped Censored
Data,Table11.13 in Reliability in Engineering
1099 F
Design, K.C. Kapur and L. Lamberson, Wiley,
1202 S 1977 www.ipamc.org
Andrew Jardine, CBM Lab 42
“Inspection Decisions including Condition-Based Maintenance”
Order Number
www.ipamc.org
Andrew Jardine, CBM Lab 43
“Inspection Decisions including Condition-Based Maintenance”
Order Number
Now create a new table giving order number for each
failure and associated median rank
Time Order Number Median Rank
544 1 0.067
663 2 0.163
897 2+1.29=3.29 0.288
914 3.29+1.29=4.58 0.411
1084 4.58+1.6=6.18 0.565
1099 6.18+1.6=7.78 0.719
I = [(10 + 1) – 2] / (1 + 6 ) = 1.29 www.ipamc.org
Andrew Jardine, CBM Lab 44
“Inspection Decisions including Condition-Based Maintenance”
Order Number
Note:
We continue with same increment until another
suspended item is encountered.
www.ipamc.org
Andrew Jardine, CBM Lab 46
“Inspection Decisions including Condition-Based Maintenance”
Steering Clutch, L.H.
Steering Clutch, L.H.
(from a group of 6 CAT D10 Dozers)
New Today
Failure intervals (F) 7979 h, 2027 h
Suspension interval (S) 9671 h
$ 5640
www.ipamc.org
Andrew Jardine, CBM Lab 49
“Inspection Decisions including Condition-Based Maintenance”
Checking the fit of a distribution:
Use of the K-S test.( Section A2.6)
We have obtained 5 component failure times, in hours.
Ordered they are:
1,5,6,8 and10
The mean and standard deviation of these times are 6 and
3.4.
Can we accept the hypothesis that the component fails
according to a normal distribution with mean = 6 hours and
standard deviation = 3.4 hours?
www.ipamc.org
Andrew Jardine, CBM Lab 50
“Inspection Decisions including Condition-Based Maintenance”
Kolmogorov – Smirnov
Goodness-Of-Fit Test
Need to check:
^
|F(ti) – F(ti)|
Max of =d
^
|F(ti) – F(ti -1)|
www.ipamc.org
Andrew Jardine, CBM Lab 51
“Inspection Decisions including Condition-Based Maintenance”
Kolmogorov – Smirnov
Goodness-Of-Fit Test
^ ^ )|
|F(ti)-F(ti -1)| F(t) |F(ti) -F(t i
1
0.9
F(t) 0.8
0.7
0.6 ^
0.5 F(ti)
0.4
0.3
0.2 ^
0.1 F(ti-1)
0
ti -1 ti t
www.ipamc.org
Andrew Jardine, CBM Lab 52
“Inspection Decisions including Condition-Based Maintenance”
OREST Outputs
www.ipamc.org
Andrew Jardine, CBM Lab 53
“Inspection Decisions including Condition-Based Maintenance”
CP Rail
www.ipamc.org
Andrew Jardine, CBM Lab 54
“Inspection Decisions including Condition-Based Maintenance”
A SUCCESS STORY IN THE ANALYSIS OF
FAILURE DATA: CP RAIL
COMPONENT POWER TRACTION TURBO-
ASSEMBLY MOTOR CHARGER
ALTERNATIVES • REPLACE
WITH NEW MAJOR SEND TO GM
• WASH & OVERHAUL FOR REBUILD
WEAR
• OVERHAUL
12 25 9 13 19 Non-committal
System
9 12 13 19 25 Happy
System
Sad
25 19 13 12 9 System
Air heater
Soot deposits
steam
Boiler
www.ipamc.org
Andrew Jardine, CBM Lab 57
“Inspection Decisions including Condition-Based Maintenance”
200
150
$/Week
$/Week
100
50
c(t) = btm
0
Time (weeks) 1 2 3 4 Weeks
$/Week
$/Week
20
10 c(t) = A – Be-kt
c(t) = 1.5 + 3.2t
0 2 4 6 8 10 Time (weeks)
Time (weeks)
0.8
0.6
0.4
0.2
tr
0
time
1
Total cost
0.8
0.4
0.2
Replacement cost
(cleaning cost)
0
tr
Optimal tr
tr – interval between replacement. www.ipamc.org
Andrew Jardine, CBM Lab 61
“Inspection Decisions including Condition-Based Maintenance”
Model Development
Cr c(t) : operating cost at time t
Cost Cr : total cost of replacement
tr : interval between
c(t) component replacement
t
tr
C(tr) : total cost / unit time
∫
tr
c ( t ) dt + C r
C (tr ) = 0
tr www.ipamc.org
Andrew Jardine, CBM Lab 62
“Inspection Decisions including Condition-Based Maintenance”
Model Development
Optimal value of tr at C´(tr) = 0.
c(tr) = C(tr)
www.ipamc.org
Andrew Jardine, CBM Lab 63
“Inspection Decisions including Condition-Based Maintenance”
Example
Air filter change:
Cr = $20 Km / month driven = 2000 Gas cost = $0.75/litre
Assuming 15 km/litre when new, then month 1 cost = $100.00
Next period O and M cost, c(t), is greater than average cost to date, C(t).
Cost ($)/month
120 111.75
112.5
111.67
0 1 2 3 4
t (month)
2C
Note: If c(t) = a + bt, then tr * =
b
r
www.ipamc.org
Andrew Jardine, CBM Lab 65
“Inspection Decisions including Condition-Based Maintenance”
REVIEW
www.ipamc.org
Andrew Jardine, CBM Lab 66
“Inspection Decisions including Condition-Based Maintenance”
Reliability Improvement Through
Preventive Replacement
• Weibull Analysis: Shape parameter β
• Need β > 1
• Constant Interval Policy
• Age-Based Policy
www.ipamc.org
Andrew Jardine, CBM Lab 67
“Inspection Decisions including Condition-Based Maintenance”
Problems using OREST
1) A sugar refinery centrifuge is a complex
machine composed of many parts that are
subject to sudden failure. A particular
component, the plough-setting blade, is
considered to be a candidate for preventative
replacement, and you are required to
determine an optimal replacement policy. The
policy you are to consider is sometimes
termed a block replacement or constant
intervals, say tp, with failure replacements
taking place when necessary. Determine the
optimal policy so that total cost per unit time
is minimized given the following data:
www.ipamc.org
Andrew Jardine, CBM Lab 68
“Inspection Decisions including Condition-Based Maintenance”
Problems using OREST
a) The labor and material cost associated with a
preventive or failure replacement is $200.
b) The value of production losses associated
with a preventative replacement is $100,
while that for a failure replacement is $700.
c) The failure distribution of the setting blade
can be described adequately by a Weibull
distribution with mean = 150 hours and
standard deviation = 15 hours.
Also indicate the approximate cost of your
optimal policy as a percentage of a replace-
only-on-failure policy.
www.ipamc.org
Andrew Jardine, CBM Lab 69
“Inspection Decisions including Condition-Based Maintenance”
Problems using OREST
2) Heavy-duty bearings in a steel forging plant have failed
after the number of weeks of operation provided in Table
1.
a) Use OREST to estimate the following Weibull
parameters: β, η, and mean life.
b) The cost of preventative replacement is $100 and the
cost of failure replacement is $1000. Determine the
optimal replacement policy.
Table 1: Bearing Failure Times
Age at Failure (weeks)
8
12
14
16
24
One unfailed at 24 weeks
www.ipamc.org
Andrew Jardine, CBM Lab 70
“Inspection Decisions including Condition-Based Maintenance”
Problems using OREST
c) The forge is cleaned and serviced once per week.
Preventative replacement of the bearing can be
carried out as part of this maintenance activity. At
what age should the bearing be replaced, given that,
in addition to direct-cost considerations, there is a
safety argument for minimizing failure. Support your
conclusions by giving the cost and the proportions of
failure replacements for some alternative policies.
d) There are two similar forging plants and each works
for 50 weeks per year. Estimate the number of
replacement parts required per year if the policy is
preventative replacement at age 6 weeks. How many
failure replacements will occur per year (steady-state
average) under this policy?
www.ipamc.org
Andrew Jardine, CBM Lab 71
“Inspection Decisions including Condition-Based Maintenance”