Calculating The Probability of Failure On Demand (PFD) of Complex Structures by Means of Markov Models

Calculating the "Probability of Failure on Demand" (PFD) of complex
structures by means of Markov Models

Andreas Hildebrandt
Pepperl+Fuchs GmbH
KonigsbergerAllee 87
D-68307 Mannheim
Abstract - For the assessment of the "Safety Integrity
Level" (SIL) in accordance with the standard EN 61508 it
is among other things also necessary to calculate the
"Probability of Failure on Demand" (PFD) of a safety
related function. Thereto a set of equations is given in the
standard mentioned above. If no appropriate formula is
available, the calculation of the PFD can be done by
means of a so called Markov Model. Especially for
heterogeneous systems the Markov Model is an
appropriate method to do the calculation of the PFD
without the need of using a special formula.
without voting) one of these equations can be used.
Unfortunately some variants of structures are missing in

the standard. In this case the calculation of the PFD can
be done by means of a so called Markov Model.
Especially for heterogeneous systems the Markov Model
is an appropriate method to do the calculation of the PFD
without the need of using a special formula.
Even though the understanding of a Markov Model is
not very difficult in principle the definition of the various
states and the corresponding transition probabilities can
be a little bit tricky. Mainly the consideration of the
common cause failures leads to additional difficulties and
will be discussed in detail.
To show how to define the various states of a Markov

Model and how to derive the appropriate transition
probabilities from given device specifications, the PFD of
a one channel system is calculated by using a Markov
Model. It is shown that the result of the Markov Model is in
accordance with the formula given in standard EN 61508.
Within the scope of this paper a Markov Model for a

heterogeneous 1 out of 2 - System is presented and the
results of this model are compared with the results
derived by a formula given in the EN 61508.
In a second step a Markov Model for a 1 out of 2 System (10o2) is presented. For multi channel systems
the common cause failures have to be considered. It is
The following chapter will start with a one channel

system to explain the handling of a Markov Model in
principle. Later on a heterogeneous two channel system
will be discussed.
shown that this leads to additional states in the Markov

Model because the return to the initial state is different for
common cause failures and failures of individual
channels.
Last but not least an example for a heterogeneous
10o2 - system is presented.
Finally several calculation results produced with the

Markov Model mentioned above are compared with those
derived from the formulas given in the standard. This is
done by choosing the same failure rates for both channels
so that the system becomes homogenous. For dangerous
undetected failures (ADU) the results of the Markov Model
are equal to those derived from the formula given in the
standard. For dangerous detected failures (ADD) the
results of the Markov Model are only half the values of the
formula. This is due to a simplification of the formula
which leads to an inaccuracy that is usually negligible.
Index Terms
11.
For a one channel system the calculation of the PFD is

usually not done by means of a Markov Model but with a
formula given in the standard EN 61508. Nevertheless a
one channel system is a good example to explain the
application of a Markov Model. In addition the verification
of the Markov Model is quite simple in this case because
the formula of the 1001 - Structure is well understood and
can be easily used as a benchmark.
EN 61508, PFD, Probability of Failure
on Demand, Heterogeneous Structure, Homogenous
Structure, Markov Model, Common Cause Failure,

Dangerous Detected Failure, Dangerous Undetected
Failure, 1 oo2 - System, CARMS.
I.
CREATING AMARKOVMODEL FORA SIMPLE

ONE CHANNEL STRUCTURE (1001-SYSTEM)
Starting with a properly working system the first step is

to determine the different kinds of failures which can
happen. Depending on the failure, the system will move
from the initial state to a different state. For a one channel
system there are only two different kinds of failures
possible:
INTRODUCTION
For the assessment of the "Safety Integrity Level" (SIL)

in accordance with the standard EN 61508 it is among
other things also necessary to calculate the "Probability of
Failure on Demand" (PFD) of a safety related function.
Thereto a set of equations is given in the standard
mentioned above. Depending on the structure of the
safety related loop (single channel or multi channel with or
1) Detected Fault: The fault will be detected by

periodical diagnostic. After detecting the fault it takes the
mean time to repair (MTTR) to restore the system. Due to
the repair the system will go back to the initial state # 0.
1
2) Undetected Fault: The fault will be detected by

the proof test. As long as no proof test is performed the
system is down. Therefore the mean down time will be
half the proof test time T1 plus the mean time to repair
MTTR if an undetected failure occurs (in other words, the
mean down time is T, / 2 + MTTR). After repair the
system goes back to the initial state #0.
PFD
PFD Mark
MTTR
DU
1 + MTTR +
)
2
DD
MTTR
In other words in case of low failure rates (A t << 1) the

result of the Markov Model turns into the equation given in
the standard EN 61508.
iiundetect.
111. CREATING A MARKOV MODEL FOR A

HETEROGENOUS TWO CHANNEL
STRUCTURE (1 002-SYSTEM)
In case of a SIL 3 requirement it is often possible to use

two SIL 2 devices in parallel. In order to minimize the
probability for a so called common cause failure it is a
good idea to use different kinds of devices for each
channel. In this case the failure rates for channel #1 and
channel #2 are different in general. Unfortunately there
are no formulas for heterogeneous systems available up
to now. Therefore the calculation of the PFD has to be
done with the help of alternative methods. A commonly
used method to calculate the PFD of complex structures
is the Markov Model.
As mentioned above there is no need to do the

calculation of the PFD for a one channel system by
means of a Markov Model but this is going to change if
the PFD of a multi channel system must be calculated. In
case of a homogenous system there are still some
formulas available as long as the number of channels is
not too high. However for heterogeneous structures there
are no formulas given in the standard IEC 61508 even if it
is only a two channel system. Therefore the use of a
Markov Model is advisable.
Fig. 1
The corresponding probability matrix is given as:
X DD
1 XDD DD
-X DU
-
equation above passes over to:
lool
Markov
!DU
T, M
DU 2
) DD
If the numerator is significantly smaller than one the
The transition probability from the initial state (state #0)

to the state #1 and state #2 is given by ADD and ADU
respectively. The probability for the way back is the
reciprocal of the mean down time. In case of a detected
fault the mean down time is MTTR. For an undetected
fault it is (Ti / 2 + MTTR). This leads to the Markov
Diagram shown in Figure 1.
-6
MTTR
1
0.5 T1 + MTTR
X DU
1
MTTR
0
1-
The evaluation of the appropriate Markov Model can be

done analogous to the considerations described in
chapter 11. The main difference to a one channel system is
the fact that for multi channel systems the so called
"common cause failure" has to be taken into account.
Moreover the Mean down time of the system in case of
two independent undetected dangerous faults is no longer
(Ti / 2 + MTTR) but (Ti / 3 + MTTR). For an undetected
common cause failure the system behaves like a single
channel system and as a result of this, the mean down
time is (Ti / 2 + MTTR) as ever. This leads to the Markov
Diagram shown in Fig. 2 (for an enlarged figure see
appendix A).
0.5 T1 + MTTR
Note: The sum of probabilities in each line must be one.

Therefore the elements of the diagonal were calculated
accordingly.
The steady state probability of each state can easily be

calculated by matrix multiplication:
.r*[T ]n
=limLPi0oo1 J
The evaluation of this formula is usually done with the

help of an appropriate software tool like MATHCAD,
MAPLE, CARMS [1] or something like this.
The corresponding probability matrix is given in

Appendix B. The Markov Model for the heterogeneous
system can also be used for a homogeneous structure by
equating the failure rates from channel #1 with the failure
rates of channel #2. In this case the results from the
Markov Model are comparable to the results of the
formula for the two channel system given in the standard
EN 61508.
A different method to solve the Markov Model is a set of

equations [2]. This leads to the following result for the
steady state of the one channel system (1001) mentioned
above:
becomes substantial. From there the actual value for the

PFD under best case condition is about 1.1*10 which is
approximately half the calculated value.
/
/,U CCE
For the heterogeneous 10o2 system the PFD

calculated by means of the Markov - Model shown in
>Fig. 2 comes to:
2 undet.
detect.
C2
PFDheerge=526.-106 ;--5. 1O06
L~~~~~~~~~~~~~~~~htroe
, S
o.kM
As expected this value is within the limits for the PFD

given by the best case and worst case condition, that
~~~~D
~~~~means:
C1dX tt.
PFDbestcase <PFDheterogen <PFDworstcase
V.
NOMENCLATURE
SIL
Safety integrity level.

Safety instrumented function.
SIS Safety instrumented system.
SFF Safe failure fraction
PFD Probability of failure on demand
PDH probability of dangerous failure per hour (1/h)
failure rate of safe failures (1 h)
Xs
failure rate of dangerous failures (1 h)
XD
XDD failure rate of detected dangerous failures (1 h)
XDU failure rate of undetected dangerous failures (1 h)
tCE channel equivalent mean down time (h)
MTTR mean time to restoration (h)
TI
proof-test interval (h)
p3 fraction of undetected failures that have a
ACl undet. Cldedett.0/
'2/ o.k.SIF
Fig. 2
IV. EXAMPLE: HETEROGENEOUS TWO CHANNEL

LEVEL CONTROL SYSTEM
Assume a SIL 3 level control system with two different

SIL 2 transmitters. The failure rates of the transmitter #1
and #2 are given as:
ADUl = 21 07 1 /h
ADU2 = 7-109 1/h
ADD1 = 7-1 0 1 /h
ADD2 = 210-5 1/h
common cause
CI
cause
Channel #1
C2 Channel #2
UCCF undetected common cause failure
(Transmitter #2 has a good diagnostic coverage.

Therefore the failure rate for the undetected faults
becomes lower and the failure rate for the detected faults
increases)
VI. REFERENCES
The factors D and fD for the common cause failures are

assumed to be 1% because the diversity is quite good
due to the heterogeneous system design. The proof test
interval T1 is defined as 1 year and the mean time to
repair MTTR as 8 hours. (Default value of the EN 61508)
[1]
Jan Pukite, Paul Pukite, "Modelling for Reliability

IEEE Press, ISBN 0-7803-3482-5
Analysis",
http://umn.edu/-puk/carms.html
[2]
Obviously the PFD of the system must be within the

values for a homogeneous system built up with two
identical transmitters of type #1 or type #2 respectively.
That means it is quite easy to calculate the limit values for
the best case and the worst case by using the formula for
the homogeneous 10o2 - system given in the standard
EN 61508. For the example mentioned above the results
William M. Goble, "Control Systems Safety

Evaluation and Reliability", ISBN 1-55617-636-8,
www.isa.org
VI. VITA
The author graduated from University of Kaiserslautern,
are:
PFD vot= 1 05 10~

O.l*l PFDbestcase =1,97 10-6 2 10-6
fraction of detected failures that have a common
Germany in 1990 and gets the PhD Degree in 1996.

Sinceasthen
he is with
Pepperl+Fuchs
GmbH,
Mannheim,
,first
a design
engineer,
later on head
of the
product
worstcase
release department and now leader of the department

"training and committee worK'. He is author of several
previous papers and is a member of the DKE Standards
subcommittee UK921.3 He is chairman of the ZVEI
working group EMC.
Remark: Due to a simplification in the formula given in

the standard EN 61508 for the 10oo2 system, the influence
of the detected failures on the PFD is twice as high as it is
in reality. Therefore the calculated PFD is too pessimistic
if the contribution of the detected failures to the PFD
Appendix A
Cl o.k.
~~C1 udetec.
etect.4
Cl~~
Clo.k
Cl
DD
Markov -
C1
undetect
detecfrht. rqn
s102sse
4~~~~~~~U
Appendix B
T1 +2 MTTR
MTTR
T1 +2 MTTR
MTTR
T1 + 3
(1-f3)XDUI
0
(1
D)XDDI
0
(1
)XDU2
0
(1
DD)XDDI
0
0
0
0
X
0DU2
DDI
XDD2
~XDU I
2
MTTR
o0 2
DD2
XDUl
DDI
DU2
XDU
MTTR
T1 +2
MTTR
10
MTTR
2
T1 +2 MTTR
MTTR
MTTR
0
0
0
T1 +2
MTTR
0
Probability Matrix
Note: The sum of the probabilities in each line must be one. Therefore the elements of the diagonal have to be calculated
accordingly (elements of the diagonal are still missing in the matrix above).
E. g. for row number k that means:
9
Pk,k
1 ZPk i
i=5

Calculating The Probability of Failure On Demand (PFD) of Complex Structures by Means of Markov Models

Uploaded by

Copyright:

Available Formats

Calculating The Probability of Failure On Demand (PFD) of Complex Structures by Means of Markov Models

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Calculating The Probability of Failure On Demand (PFD) of Complex Structures by Means of Markov Models

Uploaded by

Copyright:

Available Formats

Calculating the "Probability of Failure on Demand" (PFD) of complex

structures by means of Markov Models

without voting) one of these equations can be used.

Unfortunately some variants of structures are missing in

To show how to define the various states of a Markov

Within the scope of this paper a Markov Model for a

The following chapter will start with a one channel

shown that this leads to additional states in the Markov

Last but not least an example for a heterogeneous

10o2 - system is presented.

Finally several calculation results produced with the

For a one channel system the calculation of the PFD is

EN 61508, PFD, Probability of Failure

on Demand, Heterogeneous Structure, Homogenous

Structure, Markov Model, Common Cause Failure,

CREATING AMARKOVMODEL FORA SIMPLE

Starting with a properly working system the first step is

For the assessment of the "Safety Integrity Level" (SIL)

1) Detected Fault: The fault will be detected by

2) Undetected Fault: The fault will be detected by

In other words in case of low failure rates (A t << 1) the

111. CREATING A MARKOV MODEL FOR A

In case of a SIL 3 requirement it is often possible to use

As mentioned above there is no need to do the

equation above passes over to:

The transition probability from the initial state (state #0)

The evaluation of the appropriate Markov Model can be

Note: The sum of probabilities in each line must be one.

The steady state probability of each state can easily be

The evaluation of this formula is usually done with the

The corresponding probability matrix is given in

A different method to solve the Markov Model is a set of

becomes substantial. From there the actual value for the

For the heterogeneous 10o2 system the PFD

PFDheerge=526.-106 ;--5. 1O06

As expected this value is within the limits for the PFD

PFDbestcase <PFDheterogen <PFDworstcase

Safety integrity level.

ACl undet. Cldedett.0/

IV. EXAMPLE: HETEROGENEOUS TWO CHANNEL

Assume a SIL 3 level control system with two different

(Transmitter #2 has a good diagnostic coverage.

The factors D and fD for the common cause failures are

Jan Pukite, Paul Pukite, "Modelling for Reliability

Obviously the PFD of the system must be within the

William M. Goble, "Control Systems Safety

PFD vot= 1 05 10~

fraction of detected failures that have a common

Germany in 1990 and gets the PhD Degree in 1996.

release department and now leader of the department

Remark: Due to a simplification in the formula given in

You might also like