Professional Documents
Culture Documents
Mario Vianello
vianello.clm@tin.it
PRODUCT
QUALITY DESIGN
01OFHLO
These slides can not be reproduced or distributed without permission in writing from the author.
COURSE CONTENTS
These slides can not be reproduced or distributed without permission in writing from the author. 2
PRODUCT QUALITY DESIGN
3
FUNDAMENTALS OF RELIABILITY
AND ROBUST DESIGN
These slides can not be reproduced or distributed without permission in writing from the author. 3
CHAPTER CONTENTS
These slides can not be reproduced or distributed without permission in writing from the author. 4
CHAPTER CONTENTS
These slides can not be reproduced or distributed without permission in writing from the author. 5
3. Fundamentals of Reliability and Robust Design
3.1
RELIABILITY AND ROBUST
DESIGN DEFINITIONS
(related to the Quality)
These slides can not be reproduced or distributed without permission in writing from the author. 6
3.1. Reliability and Robust Design definitions
3.1.1
Probabilistic Design
(short mention)
These slides can not be reproduced or distributed without permission in writing from the author. 7
3.1.1. Probabilistic Design (short mention)
PROBABILISTIC DESIGN
The safety factor h = S/L is replaced by the Failure Probability Pf = f (S-L).
S = strength; L = load; zD = S - L = difference between strength and load
0,40
L = load
(mL =12.50 kg/mm2,
sL = 1.05 kg/mm2)
0,35
distribution
PROBABILITY DENSITY
0,003
0,002
0,25 zD = S - L
0,001 mD = mS - mL S = strength
sDzD= =sSS2 +- sL2 (mS =19.00 kg/mm2,
0,000
2
-2,50 -2,25 -2,00 -1,75 -1,50 -1,25 -1,00 -0,75 -0,50 -0,25 0,00
0,20 ( mDD =6.50
(m = 6.50kg/mm
kg/mm2,, sS = 2.25 kg/mm2)
ssDD = 2.48 kg/mm
kg/mm22)
0,15
strength
0,10 distribution
Note that, if the strength distribution (keeping constant its mean) widens or narrows as its
variability increases or decreases, the failure probability varies (even dramatically), while
the safety factor remains unchanged.
These slides can not be reproduced or distributed without permission in writing from the author. 8
3.1.1. Probabilistic Design (short mention)
The stress distribution refers to the stress peaks, which are the critical factor
able to compromise the strength. Think, as an example, to a strut of the landing
gear of an airplane. The first impact at the landing (involving the greatest stress) is the
one that can produce an immediate break or a "cumulative damage" that will reduce
the strength at the next loads. Other shocks beyond the first are certainly less impor-
tant.
The intersection area or overlapping area of the two normal distributions
that represent strength and stress (highlighted in yellow in the previous
slide) contains both:
pairs of points where the strength exceeds the stress (and therefore
without problems),
pairs of points where the stress exceeds the strength (with breaking
or at least a permanent damage)
and therefore the area of this intersection is not proportional to the failu-
re probability Pf.
Instead, the negative area of the difference normal distribution (highlighted
in red on the previous slide) is proportional to the failure probability Pf
and is obviously smaller than the yellow one above.
These slides can not be reproduced or distributed without permission in writing from the author. 9
3.1.1. Probabilistic Design (short mention)
independent random variables, normally distributed
generic real deterministic constant
ALGEBRAIC
FUNCTION
Mean Standard deviation Useful expressions
to evaluate means
and standard deviations
of random variables
normally distributed
These expressions can be applied
to algebraic variables
in the individual steps
of the usual mathematical expressions
(maybe with the help of EXCEL).
Obviously, for complex cases, it is
preferable to use appropriate software
commercially available.
These slides can not be reproduced or distributed without permission in writing from the author. 10
3.1.1. Probabilistic Design (short mention)
VARIABILITY IN MATERIALS PROPERTIES STATISTICAL VALUES FOR THE M
Coefficient of
Property and material variation
Example of Tables
s/m assessing
Tensile Ultimate Strength of Metallic Materials 0.05 the dispersion
Tensile Yield Strength of Metallic Materials 0.07 of material properties
Endurance Limit for Steel 0.08
through
Coefficient of Variation
Brinnell Hardness of Steel 0.05
s/m
Fracture Toughness of Metallic Materials 0.07
These slides can not be reproduced or distributed without permission in writing from the author. 11
3.1. Reliability and Robust Design definitions
3.1.2
Detailed definitions
of Reliability
These slides can not be reproduced or distributed without permission in writing from the author. 12
3.1.2. Detailed definitions of Reliability
The Reliability R
of a product (system or component)(1)
is its ability
These slides can not be reproduced or distributed without permission in writing from the author. 14
3.1.2. Detailed definitions of Reliability
These slides can not be reproduced or distributed without permission in writing from the author. 17
3.1.2. Detailed definitions of Reliability
These slides can not be reproduced or distributed without permission in writing from the author. 18
3.1.2. Detailed definitions of Reliability
3.1.3
Grading of failures
according to “severity”
These slides can not be reproduced or distributed without permission in writing from the author. 21
3.1.3. Grading of failures according to “severity”
Classification of the failure severity in automotive
(examples for main categories)
Request for repair
SEVERITY DEFINITION E X A M P L E S (to assess the severity
in the course of telephone
interviews with customers)
SAFETY The safety of the driver - Braking inefficient; immediate
- Deceleration sensor (pretensioners)
and/or passengers and/or failure;
others is compromised. - Stop lights switch failure;
- Beam headlamps not working.
CRITICAL These failures force to su- - The engine will not start; immediate
- You can not insert any gear;
(broke down) spend the mission (e.g. - Doors blocked: we do not enter.
vehicle can not start or re-
start).
SERIOUS Functional failures that - Inability to lock the vehicle; as soon as possible:
- Engine: irregular operation; certainly without waiting
need to be repaired within - km/mileometer: failure; another occasion (e.g.
a short time to avoid com- - Door mechanical window not working; scheduled maintenance).
promising the use of the - Dents on the body (e.g. at the delivery).
vehicle or aesthetic failu-
res of particular relevance.
MARGINAL These failures are detec- - Door panel: noise; added to another oc-
- Gear lever boot: detachment/tear;
ted by the customer, but - Small imperfections in the finish or in
casion (e.g. expiry of sche-
not perceived as serious. duled maintenance).
the body.
These slides can not be reproduced or distributed without permission in writing from the author. 22
3.1.3. Grading of failures according to “severity”
Detailed classification (by the Fiat Standard about FMEA)
These slides can not be reproduced or distributed without permission in writing from the author. 23
3.1. Reliability and Robust Design definitions
3.1.4
Mission profile
These slides can not be reproduced or distributed without permission in writing from the author. 24
3.1.4. Mission profile
A.M.O.A. project and black box
A.M.O.A. = Automobile Mode of Operation Achievement
These slides can not be reproduced or distributed without permission in writing from the author. 25
A.M.O.A.: Automobile Mode of Operation Achievement
M A I N F U N C T I O N S D E T E C T E D
ENGINE rpm depending on gear and depression upstream of
the combustion chamber;
water temperature with the number of presences
and the time spent in predefined classes;
temperature of the engine compartment.
0 5 1,350 9% 2 18
~ 30%
6 10 2,100 14% downtown
7 23
11 25 3,150 21% 15 28
26 50 2,100 14% 32 41
~ 30%
51 100 1,500 10% highway
58 53
over 100 km 4,800 32% 175 78
TOTAL 15,000 100%
These slides can not be reproduced or distributed without permission in writing from the author. 27
3.1.4. Mission profile
Mission profile for a class C gasoline car (Italian Market)
(average yearly mileage = 15,000 km; overall average speed = 35.5 km/h).
BAR CHART with the breakdown of average annual driving distance in the classes of journey length
6.000
30% downtown 30% highway
4.000
3.150 (21%)
3.000
1.000
0
A B C D E F
(0÷5 km) (6÷10 km) (11÷25 km) (26÷50 km) (51÷100 km) (oltre 100 km)
These slides can not be reproduced or distributed without permission in writing from the author. 28
3.1.4. Mission profile
Average yearly mileage = 15,000 km
Example of COMPLETE
Average number of missions per year = 1,250 missions/year
MISSION PROFILE Average number of missions in 10,000 km = 833 missions/(10,000 km)
for a whole vehicle
(TIPO gasoline, Averge mission mileage = 12 km/mission
class C, 1987) Average speed (waitings not included) = 35.5 km/h
These slides can not be reproduced or distributed without permission in writing from the author. 29
3.1. Reliability and Robust Design definitions
3.1.5
Remarks on Reliability
practical measurements
These slides can not be reproduced or distributed without permission in writing from the author. 30
3.1.5. Remarks on Reliability practical measurements
Design intrinsic
Rd DESIGN reliability
mission reliability
(improves with redundancies)
RELIABILITY
measured reliability
by the number of repairs
per 100 vehicles
at the Customer Service
(worsens with redundancies)
Let us imagine that a twin-engined plane (equipped with two engines of the
kind described above), to be able to take off and land, needs that at least
one of the engines does work.
As to prevent the airplane to fly, it is necessary that both engines fail, the
probability that the plane crashes (mission unreliability) is equal to P2,
but the probability of a request for repair (measured unreliability ) is ap-
proximately equal to 2.P.
Let us consider now a four-engined aircraft (4 motors of the kind mentio-
ned above), which likewise, in order to take off and land, needs that at least
one of the engines does work.
As it is necessary that all four engines will fail to prevent the aircraft to fly,
the probability that the aircraft crashes (mission unreliability) is now
equal to P4 (much less than the previous one), but the probability of a
request for repair (measured unreliability) is now roughly equal to 4.P
(twice the previous one !).
These slides can not be reproduced or distributed without permission in writing from the author. 34
3.1.5. Remarks on Reliability practical measurements
These slides can not be reproduced or distributed without permission in writing from the author. 35
3.1.6. “Robustness” concept (Robust Design)
For possible responses to each identified expected function can
be defined:
The ideal value = nominal dimension
A first range, very narrow, within which
virtually no customer is able to percei-
ve the difference between the respon-
ses (this range is of poor importance
from an operational standpoint).
The tolerability range, rather wide. A
annoying zones
response out of it is so unsatisfactory
for the customer to induce him to se-
riously consider the opportunity to chan-
ge supplier. For this reason, the zones
outside it are called annoying zones.
It is appropriate that Quality & Reliability targets refer to the tolerability range
(the wide one) associated with an extremely low probability (typically 3.4 ppm)
of "breached", in order that there are no doubt that, in case of overruns, analysis
and corrective actions will be promptly undertaken.
These slides can not be reproduced or distributed without permission in writing from the author. 36
3.1.6. “Robustness” concept (Robust Design)
B A S I C I N F O R M A T I O N
These slides can not be reproduced or distributed without permission in writing from the author. 38
3.1. Reliability and Robust Design definitions
3.1.6
“Robustness” concept
(Robust Design)
These slides can not be reproduced or distributed without permission in writing from the author. 39
3.1.6. “Robustness” concept (Robust Design)
These slides can not be reproduced or distributed without permission in writing from the author. 40
3.1.6. “Robustness” concept (Robust Design)
A product is robust if it’s little affected
by causes of performance variability,
generated both by:
design and/or process parameters named control
factors,
environmental factors and mission characteristics
or process parameters whose control has been
judged difficult or too expensive to be controlled: all
are named noise factors.
Noise factors examples :
environment: pressure, temperature, humidity, vibrations, pollution,
electromagnetic fields, etc.;
items: dimensions, roughness, hardness, strength, etc.;
usage: wear-out, overstress, etc.;
dust ...
These slides can not be reproduced or distributed without permission in writing from the author. 41
3.1.6. “Robustness” concept (Robust Design)
Turning off the engine of a car with the turn signal still running, so-
mething remains polarized in the engine control unit, so that the
next restarting the engine can take several attempts with a loss of
time very annoying.
Turn off the engine with the turn signal “on” is a maneuver rare
but allowed, so the design of this control unit is “weak” and must
be improved.
These slides can not be reproduced or distributed without permission in writing from the author. 43
3.1.6. “Robustness” concept (Robust Design)
These slides can not be reproduced or distributed without permission in writing from the author. 44
3.1.6. “Robustness” concept (Robust Design)
These slides can not be reproduced or distributed without permission in writing from the author. 45
3.1.6. “Robustness” concept (Robust Design)
Output
Input SYSTEM Y = system response
M = signal factors (product or process) (quality characteristic)
These slides can not be reproduced or distributed without permission in writing from the author. 47
3.1.6. “Robustness” concept (Robust Design)
Structure of “influencing factors”
INFLUENCING FACTORS
UNACCEPTABILITY
TOLERANCE
CUSTOMER'S
EXPECTED
FUNCTION
RESPONSE
ON THE
INPUT PRODUCT TOLERABILITY
RANGE
(borrowed from 6 s )
(*) Signal factors and noise factors are always characterized by a UNACCEPTABILITY
predefined range of variability, outside of which is not generally
guaranteed a satisfactory functioning of the system.
These slides can not be reproduced or distributed without permission in writing from the author. 48
3.1. Reliability and Robust Design definitions
3.1.7
Concept, Parameter and
Tolerance Design
These slides can not be reproduced or distributed without permission in writing from the author. 49
3.1.7. Concept, Parameter and Tolerance Design
The 3 phases
Selects among architectural and technological al-
CONCEPT ternatives, in order to satisfy customer functional
requirements and minimize variability.
DESIGN Main tools:
(or System Design) Norms;
Preventive reliability analysis;
Diagnosis of failures in field;
Results of problem solving activities (especially D.O.E.).
The CONCEPT phase is often mentioned in Literature,
but without underlying its fundamental importance.
Optimizes the nominal values of control factors,
PARAMETER without tolerance reduction (and related extra
costs).
DESIGN
This phase is used only if the previous one could
TOLERANCE not reach the stated targets. Starting from Parame-
ter Design results, the Tolerance Design reduces
DESIGN tolerances mostly for the control factors with the
best cost/benefits ratios.
Of course these activities are much more expen-
sive than the previous ones.
These slides can not be reproduced or distributed without permission in writing from the author. 50
3.1.7. Concept, Parameter and Tolerance Design
These slides can not be reproduced or distributed without permission in writing from the author. 51
3.1.7. Concept, Parameter and Tolerance Design
like D.O.E.
A rational
approach
than necessary (concept
PARAMETER DESIGN jump);
phase
without any method b) Adoption of a poor solution
from the customer’s point
of view.
These slides can not be reproduced or distributed without permission in writing from the author. 52
3.1. Reliability and Robust Design definitions
3.1.8
Taguchi’s Loss Function
These slides can not be reproduced or distributed without permission in writing from the author. 53
3.1.8. Taguchi’s Loss Function
Tolerances
represent only a “test”
to accept or to reject,
but it is not true
that all the values within them
satisfy the customers
in the same manner.
These slides can not be reproduced or distributed without permission in writing from the author. 54
3.1.8. Taguchi’s Loss Function
( nomial value )
Target value
Nominal value
TOLERANCE LIMITS
TOLERANCE LIMITS
These slides can not be reproduced or distributed without permission in writing from the author. 55
3.1.8. Taguchi’s Loss Function
Example of calculating warranty-costs
(incurred in Customer Office)
( nomial value )
Target value
TOLERANCE LIMITS
Let us suppose that, if the value of the characteristic is as shown in the figure,
there is a 1% risk that this component, when assembled with another, gene-
rates a failure (complained by the customer), and its repair costs 100.00 €.
Therefore the cost incurred by the Customer Office can be calculated as:
1
--------- . 100,00 = 1,00 €/component
100
These slides can not be reproduced or distributed without permission in writing from the author. 56
3.1.8. Taguchi’s Loss Function
These slides can not be reproduced or distributed without permission in writing from the author. 57
3.1.8. Taguchi’s Loss Function
LOSS FUNCTION
NOMINAL VALUE
These slides can not be reproduced or distributed without permission in writing from the author. 58
3.1.8. Taguchi’s Loss Function
The public showed a strong preference for color Sony television sets produced in Japan, com-
pared to the corresponding produced in the U.S., although both comply with the same specifica-
tions for color intensity.
Quality
levels
These slides can not be reproduced or distributed without permission in writing from the author. 59
3.1.8. Taguchi’s Loss Function
L = K [ (m - m)2 + s2]
COSTS (SOCIAL LOSSES) = Loss Function
( nominal value )
Distribution of quality
Target value
characteristic, y,
with mean m and variance s2
Loss Function
m m = process mean
These slides can not be reproduced or distributed without permission in writing from the author. 60
3.1.8. Taguchi’s Loss Function
TIPO DI FUNZIONE RAPPRESENTAZIONE CARATTERISTICHE
L(y)
A0
L(y)
- valore ideale = (L(y) = 0)
- non può assumere valori
TANTO PIÙ GRANDE L(y) = k (1/y negativi
These slides can not be reproduced or distributed without permission in writing from the author. 61
3.1.8. Taguchi’s Loss Function
n n- 1 2
Q k m y 0
ACCEPTANCE
s
REJECTION REJECTION 2
AREA AREA AREA
n-1
n
Lower
tolerance limit Nominal value Upper
tolerance limit
Q k DX S 2 2
QS
These slides can not be reproduced or distributed without permission in writing from the author. 62
3.1.8. Taguchi’s Loss Function
To summarize
These slides can not be reproduced or distributed without permission in writing from the author. 63
3.1. Reliability and Robust Design definitions
3.1.9
Practical recommendations
These slides can not be reproduced or distributed without permission in writing from the author. 64
3.1.9. Practical recommendations
These slides can not be reproduced or distributed without permission in writing from the author. 65
3.1.9. Practical recommendations
None of the world automobile industry
would be able to achieve reliability targets
as demanding as those currently imposed by the market,
if it should set
the task of preventing and checking for reliability
(for each setting of a new car model)
each time from scratch on all components/systems,
as if for the first time.
These slides can not be reproduced or distributed without permission in writing from the author. 67
3.1.9. Practical recommendations
3.2
SERIES SYSTEM
AND PARALLEL SYSTEM
These slides can not be reproduced or distributed without permission in writing from the author. 69
3.2. Series system and parallel system
COMPONENT SYSTEM
Object, Set of connected
also complex, components designed
whose reliability to satisfy a function,
can be assessed by whose reliability can not
directly using be assessed by
the statistical data directly using available
from the field. experimental data.
These slides can not be reproduced or distributed without permission in writing from the author. 70
3.2. Series system and parallel system
A B N
These slides can not be reproduced or distributed without permission in writing from the author. 71
3.2. Series system and parallel system
A B N
Being Rs = P Ri,
the system failure probability is immediately obtained as:
Fs = 1 - Rs = 1 - P Ri = 1 - P (1 - Fi)
These slides can not be reproduced or distributed without permission in writing from the author. 72
3.2. Series system and parallel system
SERIES SYSTEM: case of only two elements (with different reliability)
A B
Rs = RA . RB,
Rs = (1 - FA) . (1 - FB) = 1 - FA - FB + FA . FB
Fs = 1 - Rs = FA + FB - FA . FB
The sensitivity S of the system reliability, Rs, to the values Ri of the reliability of every element
is given by the derivative of the Rs with respect to the reliability of the specific element i.
These slides can not be reproduced or distributed without permission in writing from the author. 74
3.2. Series system and parallel system
These slides can not be reproduced or distributed without permission in writing from the author. 75
3.2. Series system and parallel system
Being Fs = P Fi,
the system reliability is immediately obtained as:
Rs = 1 - Fs = 1 - P Fi = 1 - P (1 - Ri)
These slides can not be reproduced or distributed without permission in writing from the author. 76
3.2. Series system and parallel system
A
B
Fs = FA . FB,
Fs = (1 - RA) . (1 - RB) = 1 - RA - RB + RA . RB
Rs = 1 - Fs = RA + RB - RA . RB
This expression is the same as the sum of probabilities
(“union” event) for mutually independent events:
P(AB) = P(A) + P(B) - P(A) . P(B) .
The sensitivity S of the system reliability, Rs, to the values, Ri, of the reliability of every element
is given by the derivative of the Rs with respect to the reliability of the specific element i.
The derivative of: Rs = 1 - [(1 - R1) . (1 - R2) . … . (1 - Ri ) . … . (1 - Rn)] ith respect to Ri,
can be written as:
d Rs d (1-Ri)
--------- = 0 - [(1-R1) . (1-R2) . … . (1-Ri-1) . (1-Ri+1) . (1-Rn) . ------------- =
d Ri d Ri
= - (1-R1) . (1-R2) . … . (1-Ri-1) . (1-Ri+1) . (1-Rn) . (0 - 1) =
1 - Rs
= (1-R1) . (1-R2) . … . (1-Ri-1) . (1-Ri+1) . (1-Rn) = -------------
1 - Ri
These slides can not be reproduced or distributed without permission in writing from the author. 78
3.2. Series system and parallel system
PARALLEL SYSTEM: sensitivity
A
B
We see that the searched derivative is given by the product of all (1 - Rj)
except (1 - Ri).
As the product of all (1 - Rj) is (1 - Rs), the derivative can be written as:
(1 - Rs) / (1 - Ri) .
This can be confirmed by observing that the reliability of a whole parallel-system is always higher
than the reliability of its individual elements (and also of that one more reliable). Thus, if we want
to improve the system reliability, the most convenient thing to do is to improve the reliability of
the component currently more reliable, i.e. the best, because, as long as it works, the whole
system continues to operate.
These slides can not be reproduced or distributed without permission in writing from the author. 79
3.2. Series system and parallel system
SERIES & PARALLEL SYSTEMS: summary
A B N
Rs = P Ri Fs = 1 - Rs = 1 - P Ri = 1 - P (1 - Fi)
d Rs Rs
--------- = -------
Ri Ri
A Fs = P Fi
B Rs = 1 - Fs = 1 - P Fi = 1 - P (1 - Ri)
d Rs 1 - Rs
N --------- = -----------
d Ri 1 - Ri
These slides can not be reproduced or distributed without permission in writing from the author. 80
3.2. Series system and parallel system
These slides can not be reproduced or distributed without permission in writing from the author. 82
3.2. Series system and parallel system
EXERCISE - Of the system represented below, we want to know:
1. the overall system’s failure frequency (at 500 hours), FS;
2. the component (among A, B and C) that is more suitable to improve in order to reduce
the overall system’s failure frequncy, FS, supposing we can improve only one component.
FAILURE FREQUENCIES
A = 0.03
FA = 0.03 repairs/500 hours
C = 0.0005 FB = 0.02 repairs/500 hours
B = 0.02 FC = 0.0005 repairs/500 hours
X C = 0.0005
RS = P Ri = RX . RC = (1 - FX) . (1 - FC) =
= (1 - 0.0006) . (1 - 0.0005) = 0.9994 . 0.9995 = 0.9989 Series system X-C
If, on the last expression above (in bold), we perform a sensitivity analysis by partial derivati-
ves, using the values of reliability obtainable from the previous slide (RA = 1 - FA = 1,00 - 0,03 =
0,97; RB = 1 - FB = 1,00 - 0,02 = 0,98; RC = 1 - FC = 1,0000 - 0,0005 = 0,9995), we obtain:
d RS
---------- = (1 - RB) . RC = 0.02 . 0.9995 = 0.01999
d RA
d RS
---------- = (1 - RA) . RC = 0.03 . 0.9995 = 0.02989
d RB
d RS
---------- = (RA + RB - RA . RB) = 0.97 + 0.98 - 0.97 . 0.98 = 0.9994
d RC
As the partial derivative with the highest value is by far the one with respect to the ele-
ment C (0,9994), the sensitivity analysis would indicate the opportunity to act on the
element C, instead of the element B, as shown in the previous slide!
These slides can not be reproduced or distributed without permission in writing from the author. 84
3.2. Series system and parallel system
To clarify the issue, in the table below, the reliability of the system was recalculated as
a result of improvements (small and large), implemented on the elements in runoff, B
and C, but established by two different criteria.
EQUAL IMPROVEMENT (IN ABSOLUTE VALUE) FOR THE ELEMENTS A AND C IMPROVEMENTS PROPORTIONAL TO THE INITIAL FAILURE FREQUENCY OF EACH ELEMENT
SMALL IMPROVEMENT = 0,00002 It is better to act on element C SMALL IMPROVEMENT = 1,50% It is better to act on element B
BIG IMPROVEMENT = 0,0004 Even more, in this case, it is better to act on element C BIG IMPROVEMENT = 20,00% Even more, in this case, it is better to act on element B
These slides can not be reproduced or distributed without permission in writing from the author. 85
3.2. Series system and parallel system
In fact, partial derivatives express the angular coefficient of the benefits obtainable on the
system reliability. The value of the angular coefficient represents also the order of impor-
tance of the individual elements of the system if we assume an equal increase in reliability
for each element. But, in production, the reliability improvements are generally proportional
to the failure frequency of the element, since it is easier to improve an unreliable element
than a very reliable one!
Rs versus B Rs versus C
0,0015 0,0013
0,0014 0,0012
0,0013
0,0011
0,0012
0,0010
0,0011
0,0009
0,0010
0,0008
y = 0,9994x + 0,0006
0,0009
0,0008 0,0007
0,0007 0,0006
0,0006
0,0005
0,0005
y = 0,03x + 0,0005
0,0004
0,0004
0,0003
0,0003
0,0002
0,0002
0,0001 0,0001
0,0000 0,0000
0,0000 0,0025 0,0050 0,0075 0,0100 0,0125 0,0150 0,0175 0,0200 0,0225 0,0250 0,0275 0,0300 0,00025 0,00030 0,00035 0,00040 0,00045 0,00050 0,00055 0,00060
The angular coefficient of the element C (0,9934) is obviously much higher than that of the
element B (0.03): but, clearly, improving C, which already has a very high reliability, is con-
siderably more difficult than improve B.
Therefore, in presence of mixed systems (and not all series or all parallel), we must use
great caution in adopting the results of the partial derivatives.
These slides can not be reproduced or distributed without permission in writing from the author. 86
3.2. Series system and parallel system
We have seen that the Reliability function, R(F1, F2, F3, …, Fn), can be differentiated with respect
to each Fi, writing:
dR
dR = ------- dFi
dFi
However dFi expresses here the infinitesimal variation of failure frequency, Fi, in absolute terms.
The same infinitesimal variation of failure frequency, Fi, could be expressed, as a fraction (or si-
milarly as a percentage) of Fi, in the form (dFi/Fi). At this point, we can develop the previous ex-
pression and write:
dR dR dFi
dR = ------- dFi = Fi ------- -------
dFi dFi Fi
Thus we see that the effect of improvements in terms of fraction (or percentage) of the initial
nominal failure frequency is proportional not only to the differential term, but also to the initial
nominal value of the failure frequency, Fi.
With reference to the system of the exercise (where it was: R = (1 - FA . FB) . (1 - FC); FA = 0,03;
FB = 0,02; FC = 0,0005), we obtain the following angular coefficients:
In absolute terms In percentage terms
dR dR
------- = FB . (1 - FC) = 0,019990 FA . ------- = FA . FB . (1 - FC) = 0,0005997 In percentage terms,
dFA dFA the elements A and B
dR dR have equal
------- = FA . (1 - FC) = 0,029985 FB . ------- = FB . FA . (1 - FC) = 0,0005997 partial derivatives.
dFB dFB
dR dR
------- = (1 - FA . FB) = 0,999400 FC . ------- = FC . (1 - FA . FB) = 0,0004997
dFC dFC
These slides can not be reproduced or distributed without permission in writing from the author. 87
3. Fundamentals of Reliability and Robust Design
3.3
SOURCES
OF INFORMATION
FROM THE FIELD
These slides can not be reproduced or distributed without permission in writing from the author. 88
3.3. Sources of information from the field
Essential characteristics
of information sources from the field
1) Acquisition time
2) Types and quality of information
3) Availability in different markets
These slides can not be reproduced or distributed without permission in writing from the author. 90
3. Fundamentals of Reliability and Robust Design
3.4
MULTI-TARGET
OPTIMIZATIONS
These slides can not be reproduced or distributed without permission in writing from the author. 91
3.4. Multi-target optimizations
3.4.1
The basic problem
These slides can not be reproduced or distributed without permission in writing from the author. 92
3.4.1. Multi-target optimizations: the basic problem
OPTIMIZATION METHODS
have the purpose of:
finding the conditions, i.e. the values
xi of independent variables, which
make so that the responses yi (de-
pendent variables) of expected func-
tions are very close to their assigned
targets,
taking into account of possible con-
straints.
These slides can not be reproduced or distributed without permission in writing from the author. 93
3.4.1. Multi-target optimizations: the basic problem
These slides can not be reproduced or distributed without permission in writing from the author. 94
3.4.1. Multi-target optimizations: the basic problem
These slides can not be reproduced or distributed without permission in writing from the author. 95
3.4.1. Multi-target optimizations: the basic problem
PROBLEMS CLASSIFICATION
3.4.2
Logical and mathematical
tools
These slides can not be reproduced or distributed without permission in writing from the author. 97
3.4.2. Multi-target optimizations: logical & math. tools
Principles
These slides can not be reproduced or distributed without permission in writing from the author. 99
3.4.2. Multi-target optimizations: logical & math. tools
These slides can not be reproduced or distributed without permission in writing from the author. 100
3.4.2. Multi-target optimizations: logical & math. tools
These slides can not be reproduced or distributed without permission in writing from the author. 101
3.4.2. Multi-target optimizations: logical & math. tools
Definitions:
These slides can not be reproduced or distributed without permission in writing from the author. 102
3.4.2. Multi-target optimizations: logical & math. tools
These slides can not be reproduced or distributed without permission in writing from the author. 103
3.4.2. Multi-target optimizations: logical & math. tools
These slides can not be reproduced or distributed without permission in writing from the author. 104
3.4.2. Multi-target optimizations: logical & math. tools
“Replacement” techniques
of a multi-target problem
Weights method;
Constraints method.
These slides can not be reproduced or distributed without permission in writing from the author. 105
3.4. Multi-target optimizations
3.4.3
“Weights” method
These slides can not be reproduced or distributed without permission in writing from the author. 106
3.4.3. Multi-target optimizations: “weights” method
These slides can not be reproduced or distributed without permission in writing from the author. 107
3.4.3. Multi-target optimizations: “weights” method
From a general point of view, mathematical aspects are not a big deal,
but the main difficulty is to assign to the "weights" values consistent with
the physical reality of the phenomenon to be optimized.
For example, in an engine, what are the correct values of the performances
weights (for each of the features listed on the right) that allow fuel economy
the subsequent mathematical processing to identify the best pollution
configuration capable of inducing the customer to purchase ? costs
If the assignment of "weights" is not consistent with the values that induce custo-
mers to purchase, the mathematical solution will certainly result in low efficiency.
These slides can not be reproduced or distributed without permission in writing from the author. 108
3.4. Multi-target optimizations
3.4.4
“Constraints” method
These slides can not be reproduced or distributed without permission in writing from the author. 109
3.4.4. Multi-target optimizations: “constraints” method
These slides can not be reproduced or distributed without permission in writing from the author. 110
3.4.4. Multi-target optimizations: “constraints” method
These slides can not be reproduced or distributed without permission in writing from the author. 111
3.4.4. Multi-target optimizations: “constraints” method
These slides can not be reproduced or distributed without permission in writing from the author. 112
3.4. Multi-target optimizations
3.4.5
Comparison of two methods
and their integrated use
These slides can not be reproduced or distributed without permission in writing from the author. 113
3.4.5. Multi-target optimizations: comparison of 2 methods
These slides can not be reproduced or distributed without permission in writing from the author. 114
3.4.5. Multi-target optimizations: comparison of 2 methods
The procedure is easier to manage if we proceed step by step.
a) It is preferable to start with the constraints method, listing all the con-
straints “we ideally want“.
b) If, as happens in most cases, this way will not get to any solution, we
have to establish an order of importance (from the customer’s point of
view) of each constraint: somehow, this relates to the weights method.
c) In parallel, we can define, for each considered constraint, a softening,
changing “ideally desired" values in "acceptable“ values.
d) It begins to take on, for each constraint (one at a time starting from the
less important ones) "acceptable“ values: if a solution is found we stop,
otherwise we proceed, "softening" the subsequent constraint in order of
importance.
e) If not even this gives us a viable solution, we must replace, for each
constraint, the above "acceptable“ values with some "minimum accep-
table" values less demanding than the previous ones and re-start again
from step d.
f) If even this will not provide an acceptable trade-off, we must give up so-
mething else, perhaps with a minimum of importance.
These slides can not be reproduced or distributed without permission in writing from the author. 115
3.4.5. Multi-target optimizations: comparison of 2 methods
These slides can not be reproduced or distributed without permission in writing from the author. 116
3. Fundamentals of Reliability and Robust Design
3.5
RELIABILITY TARGETS
DEPLOYMENT
These slides can not be reproduced or distributed without permission in writing from the author. 117
3.5. Reliability targets deployment
Deployment
Targets
example by comparisons with the best competitors) and
subsequent allocation of targets at the component le-
vel, based on results from the field and detailed compa-
rison with the best competitors: TOP-DOWN approach;
departments)
the improvements that can be implemented in the de-
sign and/or in the production process, and subsequent
evaluation of the achievable target at the whole
vehicle level: BOTTOM-UP approach.
These slides can not be reproduced or distributed without permission in writing from the author. 118
3.5. Reliability targets deployment
EXAMPLE OF Archetype
TARGETS 200
DEPLOYMENT:
Group 1 Group n
TOP-DOWN
Steering
Engines Brake
system Painting
"Fire" system
APPROACH
"Sulmona"
11,575 4,500 7,475 3,025
Steering
Steering
Steering shaft Steering
Others shafts and Others
wheel support box
couplings
bushings
0,925 0,425 1,200 1,025 0,625 0,300
80/200 = 0,4
Archetype's values New car
multiplied by 0,4
model
80
Group 1 Group n
Steering
Engines Brake
system Painting
"Fire" system
"Sulmona"
4.63 1.80 2.99 1.21
Steering
Steering
Steering shaft Steering
Others shafts and Others
wheel support box
couplings
bushings
0.37 0.17 0.48 0.41 0.25 0.12
These slides can not be reproduced or distributed without permission in writing from the author. 119
3.5. Reliability targets deployment
BLOCK DIAGRAM OF DEVELOPMENT TECHNIQUE
I N P U T S AA CT T TI V I TI I TE À
I V S OUT P U T S
OBIETTIVO
Whole
avehicle
livello
TARGETS
totale veicolo
Distinction between
Suddivisione
carry-over and specific
carry-over e specifici
COMPUTER DEPLOYMENT
components
INFORMATIZZATA
OBIETTIVI
TARGETS
Some component sì
Failure frequencies Reliability has a more demanding
più severo
yes ALLOCATION
Frequenze di guasto DEPLOYMENT ASSEGNAZIONE
(from targets targetobiettivo
than OF
DI AFFIDABILITÀ
rispetto
da Rete Assistenziale
Customer Service)
obiettivi di affidabilità
the archetype? OBIETTIVI
DEPLOYMENT precedente ? RELIABILITY
PROCEDURE
DI AFFIDABILITÀ
TARGETS
A LIVELLO
FOR DI
DEPLOYMENT
no
OF RELIABILITY
no COMPONENTE
Warranty costs INDIVIDUAL
Costi COMPONENTS
(from
da Rete Assistenziale
Customer Service)
PROCEDURA
Average Failure
Gravità Targets
Severity
THE
RITARATURA
da Tabelle di
(for each component) TUNING
FORDI
Competitors data
Dati concorrenza
(if available)
(dove disponibili)
These slides can not be reproduced or distributed without permission in writing from the author. 120
3. Fundamentals of Reliability and Robust Design
3.6
MAIN INDICATORS
OF THE REACHED
RELIABILITY LEVEL
These slides can not be reproduced or distributed without permission in writing from the author. 121
3.6. Main indicators of the reached reliability level
3.6.1
R(t), F(t) and M(t)
functions
These slides can not be reproduced or distributed without permission in writing from the author. 122
3.6.1. R(t), F(t) and M(t) functions
Non-repairable systems: basic functions
R(t)Curve F(t)functions
and F(t) e R(t)
1,0
0,9
Failure probability
Probabilità F(t) == 11 -- R(t)
di guasto F(t) R(t)
PROBABILITÀPROBABILITY
0,8
CUMULATA
0,7
0,6
0,5
CUMULATIVE
0,4
0,3
0,2
Affidabilità
Reliability R(t)
R(t) == 11 -- F(t)
F(t)
0,1
0,0
0 50.000 100.000 150.000 200.000 250.000 300.000 350.000
MILEAGE, km (or
PERCORRENZA, kmhours
(o tempo o cicli)
or cycles)
Function R(t) always starts from the value 1 and becomes equal to zero when all tested items are bro-
ken (in this case outside the right side of the graph). Function F(t) always starts from zero and becomes
equal to 1 when all tested items are broken (in this case outside the right side of the graph).
These slides can not be reproduced or distributed without permission in writing from the author. 123
3.6.1. R(t), F(t) and M(t) functions
“Non-repairable” systems
R(t)Curve F(t)functions
and F(t) e R(t)
1,0
0,9
Failure probability
Probabilità F(t) == 11 -- R(t)
di guasto F(t) R(t)
PROBABILITÀPROBABILITY
0,8
CUMULATA
0,7
0,6
0,5
CUMULATIVE
0,4
0,3
0,2
Affidabilità
Reliability R(t)
R(t) == 11 -- F(t)
F(t)
0,1
0,0
0 50.000 100.000 150.000 200.000 250.000 300.000 350.000
MILEAGE, km (or
PERCORRENZA, kmhours
(o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 124
3.6.1. R(t), F(t) and M(t) functions
f(t)CURVE
& h(t) f(t)
functions
e h(t)
Both functions are often
GUASTO RATE
1,2E-05
denoted by l,
1,1E-05 which creates confusion.
DIHAZARD
1,0E-05
9,0E-06
e TASSO&
6,0E-06
DENSITÀDENSITY
3,0E-06
PROBABILITY
2,0E-06
1,0E-06
0,0E+00
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. (1) o “failure rate” . 125
3.6.1. R(t), F(t) and M(t) functions
(1) Alessandro Birolini - RELIABILITY ENGINEERING: Theory and Practice, Fourth Edition, Springer, 2004
These slides can not be reproduced or distributed without permission in writing from the author. 126
3.6.1. R(t), F(t) and M(t) functions
These slides can not be reproduced or distributed without permission in writing from the author. 128
3.6.1. R(t), F(t) and M(t) functions
1,1E-05
1,0E-05
h(t) = constant
9,0E-06
DI PROBABILITÀ
8,0E-06
rate
7,0E-06
Hazard
6,0E-06
5,0E-06
DENSITÀ
4,0E-06
f(t)
3,0E-06
2,0E-06
1,0E-06
0,0E+00
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 129
3.6.1. R(t), F(t) and M(t) functions
These slides can not be reproduced or distributed without permission in writing from the author. 130
3.6.1. R(t), F(t) and M(t) functions
Weights =
Each of the 3 Weibull must
be “weighted” according to
the proportion of different
kind of failures (infant morta-
lity, random and wear-out) on
the considered system.
0,000010
0,000005
0,000000
0 50.000 100.000 150.000 200.000 250.000 300.000 350.000
These slides can not be reproduced or distributed without permission in writing from the author. 132
3.6.1. R(t), F(t) and M(t) functions
The simultaneous presence of all three Weibull, which distinguishes this model, has
the result that the bottom of the function h(t) can not generally be perfectly flat, as in
the case of a single exponential distribution (only random failures). We can also
observe that a horizontal stroke of the function h(t) = f(t)/R(t) usually excludes a
corresponding horizontal stroke of the function f(t), since it should remain constant
R(t), which is impossible.
2,5E-5
2,0E-5
h(t)
GUASTO
Wear-out
TASSO DIRATE,
1,5E-5
Infant mortality
HAZARD
1,0E-5
Useful life
5,0E-6
(constant)
0,0E+0
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000
PERCORRENZA,
MILEAGE,km km
km( o (or (o tempo
tempo
hours o cicli)
oorcicli )
cycles)
A graph like that above can not arise from the model with the three Weibull discus-
sed before. Nevertheless it is the most commonly mentioned in Literature. It is not
unreasonable, but implies that, between the end of the area of infant mortality
and the beginning of that of wear-out, there are only random failures. That is to
say that the model is composed of three Weibull juxtaposed and truncated at the
end of each zone, with no overlap between them. In the central zone, the Weibull
would present b = 1, becoming an exponential distribution.
These slides can not be reproduced or distributed without permission in writing from the author. 133
3.6.1. R(t), F(t) and M(t) functions
From a technical point of view, the previous hypothesis about the horizontal
stroke of the hazard rate means to assume that, in the useful life zone,
can never happen a “late” failure of infant mortality, nor an “early”
failure of wear-out.
We can not exclude that the practice presents situations of this kind, but it
seems not to be the more general case. Thus the graph of Literature is
mainly important from a conceptual point of view.
These slides can not be reproduced or distributed without permission in writing from the author. 134
3.6.1. R(t), F(t) and M(t) functions
0,9
Failure probability
Probabilità F(t) == 11 -- R(t)
di guasto F(t) R(t)
PROBABILITÀPROBABILITY
0,8
CUMULATA
0,7
0,6
0,5
CUMULATIVE
0,4
0,3
0,2
Affidabilità
Reliability R(t)
R(t) == 11 -- F(t)
F(t)
0,1
0,0
0 50.000 100.000 150.000 200.000 250.000 300.000 350.000
MILEAGE, km (or
PERCORRENZA, kmhours
(o tempo o cicli)
or cycles)
f(t)CURVE
& h(t) f(t)
functions
e h(t)
GUASTO RATE
1,2E-05
1,1E-05
DIHAZARD
1,0E-05
9,0E-06
h(t) della sola componente esponenziale (inconvenienti casuali )
e TASSO&
8,0E-06
FUCNTION
7,0E-06
DI PROBABILITÀ
6,0E-06
DENSITÀDENSITY
5,0E-06
4,0E-06
3,0E-06
PROBABILITY
2,0E-06
1,0E-06
0,0E+00
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 135
3.6.1. R(t), F(t) and M(t) functions
Repairable systems: basic function
M(t) function
Curva M(t)
8
Wear-out
UNIT
UNITÀ
7
M(t) values
may exceed
t
PER
al tempo
unity
FAILURES PER
6
/ unità]
at the time t
Useful life
GUASTI
]
cumulato
[repairs / unit
5
[inconvenienti
Presence of
MEDIO DIOF
failures also
cumulated
4
AVERAGE NUMBER
at 0 km [FCA]
mediamente
3 M(t) estimation
at 24 months of use
l = constant slope of the linear stroke
Infant NUMERO
1
Average warranty
mortality
driving distance
0
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 136
3.6.1. R(t), F(t) and M(t) functions
UNIT
0,9
UNITÀ
Failure probability
Probabilità F(t) == 11 -- R(t)
di guasto F(t) R(t) 7
t
PROBABILITÀPROBABILITY
PER
0,8
al tempo
CUMULATA
FAILURESPER
6
[inconvenienti / unità]
t
0,7
at the time
GUASTI
]
cumulato
unit
0,6 5
/
MEDIO DIOF
[repairs
0,5
cumulated
CUMULATIVE
NUMERO NUMBER
mediamente
0,4
3
0,3
AVERAGE
2
0,2
Affidabilità
Reliability R(t)
R(t) == 11 -- F(t)
F(t) 1
0,1
0
0,0
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
0 50.000 100.000 150.000 200.000 250.000 300.000 350.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
MILEAGE, km (or
PERCORRENZA, kmhours
(o tempo o cicli)
or cycles)
f(t)CURVE
& h(t) f(t)
functions
e h(t)
GUASTO RATE
1,2E-05
1,1E-05
DIHAZARD
1,0E-05
9,0E-06
h(t) della sola componente esponenziale (inconvenienti casuali )
e TASSO&
8,0E-06
FUCNTION
7,0E-06
DI PROBABILITÀ
6,0E-06
DENSITÀDENSITY
5,0E-06
4,0E-06
3,0E-06
PROBABILITY
2,0E-06
1,0E-06
0,0E+00
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 137
3.6.1. R(t), F(t) and M(t) functions
Repairable systems: derivative
l function
Curva = = dM(t)/dt
dM(t)/dt
8,0E-05
FUNCTION
7,0E-05
ZA DI GUASTO
x km)]
6,0E-05
DENSITY
x km)
/ (unità ]
5,0E-05
DI FREQUEN
[ repairs / (unit
PROBABILITY
4,0E-05
[inconvenienti
3,0E-05
INTENSITÀ
2,0E-05
FAILURE
1,0E-05
0,0E+00
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 138
3.6.1. R(t), F(t) and M(t) functions
UNIT
0,9
UNITÀ
Failure probability
Probabilità F(t) == 11 -- R(t)
di guasto F(t) R(t) 7
t
PROBABILITÀPROBABILITY
PER
0,8
al tempo
CUMULATA
FAILURESPER
6
[inconvenienti / unità]
t
0,7
at the time
GUASTI
]
cumulato
unit
0,6 5
/
MEDIO DIOF
[repairs
0,5
cumulated
CUMULATIVE
NUMERO NUMBER
mediamente
0,4
3
0,3
AVERAGE
2
0,2
Affidabilità
Reliability R(t)
R(t) == 11 -- F(t)
F(t) 1
0,1
0
0,0
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
0 50.000 100.000 150.000 200.000 250.000 300.000 350.000
PERCORRENZA, kmhours
MILEAGE, km (or (o tempo o cicli)
or cycles)
MILEAGE, km (or
PERCORRENZA, kmhours
(o tempo o cicli)
or cycles)
f(t)CURVE
& h(t) f(t)
functions
e h(t) l function
Curva = = dM(t)/dt
dM(t)/dt
GUASTO RATE
1,2E-05 8,0E-05
FUNCTION
1,1E-05
DIHAZARD
7,0E-05
GUASTO
DI GUASTO
1,0E-05
km)]
6,0E-05
xx km)]
9,0E-06
DENSITY
x km) ]
h(t) della sola componente esponenziale (inconvenienti casuali )
e TASSO&
ZA DI
8,0E-06
FUCNTION
ZA
(unità
5,0E-05
// (unità
7,0E-06 FREQUEN
[ repairs / (unit
DI FREQUEN
PROBABILITY
DI PROBABILITÀ
6,0E-06 4,0E-05
[inconvenienti
[inconvenienti
DENSITÀDENSITY
5,0E-06
3,0E-05
4,0E-06
INTENSITÀ
3,0E-06 2,0E-05
PROBABILITY
FAILURE
2,0E-06
1,0E-05
1,0E-06
0,0E+00
0,0E+00
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000
PERCORRENZA, kmhours
(o tempo o cicli) PERCORRENZA, km
kmhours
MILEAGE, km (or (o tempo o cicli)
cicli)
or cycles)
MILEAGE, km (or or cycles)
These slides can not be reproduced or distributed without permission in writing from the author. 139
3.6.1. R(t), F(t) and M(t) functions
If all tested vehicles have covered the same distance, the con-
struction of the M(t) function does not present any problem.
Remembering that it represents the average number of failu-
res per vehicle cumulated at a given mileage, we see that it
is enough to detect the number of failures up to the assigned
mileage and divide it by the total number of tested vehicles.
The value so calculated is a point on the diagram. This calcula-
tion is then repeated at specified mileages and the points so
determined can then be connected by a curve, which is the M(t)
function.
These slides can not be reproduced or distributed without permission in writing from the author. 140
3.6.1. R(t), F(t) and M(t) functions
On the contrary, if the vehicles cover different distances (e.g.
because they are needed for other tests, or because they had an ac-
cident, or simply because we accept a lower confidence at highest
mileage in order to progressively reduce the sample and thus the
costs), then the calculation is a little more complicated and we
have two alternatives depending on the situation.
If we operate manually (without computer), we must develop
the calculation by mileage classes (see next slides).
If we use a computer, we can make a calculation quite similar
to that described above, except to evaluate a new point on the
diagram each time the number of tested vehicles changes
(= mostly decreases), rather than at the end of every mileage
class. In this way, the number of calculations to be performed
considerably increases compared to the manual method (but
this is irrelevant because it is the computer that makes them)
and the accuracy of results is significantly improved.
These slides can not be reproduced or distributed without permission in writing from the author. 141
0,5 3.6.1. R(t), F(t) and M(t) functions
for other uses during the test. The diagram below shows mileage
2
2
0 0
N° vetture oltre
le varie soglie di 8 7 7 6 4 3 0
percorrenza
NUMBER OF CARS
These slides can not be reproduced or distributed without permission in writing from the author. 142
3.6.1. R(t), F(t) and M(t) functions
At the conclusion of the test, we can distribute the total failures detected in
the same mileage classes of the previous diagram, as shown in the
histogram below:
4
3
3
2
2
1 1 1
1
0 0
Of course, a cumulative curve as the dotted one above would be misleading, be-
cause it does not take into account that the number of cars (= denominator) becomes
progressively smaller and smaller with the progress of the test and it is therefore ex-
pected a slope much steeper than the almost linear one, shown in the Figure.
These slides can not be reproduced or distributed without permission in writing from the author. 143
3.6.1. R(t), F(t) and M(t) functions
These slides can not be reproduced or distributed without permission in writing from the author. 144
2,0
1,5
3 / 7.5 =1,00.400
0.125 + 0.400 = 0.525
0,5
5
1 / 8 = 0,125
4
3
3
2
2
1 1 1
1
0 0
N° vetture oltre
le varie soglie di 8 7.5 7 7 6 4 3 0
NUMBER OF CARS
percorrenza
1,5
1,0
The last segment may be
less reliable
0,5
for the small number
of cars remaining in test.
km Incrementi
Increments Cumul.
5 A nuovo : 0 1/8,0 = 0,125 0,125
4 5.000 3/7,5 = 0,400 0,525
10.000 1/7,0 = 0,143 0,668
3
3 15.000 2/6,5 = 0,308 0,976
2 20.000 1/5,0 = 0,200 1,176
2
25.000 0/3,5 = 0,000 1,176
1 1 1
1
30.000 1/1,5 = 0,667 1,842
0 0
N° vetture oltre
le varie soglie di 8 7.5 7 7.0 7 6.5 6 5.0 4 3.5 3 1.5 0
NUMBER OF CARS
percorrenza
These slides can not be reproduced or distributed without permission in writing from the author. 147
3.6.1. R(t), F(t) and M(t) functions
Points so calculated may then be interpolated, for example with a cubic
2,0
1,8
1,6
1,4
Failures/vehicles
1,2
1,0
0,8
0,6
0,4
y = 1 . 10-13 . x3 - 5 . 10-09 . x2 + 0,0001 . x + 0,1118
R2 = 0,9721; r = 98,6%
0,2
0,0
0 5.000 10.000 15.000 20.000 25.000 30.000
Mileage [km]
These slides can not be reproduced or distributed without permission in writing from the author. 148
3.6.1. R(t), F(t) and M(t) functions
CONFIDENCE INTERVAL (CONFIDENCE BAND)
FOR M(t) CURVES
Denoting by:
n = the total number of tested units until time t
T = n.t = the equivalent total test time
r = the total number of failures detected at time t
M(t) = the average number of failures per unit cumulated at time t
1-a = the one sided (= unilateral) Confidence Level (a = significance)
c2 = the inverse chi-square function,
the Confidence Interval Lower and Upper Limits for a test (which may be truncated
at whatever time t) are given by the following expressions:
M(t) M(t)
M(t)L = ---------- ; M(t)U = ----------
KU KL
where KU and KL may be found in specific Tables or calculated as shown in the next
slide.
These slides can not be reproduced or distributed without permission in writing from the author. 150
3.6.1. R(t), F(t) and M(t) functions
CONFIDENCE INTERVAL (BAND) FOR M(t) CURVES: calculation of KU and KL
2.r 2.r
KU = ------------- ; KL = -------------
c2{2r; 1-a} c2{2r+2; a}
The previous expressions are easily understood by remembering that the c2 values of
the denominators represent respectively the Lower and Upper Limit of the Confidence
Interval for the expected number of failures in the specific conditions.
These two expressions become meaningless when no failures have been detected,
i.e. when r = 0.
In this case, only the Confidence Interval Upper Limit is meaningful and it can be
calculated by the following expression:
c2{2; a}
M(t)U = ---------- for r = 0
2.T
These slides can not be reproduced or distributed without permission in writing from the author. 151
3.6.1. R(t), F(t) and M(t) functions
CONFIDENCE INTERVAL (BAND) FOR M(t) CURVES: final discussion
From the previous expressions (with r ≠ 0), we obtain the following expression for the
Confidence Interval Upper Limit:
M(t) c2{2r+2; a}
M(t)U = -------- = -------------- . M(t)
KL 2.r
It’s evident that the factor which multiplies M(t) is not (directly) related to the sam-
ple size, but only to the number of failures (and, of course, to the assigned Con-
fidence Level). Therefore the Confidence Interval is proportional to:
to the mean value, M(t), of the cumulative failure frequency, which de-
pends on both the number of detected failures and the number of tested units(1)
(as it is their ratio) …
… and to a coefficient, which only depends on the number of detected failu-
res (through r and c2) and on the required Confidence Level (through a).
The number of detected failures may be calculated as the product of the mean
value M(t) by the number of units in the sample (= sample size)(2): therefore, when
the sample size is the same (i.e. constant), smaller values of M(t) have propor-
tionally larger Confidence Intervals.
(1) Of course, the number of tested units affects the accuracy of the M(t) estimate, but not directly its value.
(2) This product is equal to the product of the Probability Density Function, l, multiplied by the time t (e.g.: km, hours, cycles)
and then multiplied by the number of tested units.
These slides can not be reproduced or distributed without permission in writing from the author. 152
3.6.1. R(t), F(t) and M(t) functions
These slides can not be reproduced or distributed without permission in writing from the author. 153
3.6.1. R(t), F(t) and M(t) functions
Usefulness of the pseudo bathtub curve
mortality
FAILURE FREQUENCY DENSITY l
Failures
[repairs / (cars x 100 km)]
3.6.2
Weibull Chart
and “failure law”
These slides can not be reproduced or distributed without permission in writing from the author. 155
3.6.2. Weibull Chart and “failure law”
One of the strong points of Weibull distribution is the variety of shapes that it
can assume as a function of different values of b.
These slides can not be reproduced or distributed without permission in writing from the author. 156
3.6.2. Weibull Chart and “failure law”
b = 1 random failures
These slides can not be reproduced or distributed without permission in writing from the author. 157
3.6.2. Weibull Chart and “failure law”
Truncated sample
Having a software
program, it is easy to
estimate the
“confidence band”.
Note that
the more reliable
points are
the central ones,
because they are
confirmed
by the preceding and
by the following points.
These slides can not be reproduced or distributed without permission in writing from the author. 158
Weibull Chart
sheet
These slides can not be reproduced or distributed without permission in writing from the author. 159
3.6.2. Weibull Chart and “failure law”
3.6.2.1
Weibull knee
These slides can not be reproduced or distributed without permission in writing from the author. 160
Sometimes it may be convenient to represent two laws of failure
on a single graph: this is known as a Weibull with a knee.
0,90
0,000006
0,80
PROBABILITÀPROBABILITY
DENSITY
CUMULATA
DENSITÀ DI PROBABILITÀ
0,000005
0,70
0,60
0,000004
PROBABILITY
0,50
CUMULATIVE
0,000003
0,40
0,30
0,000002
0,20
0,000001
0,10
0,00 0,000000
0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000 275.000 300.000 0 25.000 50.000 75.000 100.000 125.000 150.000 175.000 200.000 225.000 250.000 275.000 300.000
PERCORRENZA
COVERED [km][km]
DISTANCES PERCORRENZA
COVERED [km] [km]
DISTANCES
In the first few kilometers, the wear-out is “latent” and does not produce perceptible manifestations,
so that the infant mortality failure mode (red curve) is dominant,
but after a certain distance (about 110,000 km), is clearly predominant
the wear-out failure mode (blue curve), overwhelming the effect of infant mortality,
now completely negligible: remember that the ordinate is double-logarithmic!
These slides can not be reproduced or distributed without permission in writing from the author. 162
3.6.2.1. Weibull knee
Cover gasket
We underline that:
the classical (and correct) use of the Weibull straight line
is with only one failure mode (component/defect);
for this reason, we say that it represents the “failure law”
by means of its slope (or exponent) b.
These slides can not be reproduced or distributed without permission in writing from the author. 164
3.6.2. Weibull Chart and “failure law”
3.6.2.2
Bx = n [km, hours, cycles]
These slides can not be reproduced or distributed without permission in writing from the author. 165
3.6.2.2. Bx = n [km, hours, cycles]
These slides can not be reproduced or distributed without permission in writing from the author. 166
3.6.2. Weibull Chart and “failure law”
3.6.2.3
Minimum number of “failed”
components
to draw a straight line
on a Weibull Chart
These slides can not be reproduced or distributed without permission in writing from the author. 167
3.6.2.3. Min. of “failed” components to draw a line on a Weibull Chart
A thumb rule,
based on common sense and almost universally recognized,
is that, to draw a straight line on Weibull Chart, it is essential
to have at least 5 points, i.e. 5 “failed” components.
Working with only 4 “failed” components
poses some risks not to be underestimated,
while a smaller number is absolutely unadvisable.
Although not explicitly mentioned,
this rule is broadly followed in Literature,
where most of Manuals rarely provide
Tables or Diagrams with less than 5 failed components.
Therefore, in designing the tests,
we must plan
a number of components great enough
to generate the required 5 failed components,
in an acceptable time.
These slides can not be reproduced or distributed without permission in writing from the author. 168
3.6.2. Weibull Chart and “failure law”
3.6.2.4
3-parameters
Weibull distribution
These slides can not be reproduced or distributed without permission in writing from the author. 169
3.6.2.4. 3-parameters Weibull distribution
These slides can not be reproduced or distributed without permission in writing from the author. 170
3.6.2.4. 3-parameters Weibull distribution
These slides can not be reproduced or distributed without permission in writing from the author. 171
3.6.2.4. 3-parameters Weibull distribution
m = t0 + h . G(1 + 1/b)
These slides can not be reproduced or distributed without permission in writing from the author. 172
3.6.2.4. 3-parameters Weibull distribution
Graphical construction
There are situations where, if we use the expression of
2-parameter Weibull, the points in the Weibull Chart are
no longer arranged along a straight line, but along a
curve with the concavity downwards.
These slides can not be reproduced or distributed without permission in writing from the author. 174
3.6.2.4. 3-parameters Weibull distribution
These slides can not be reproduced or distributed without permission in writing from the author. 175
3.6.2.4. 3-parameters Weibull distribution
The graphical method (rather rough but easy and straightforward) simply determines
the vertical asymptote of the curve interpolating points whose intersection with the time
axis identifies the value of t0. In this case, t0 is estimated at around 2,300 km.
It is simply
to draw a curve!
t0
These slides can not be reproduced or distributed without permission in writing from the author. 176
3.6.2.4. 3-parameters Weibull distribution
To obtain an estimate of t0 and thus be able to use the 3-parameters Weibul, we can
employ an analytical procedure, as follows.
1. We choose the two extreme points,
A and B, on the Weibull Chart and
B recognize the corresponding values
of abscissa: tA 2,500 km and
tB 35,000 km.
D/2
2. We determine the midpoint M of the
M distance in mm between A and B,
D/2
i.e. of the segment A’ — B, regar-
dless of the double-logarithmic scale
of the ordinates and we find its ab-
A A’ scissa value equal to tM 4,500 km.
3. We can now apply the expression:
(tB - tM) . (tM - tA)
t0 = tM - ---------------------------
tA tM tB
(tB - tM) - (tM - tA)
4. Once obtained t0 = 2,360, we can
use the expression of the 3-para-
meters Weibull, obtaining a shape
parameter b = 0.933.
These slides can not be reproduced or distributed without permission in writing from the author. 177
3.6.2.4. 3-parameters Weibull distribution
The preceding Weibull Chart, on the left, is compared with that obtained by the 3-pa-
rameters expression, where it is evident the best alignment of the data.
These slides can not be reproduced or distributed without permission in writing from the author. 178
3.6. Main indicators of the reached reliability level
3.6.3
The most common
Reliability indicators
(MTTF, MTBF, Bx, etc.)
These slides can not be reproduced or distributed without permission in writing from the author. 179
3.6.3. The most common Reliability indicators
As before,
it is appropriate to distinguish between
non-repairable and repairable
systems/components.
These slides can not be reproduced or distributed without permission in writing from the author. 180
3.6.3. The most common Reliability indicators
“Non-repairable” systems
(1) The fractions of failed parts are inferred from experimental results (Descriptive Statistics), but, in their
place, it can also be used a probability estimate (Statistical Inference), especially in the forecast.
These slides can not be reproduced or distributed without permission in writing from the author. 181
3.6.3. The most common Reliability indicators
“Non-repairable” systems
2,50E-05
Exponential (beta = 1,0); MTTF
Esponenziale distribution MTTFkm
≈ 43.611
(b=1,0); ≈ 43.611 km
1,0
h
h ==constant = 50.000
cost. = 50.000 kmkm
First
Primi failed parts with exponential distribution
rotti esponenziale
Non-expon. distr. (b=0,6): infantinfant.; MTTF≈ ≈35.147
mort.;MTTF 35.147kmkm
f(t)==dF(t)/dt
Non-espon. (beta = 0,6): mortalità
dF(t)/dt
0,9
(for all
(per tutte 3 curves)
e tre le curve) First
Primi failed parts with non-exponential
rotti non-esponenziale distr. (inf. mort.)
di mortalità infantile
F(t)
0,8 First
Primi failed parts with non-exponential
rotti non-esponenziale di usura distr. (wear-out)
Probabilitàprobability,
0,7
density, f(t)
1,50E-05
Densità di probabilità,
0,6
0,5
Cumulative
1,00E-05
0,4
Probability
Esponenzialedistribution
Exponential (beta = 1,0); MTTFMTTF
(b=1,0); ≈ 43.611
≈ 43.611 km km
0,3
First
Primifailed parts with exponential distribution
rotti esponenziale
Non-espon. distr.
Non-expon. (beta = 0,6): mortalità
(b=0,6): infant.;
infant mort.; MTTF ≈ 35.147
MTTF ≈ 35.147
km km 5,00E-06
0,2
First
Primifailed parts with non-exponential
rotti non-esponenziale distr.
di mortalità (inf. mort.)
infantile
10.000
20.000
30.000
40.000
50.000
60.000
70.000
80.000
90.000
100.000
110.000
120.000
0
10.000
20.000
30.000
40.000
50.000
60.000
70.000
80.000
90.000
100.000
110.000
120.000
Percorrenza [km]
Mileage [km] Percorrenza [km]
Mileage [km]
1,0E-04
Esponenziale (beta = 1);
Exponential distribution MTTFMTTF
(b=1,0); ≈ 43.611
≈ 43.611
km km
= f(t)/R(t)
If failures (whose times correspond to the 8,0E-05
= f(t)/R(t)
Non-esponenziale = 3); MTTF
(beta wear-out;
Non-expon. distr. (b=3,0): ≈ 50.250
MTTF ≈ 50.250
kmkm
Primi rotti parts
First failed non-esponenziale di usura
with non-exponential distr. (wear-out)
h =hconstant
= cost. == 50.000 km
durability, because we are dealing with non- 7,0E-05
(fortutte
(per all 3e curves)
tre le curve)
5,0E-05
di rate,
wear-out. 0,0E+00
0
10.000
20.000
30.000
40.000
50.000
60.000
70.000
80.000
90.000
These slides can not be reproduced or distributed without permission in writing from the author. Percorrenza [km]
Mileage [km] 182
3.6.3. The most common Reliability indicators
“Non-repairable” systems
NOTE - As we are dealing with non-repairable systems, we should talk only ever about dura-
bility. In fact, the indicator Bq can also be applied to reliability failures, especially if they are
of a single type (defined by the coupled “component-fault”), provided that we accept to count
always only the first failure of each test unit. If the unit is then repaired and can be considered
"like new", it can be put on test again, but as if it were a new (additional) unit, which is added
to those still being tested, starting again from t = 0.
These slides can not be reproduced or distributed without permission in writing from the author. 183
3.6.3. The most common Reliability indicators
“Repairable” systems
These slides can not be reproduced or distributed without permission in writing from the author. 184
3.6.3. The most common Reliability indicators
MTTF(t) Mean Time To Failure, which MTBF(t) Mean Time Between Failures,
is the mean time to the first which is the mean time be-
(and single) failure (the mean tween two successive failu-
is calculated with reference to res. It is used only for repaira-
all tested parts). MTTF coinci- ble systems and varies along
des with the average durabi- the time axis, due to the chan-
lity, since here we are within ge of the nature of the failures,
the non-repairable systems which pass from those of infant
and of course, in this case, it mortality to the random ones
is not time dependent: but it and finally to those of wear-out.
varies over time if it is referred MTBF(t) is widely used in the
to the "survivors" at time t. It is United States as the y-axis in
mainly used for non-repaira- the diagrams of Reliability
ble systems (or similar: e.g. Growth Testing (see Par. 5.3);
car-stops), but also with repai- while Italians prefer to use its
rable systems, when the time reciprocal, l(t): see next slide.
to the first failure (on compo-
nents assembled in the facto-
ry) is different from that be-
tween the successive failu-
res (repaired by the work-
shops of Customer Care).
These slides can not be reproduced or distributed without permission in writing from the author. 185
3.6.3. The most common Reliability indicators
“Non-repairable” systems
These slides can not be reproduced or distributed without permission in writing from the author. 186
3.6.3. The most common Reliability indicators
“Non-repairable” systems “Repairable” systems
1 1
l(t) = ------------- general expression lR(t) = m(t) = -------------
MTTF(t) R=
MTBF(t)
repairable
If the failures distribution follows the exponen- which represents the density of the number of
tial law (Weibull with b = 1), l coincides with failures per unit, at the examined time. Of
the constant value of the hazard rate h(t): course, in a given time interval, the reciprocal of
the mean value of MTBF(t) represents the num-
1 1 ber of failures per unit, lR(t), on average found
l(t) = ------------- = h(t) = ---- = cost. (or detectable) in the interval. The limits of
MTTF(t) h MTBF(t) and of lR(t) when the interval width
tends to zero, around an assigned value of t,
where h is the characteristic life. Hence the represent the corresponding “instantaneous”
expression of the cumulative failure distribution values. This is the l that will be used in the
may be written in the following most common Reliability Growth Testing (see Par. 5.3).
form, which uses l in place of 1/h:
Note that, in these situations (in contrast with the "non
F(t) = 1 - e- l t repairable systems"), both MTBF(t) and lR(t) vary with
time, even according to distributions other than
exponential. It no longer makes sense to talk of ha-
zard rate, because each tested unit may have more
than one failure and then we no longer speak of frac-
tions of "broken" units (which gradually conclude the
test, as already mentioned with regard to the M(t) cur-
ve in Par 4.1.1).
These slides can not be reproduced or distributed without permission in writing from the author. 187
3.6.3. The most common Reliability indicators
These few words, taken from the wonderful book by prof. Birolini(1), reveal the
existence of a sort of overlapping-area between repairable and non-repairable
systems. For further details, please refer to the mentioned book.
(1) Alessandro Birolini - RELIABILITY ENGINEERING: Theory and Practice, Fourth Edition, Springer, 2004
These slides can not be reproduced or distributed without permission in writing from the author. 188