Sei sulla pagina 1di 27

Reliability Prediction

A Quest for Reliable Parameters


By
Yair Shai

1
Goals
• Compare the MTBCF & MTTCF
parameters in view of complex
systems engineering.
• Failure repair policy as the backbone
for realistic MTBCF calculation.
• Motivation for modification of the
technical specification requirements.

2
Promo :
Description of Parameters
t1 t2 t3 t4 t5 ....... time

 Failure Event of an Item r =Number of Failures

Repairable Items:
t
i
i
Mean Time Between Failures =
r
Non Repairable Items:
t
i
i
Mean Time To Failure =
3 r
MTBF = MTTF ??
An assumption:
Failed item returns to “As Good As
New” status after repair or renewal.
note: Time To Repair is not considered.
UP TIME

DOWN

4
Critical Failures
Moving towards System Design

A System Failure resulting in (temporary or


permanent) Mission Termination.

X
COMPUTER
SUBSYSTEM
A simple
configuration of
parallel hot

X
COMPUTER Redundancy.

A Failure: any computer failure


A Critical Failure: two computers failed
5
Critical Failures
A clue for Design Architecture
MTBCF

Mean Time Between Critical Failures


MTTCF

Mean Time To Critical Failure


SAME? Remember the assumptions
Determining the failure repair policy: COLD REPAIR

No time for repair actions during the mission


6
Functional System Design
Switch control
UNIT A ANTENA
CPU
POWER
UNIT B ANTENA
SUPPLY 4 CHANNEL
RECEVER sw
CONTROLER UNIT C ANTENA
CPU
POWER
UNIT D ANTENA
SUPPLY
POWER
SUPPLY
2/4

Operational Demand: At least two receiver units


and one antenna should work to operate the system.
7
From System Design to
Reliability Model
A ANT
CPU PS1
B ANT
CONT PS2 sw
x
CPU PS1 x C ANT
x
D ANT

Is this a Critical Failure ? 2/4

Serial model : Rs = R1x R2


Parallel model : Rs = 1- (1-R1)x(1-R2)
K out of N model : Rs = Binomial Solution
8
From RBD Logic Diagram
to Reliability Function
Simple mathematical manipulation:
Rsys(t)= f( serial / parallel / K out of N)
Classic parameter evaluation:
 WARNING !!!

MTBF   Rsys (t )dt


MTTF Is this realistic
MTBCF
MTTCF 0 ?

After[ each
S.Zacks,
repair
Springer-Verlag
of a critical failure
1991,-Introduction
The whole system
To
returns
Reliability
to status “As Good
Analysis, ParAs
3.5]
New”.
9
MTBCF vs. MTTCF
A New Interpretation
First
Common practice interpretation:

MTBCF = MTTCF = MTTCFF


Each repair “Resets” the time count to idle status
(or) Each failure is the first failure.
Realistic interpretation:

MTBCF = MTTCF
Only failed Items which cause the failure are repaired
to idle. All other components keep on aging.

10
Presentation I
Simple 3 aging components serial system model

HAD WE
A B C KNOWN
THE FUTURE…

2 3 1 2 A
2 3 2 13 B
1 1 1 3 2 C

TTCF

11
Presentation II
Simple 3 aging components serial system model

HAD WE
A B C KNOWN
THE FUTURE…
1 2 3 4 A
1 2 3 B
1 2 3 4 C

TBCF

12
Presentation III
Simple 3 aging components serial system model
HAD WE
A B C KNOWN
THE FUTURE…
1 2 3 4 A
1 2 3 B
1 2 3 4 C

TBCF

MTBCF < MTTCF


2 3 1 2 A
2 3 2 13 B
1 1 1 3 2 C

TTCF
13
Simulation Method
MONTE – CARLO

MATHCAD

MIN (X1,1 X2,1 X3,1) MIN (X1,1 X2,1 X3,1)


N=100,000 SETS

N=100,000 SETS
MIN (X1,2 X2,2 X3,2) MIN (X1,2 Δ1,2 Δ2,2)
…………………….

…………………….
MIN (X1,N X2,N X3,N)
_________________ MIN (X1,N Δ1,N Δ2,N)
_________________
1 N 1 N
  min i   min i
N i 1 N i 1
14
How “BIG” is the
Difference ?
1. Depends on the System Architecture.

2. Depends on the Time-To-Failure


distribution of each component.

3. The difference in a specific complex


electronic system was found to be ~40%

Note: True in redundant systems even when


all components have constant failure rates.
15
Why Does It Matter ?
Suppose a specification demand for a system’s
reliability :
MTBCF = 600 hour

Suppose the manufacturer prediction of the


parameter:

X
MTBCF = 780 hour -40%
ATTENTION !!! How was it CALCULATED ????

Is this MTBCF or MTTCF ????

“Real” MTBCF = 480 < 600 (spec)


16
Example 1
Aging serial system – each
component is weibull distributed

17
‫התפלגות ווייבול זהה לכל הפריטים‬

‫‪18‬‬
‫התפלגות ווייבול זהה לכל הפריטים‬

‫‪19‬‬
‫התפלגות ווייבול זהה לכל הפריטים‬

‫‪20‬‬
‫התפלגות ווייבול זהה לכל הפריטים‬

‫‪21‬‬
Example 2
Two redundant subsystems in series – each
component is exponentially distributed

22
Constant
failure rate

23
serial

Constant
failure rate

parallel

24
A Comment about
Asymptotic Availability

(*)
E{TTF } E{TBF }
A  
E{TTF }  E{TTR} E{TBF}  E{TTR}

(*) [ S.Zacks, Springer-Verlag 1991, Introduction To


Reliability Analysis, Par 4.3]

25
Repair policies
1. “Hot repair” is allowed for redundant components.

2. All components are renewed on every failure event.

3. All failed components are renewed on every failure


event.

4. Failed components are renewed only in blocks which


caused the system failure.

5. Failed subsystems are only partially renewed.

26
Conclusions

• System configuration and distribution of


components determine the gap.
• Repair policy should be specified in
advance to determine calculation method.
• Flexible software solutions are needed to
simulate real MTBCF for a given RBD.
• Predict MTBCF not MTTCF
27

Potrebbero piacerti anche