Example of PM optimization

The predominant failure mode of a critical part of a machine is of a degrading nature (erosion, for
instance) and the existing failure data allows you (after a best-of-fit test) to represent its failure
behaviour by a Weibull probability distribution with parameters: location = 0; shape = 2; scale =
3,000 hours. Costs and down time, respectively, are estimated to be 1,000 € and 5 hours per
preventive action (planned) and 3,000 € and 30 hours per corrective action (unplanned). Suppose
you accept the accuracy to be +/- 250 hours. If you elect: a) minimum cost; b) maximum
availability, as decision criterion, what should the periodicity of PM be?

As you can see in the exhibit below, 2.250 hours is the time interval that minimizes cost (98,47
€/100 operating hours) and 1.250 hours is the time interval that maximizes availability. Sensitivity
analysis can be performed in the software and we obtain the lower and upper limits for any %
allowed of increase of the minimum cost or of decrease of the maximum availability.

Curves of cost or availability varying with PM time intervals can be viewed on next page.

Weibull probability distribution Costs Availability

Failure Cumulative
Periodicity Instantaneous PM CM
density probability of
failure rate
Mean life PM CM
of PM function failure €/100 h €/100 h

0 0 0 0 0 - - - - - Parameters
1 250 5,517E-05 0,0069204 5,556E-05 125 406,94 112,84 97,97% 98,87%
2 500 0,0001081 0,0273955 0,0001111 312 213,16 112,84 98,86% 98,87% t0 = 0 hours
3 750 0,0001566 0,0605869 0,0001667 483 152,78 112,84 99,12% 98,87% a= 2
4 1.000 0,0001989 0,1051607 0,0002222 649 125,67 112,84 99,21% 98,87% b= 3000 hours
5 1.250 0,0002335 0,1593763 0,0002778 811 111,75 112,84 99,24% 98,87%
6 1.500 0,0002596 0,2211992 0,0003333 969 104,33 112,84 99,24% 98,87% Scale factor
7 1.750 0,0002767 0,2884274 0,0003889 1.122 100,51 112,84 99,23% 98,87%
8 2.000 0,000285 0,3588196 0,0004444 1.269 98,84 112,84 99,20% 98,87% Increment = 250
9 2.250 0,0002849 0,4302172 0,0005 1.411 98,47 112,84 99,17% 98,87%
10 2.500 0,0002774 0,5006482 0,0005556 1.547 98,93 112,84 99,14% 98,87% Costs
11 2.750 0,0002637 0,5684094 0,0006111 1.676 99,89 112,84 99,11% 98,87%
12 3.000 0,0002453 0,6321206 0,0006667 1.796 101,12 112,84 99,08% 98,87% PM = 1000 €/action
13 3.250 0,0002233 0,6907518 0,0007222 1.909 102,48 112,84 99,05% 98,87% CM = 3000 €/action
14 3.500 0,0001994 0,7436242 0,0007778 2.013 103,87 112,84 99,02% 98,87%
15 3.750 0,0001747 0,7903886 0,0008333 2.109 105,22 112,84 99,00% 98,87% Durations
16 4.000 0,0001502 0,8309867 0,0008889 2.195 106,47 112,84 98,98% 98,87%
17 4.250 0,0001269 0,8656013 0,0009444 2.272 107,61 112,84 98,96% 98,87% PM = 5 hours/action
18 4.500 0,0001054 0,8946008 0,001 2.340 108,61 112,84 98,95% 98,87% CM = 30 hours/action
19 4.750 8,605E-05 0,9184831 0,0010556 2.400 109,48 112,84 98,93% 98,87%
20 5.000 6,909E-05 0,9378235 0,0011111 2.451 110,21 112,84 98,92% 98,87% Conclusion
Mean failure rate = 0,000376 failures/hour
MTTF = 2.659 hours PM recomended
Optimal periodicity in the light of minimum cost = 2.250 hours
Optimal periodicity in the light of maximum availability = 1.250 hours

Sensitivity analysis Lower limit Optimum Upper limit

Tolerable increment of the minimum cost = 5% 1750 2250 3500

Tolerable decrement of the maximum availability = 1% 500 1250 Outside

The optimum time interval to perform PM is obtained by minimizing the following equation:

Where: tp is the time interval between PM actions;

t is the mean life of the components that fail before the tp;
Cc is the cost of a CM action;
Cp is the cost of a PM action;
Ft is the probability of a failure or the probability of a CM occurs;
(1 – Ft) is the probability of a PM action takes place.

F(t)

F(tp)

f(t)

0
F(tp)
0 t tp PM time interval (hours)

The mean lifet of the components that fail before tp is calculated solving the integral:

As an alternative, although less accurate, you can use a numerical method by dividing the interval
(0; tp) into N classes. The mean lifet can now be obtained from the equation:

The next figure shows how Cost varies with t and the optimum PM time interval t*.

Cost

€/hora

Least
cost

0 t* PM time interval (hours)

If you prefer maximum availability as the decision criterion instead, then you solve the next
equation for a maximum, where symbols have the same meaning as above. Besides those, TP is the
duration of a PM action and TC is the duration of a CM action (frequently the relation TC  TP
holds).

The next figure shows how Availability varies with t and the optimum PM time interval t**.

Maximum
availability(
%)

0 t** PM time interval (hours)

In the table above, you find, for instance, the cost 98,47 €/100 hours, by doing the following:

t = {[(250 + 0) / 2 x (0,0069204 – 0)] + … + [(2.250 + 2.000) / 2 x (0,4302172 – 0,3588196)]} /

0,4302172 = 1.411 operating hours

[3.000 x 0,4302172 + 1.000 x (1 – 0,4302172)] / [1.411 x 0,4302172 + 2.250 x (1 – 0,4302172)] x

100 = 98,47 €/100 operating hours

And you find 99,24% of availability this way:

t = {[(250 + 0) / 2 x (0,0069204 – 0)] + … + [(1.250 + 1.000) / 2 x (0,1593763 – 0,1051607)]} /

0,1593763 = 811 operating hours

[811 x 0,1593763 + 1.250 x (1 – 0,1593763)] / {[811 x 0,1593763 + 1.250 x (1 – 0,1593763) + 30 x

0,1593763 + 5 x (1 – 0,1593763]} x 100 = 99,24%

Rui Assis
November/05