Sunteți pe pagina 1din 12

Paper No.

CORROS90N97
322

A KNOWLEDGE-BASED SYSTEM FOR MATERIALS SELECTION

K.R Trethewey, Y. Puget, RJ.K. Wood,


Department of Engineering Materials,
University of Southampton, Highfield, Southampton S017 lBJ, UK

P.R Roberge
Department of Chemistry and Chemical Engineering,
Royal Military College of Canada, Kingston, Ontario, Canada, K7K 5L0

ABSTRACT

Materials selection is a process obviously suitable for computerization. The growing list of
available materials makes a choice increasingly susceptible to the vagaries of human decision making, as
evidenced by the continued incidence of high corrosion costs and unacceptable failures. A methodology
for computerization of the materials selection process is described and evaluated by an example involving
the choice of a polymeric paint coating for seawater service.

Keywords: Knowledge-Based System, Materials Selection, Failure Analysis, Corrosion, Materials


Performance, Engineering Design, Coatings, Seawater.

INTRODUCTION

Poor materials selection continues to be a cause of costly corrosion failure [1] and with the great
increase in available materials, especially in the coatings industry, there is an obvious need for computer-
assisted tools known as Krrowledge-Based Systems (KBSs) to assist humans and increase their efficiency
[2], Designers and engineers who need to select materials have, in the past, sought information ffom a
plethora of paper reference sources. Increasingly, these sources are being converted to computerized
tools and the Internet is rdso becoming a focus for data acquisition. As yet, however, there have been few
attempts to create machine-intelligent tools for managing these tasks.

Copyright
W 997by NACE International. Requests for permission to publish this manuscript In any form, in part or in whole must be made in writing to NACE
International, Conferences Oivision, PO. Box 218340, Houston, Texas 77218-8340. The material presented and the views expressed in this
paper are solaly those of the author(s) and are not necessarily endorsed by the Association. Print@d in the U.S.A.
The way in which a human selects a material for an application is a complex matter and many
writers have attempt~ to deiine precisely the stages involved, in such a way as might be converted into
an algorithm for a computerised methodology. Recently, a method was published by Aahby [3] which has
been influential and may yet prove to be adopted as standard. The scheme described here attempts to
build upon that workj but approaches the task in three stages

. Stage I: Considers the task caretidly and identifies the most likely way in which the component
will ultimately fail. In this paper, the failure mode is called a diagnosis.

Stage II: Identities the materird properties most relevant to each diagnosis;

Stage III: Optimises the properties for best resistance along the path to that diagnosis

Ideally, a lifetime for the component within the system specification will be known. At another level,
factors such as cost, availability and fabrication methods enter the methodology.

As pointed out by Aahby [3], a significant problem is that, whilst we generally make materials
selection decisions on the basis of materials properties, most tasks require an optimisation of more than
one property. One of the most obvious examples is the strength-to-weight ratio of materials in high
performance applications. The task demands optirnisatio~ not of strength or density properties alone, but
of a ratio of the two. Thus, whilst strength and density remain fundamentrd properties defined by a
standard test and measured precisely in laboratories, strength-to-weight ratio might be called a complex
property composed of more than one flrndsmental property and not directly measurable itself. The
immediate attraction of the computer therefore is its power to instantly assess fundamental propefiies in a
complex way and provide on-the-fly assessments of combinations of tirndamental properties. This has
been achieved in a powertid software tool [4] which relies upon a very large database of materials and
numeric values of fundamentrd properties.

The Ashby approach breaks down in many situations, notably coatings technology, when actual
values of desirable properties are not available. In situations such as these, materials selection tasks are
accomplished based upon depth of experience of those humans making the decisions and it is here that
inappropriate materirds may be selected when, for example, the designer has limited experience or when
the operational envelope of a given system is on the edge of a particular knowledge domain. It is the aim
of the work described in this paper to explore the scope for a new family of KBSS which can work with
less certain domain knowledge but still offer the best advice on materals selection decisions. As an
example, the selection of a polymeric coating for flowing seawater service has been chosen and will
hereatler be referred to as the Application.

GENERAL STRATEGY

A KBS consists of a database to store the domain knowldege and an inference engine for
manipulation of it. The detailed structure and model for the database has been described in another paper
[5]. Knowledge is elicited from a domain expert and stored in tabular form The engine makes use of
Certainty Theory to extract that knowledge and combine it using mathematical relationships which result
in assessments of the Iiielihood that a particular material will fail.

322/2
STAGE k DIAGNOSIS IDENTIFICATION

Table 1 summarizes the environments, irrtluences and failure diagnoses to which an engineering
system is susceptible. Seven fundamentrd ftilure modes have been detined by ASTM , i.e. corrosio~
creep, deformatio~ fatigue, fracture, surface darnage, and wear/erosion. Dlagnoaes are based upon failure
modes but include synonyms which assist by broadening the language to include terms in everyday usage.
Thus, corrosion has its own specific definitions, but in other situations, terms such as chemical attack or
environmental darnage might be more appropriate. For the Application, the same fkilure modes are
relevant and may contribute to the overall probability of failure in varying degrees. Figure 1 summarises
this by means of a knowledge tree. Here, coated surfaces are identified as clad, dipped, electro-deposit~
aprayed, anodked and painted. Each of these will have its own associated failure modes. For painted
surfaces, they have beerr identified as the same aeven ASTM fundamental modes, replacing corrosion
with environmental damage.

Comparing the scheme shown in Figure 1 with the available diagnoses listed in Table 1 shows that
diagnosis dl 8 applies uniquely to fracture, whilst d16 could be applied to both creep and deformation.
Diagnoses d8 and d9 both apply to surface damage and to wear/erosion, whilst d2 and d23 both apply
uniquely to environmental damage.

For a coating in seawater, only those diagnoses pertaining to the liquid environment -7 out of 23
- need to be considered, i.e. d2, d8, d9, d16, dl 7, d 18 and d23. The relative importance of each mode is
now determined by a process called knowledge elicitation. In Stage I of the computerization process
expertise fkom a domain expert is stored in a Knowledge-Base in an interviewing exercise. Questions are
posed to the expert about the likelihood of various failure modes, thus:

Question 1: Hinv important is the [increasing/&creasing] effect of [Influence] of the [Ar&rcent


PhaseJ on the [Type) [Component), such as would lead to [Diagnosis - 1], [Diagnosis - 2]... ?

The computer moves stepwiae through the list and inserts the relevant terms for each combination
of environment, influence and diagnosis, In the Application, for d9, this translates into:

How important is the increasing effect of imoact load of the seawater on the polvmenc coatina suck
as would lead to fracture, deformation or inw act failure?

It is essential to maintain the credibility of the KBS in the eyes of the human with a process called
filtering. For example, more advanced systems will have carried out filtering actions based on previously
defined descriptions of the problem so that irrelevant questions are not asked, causing users to lose
patience with the computer. In the Application, our KBS deals only with combinations involving liquid
(seawater); solid (e.g. sand particles) and gas (i.e. air) as an adjacent phase have been filtered out for
simplicity. Other more complicated Applications would require consideration of these diagnoses too.

Filtering is also used by the computer to determine how many questions it will ask, and to keep
the human informed about progress through the interview. A human who does not know how long the
interview will take or how may questions he will be asked will lose confidence very quickly. Furthermore,
the knowledge elicitation process can be quite arduous, even for a well defined problem, and great care
must be exercised by the knowledge engineer to maintain the support of hk expert.
Severrd mathematical strategies are possible for KEN tlmctiou such as those based on probability,
timzy logic and Certainty Theory. The first two are not used in this work for reasons which are described
elsewhere. Certainty Themy was adopted for this paper and is described in detail in another source [6], as
is our use of it [7]. A pivotrd parameter is the measure of belief (MB) which is a number reflecting an
experts increased belief in a hypothesis, based on some evidence. An inverse is the measure of disbelief
(MD) number reflecting an experts increased disbelief in a hypothesis, baaed on some evidence. From
these, a certainty factor (CF) is a measure of net belief according to CF=MB-MD. Rules can be
employed for manipulating these parameters and the CF factor can be used to direct a search into
promising areas or terminate if an approach is unhkely.

Certainty Theory requires that the experts answers are converted to a scale for MB tlom +1 to O
with +1 representing complete belief. Using the MD inde~ the range would be from O to -1 with -1
representing complete disbelief An expert could be given a paper questionnaire and asked to insert
numbers into a table. The knowledge engineer might feel that the expert would tind it easier to give his
answers in percentages which are subsequently converted by the computer. Perhaps ideally, the expert
answers dkwtly on the computer and in plain language. Thus, the expert selects his answer to a question
from a plain language menu:
Definitely, Alnrost certainly... fiobably not, Definitely not
Table 2 lists the available plain Iangusge descriptors and associated index values. Thus, attached
to each plain language selection, but invisible to the expert, is a CF index value which the computer (i)
stores in a new table to hold the results of the knowledge elicitation intemiew ~]) processes firther in
firther steps. It should be noted that index values x3.8 are not used here because of a problem called
saturation in which, if a value of 1.0 is used, the total certainty propagates through the KBS and gives
poor discrimination between possible outcomes. Further discussion on the subject of saturation is beyond
the scope of this paper.

The apparent simplicity of such a scheme is deceiving. The phrasing of the question is vital in
eliciting the best answers from experts. If possible, a good elicitation will reduce ambiguities by the use of
clarification or advice statements, again drawn tkom the database tables. When the elicitation exercise is
carried out, experts tind the process very demardng, even though the questions and answers appear
simple. An elicitation exercise necessarily draws upon the entire spectrum of the knowledge domain.
Experts need to call upon every part of their experience to provide what they consider a satisfacto~
answer and often reahse their own deficiencies in some areas of the domain. Again, filtering is important
so that the knowledge domain is not too broad and it is usually necessary to consult with more than one
expert to obtain a more complete expression of the domain knowledge. When faced with a lengthy
elicitation interview, experts may need encouragement so that they feel they are contributing usetldly,
even though they may be dissatisfiedwith their answers.

RULES FROM UNCERTAIN KNOWLEDGE

In Certainty Theory, rules are defined in which evidence, E, contributes to a hypothesis, H. Thus,
using diagnosis d9 as an example, the evidence might be:

There is an increasing effect of impact load of the seawater

and the hypothesis might be:

322/4
The polymeric cting nillfml by fracture, defortion or impact frnlure

In general, therefore, we make this definition

~ IF E, THEN H, Certainty Factor = CF,

In the Application, this translates to:


m There is a [increosingJ effect of [Injluence] of the [A@zcent Phose]
THEN : The [Type] [Gwnponen~ wllfdl by [Diagnosis - 1], [Diagnosis - 2]...
CERTAINTY FACTOR. CF, (From Table 2)

Table 3 lists the results of an interview for the Application. Notice that, horn a list of 24 possible
diagnoses (Table 1), the filtering process reduces these to just seven. It can be seen fkom Table 3 that in a
flowing seawater situatio~ wear and erosion has been assessed as the greatest risk to a coating with
environmental damage also a strong possibility. Temperature effects are not normally considered a
problem unless the water rises consistently above 60 C.

STAGE ~ PROPERTHM AND THEIR CLASSIFICATION

It is a manageable and finite task to compile a database containing all possible materials
properties. It is a much greater task to compile a database containing actoal values of these properties for
all available engineering materials, although this has been attempted [4]. Table 4 is a summary of all
properties which may be relevant to materials performance. The list was compiled in conjunction with the
properties contained in the Cambridge Materials Selector [4] and developed slightly to relate more
closely to this work. Thus, for example, two classes of property were identified: first, those which are
intrinsic to all materirds and which might be considered fundamental in all application second, those
properties which are dependent on application. Within each of these two classes, types of property were
identified in order to assist in the logicrd structuring. As the example below will show, the list is probably
not exhaustive and requires development for each Application. The properties are denoted by Pk, for
1<k<q,

For the second phase of the task in which the importance of properties is considered, a new rule is
defined thus:

RULE 2: IF Ez THEN H, Certainty Factor= CFZ

Notice that the same Hypothesis is used, but different Evidence is obtained. In the Application, this
translates to:
m
There is a [increming/&creosingJ effect of the [hoperty]
THEN
The [Type] [Component) wllfail by [Diagnosis -1] or [Diagnosis - 2]...
CERTAINTY FACTOR CF, (From Table 2)

32215
Table 5 smmnari ses the results of a Rrrowledge Elicitation interview with an expert, in response to the set
of questions, as detlned by questions derived from Rule 2, for polymeric coatings only.

~ombination of CFs: Itrcrementallv aaruired evidence


To make the selection of material, a judgement could be made based upon the knowledge from Rule 1 or
Rule 2 independently. However, a better judgement is possible using the available knowledge from both
interviews, Tables 3 and 5, by combmtion of Rule 1 and Rule 2. Certainty Theory tells us that certainty
propagates through similarly concluded roles, i.e. multiple rules can be written to support a hypothesis. It
is natural for humans to feel more contident about a conclusion when evidence has been obtained from
more than one source. The technique whereby multiple rides are combhed is known as incrementally
acquired evidence. When working with CF index values, the rules of combination vary depending upon
the signs. Thus

CF ~(CF,,CF2) = CF] + CF2 * (1 - CFI); both >0

= CFI + CF2 * (1 + CFI); both <0

= (CF, + CF,)/(1 - min{lCF,l, ICF,[}); one <0

If other evidence or domain knowledge is available, this technique can be used iteratively to refine the
likelihood of the hypothesis, ~ and hence to gain more contldence in the computer prediction.
THE APPLICATION: POLYMERIC COATING FOR IMMERSION

In generic classifications such as that of Aahby [4] it is difficult to use language and definitions
which cover all possible applications, so for a particular application it is necessary to adopt slightly
different language and make small modifications to the properties considered. In the domain of coatings
technology, Munger is an international coatings expert of high renoun [8], therefore, his work was
consulted for the Application used here. Munger lists numerous properties which he considers important
for selecting the best coating and wbic~ for mmparison, are also listed in Table 4. It can be seen at once
that, rdthougb terms differ slightly, the properties themselves compare dkectly with the generic properties
listed by Aahby. Because of the strong focus on the Application, additional properties are suggested, such
as the cosmetic property, p39, which is almost always of importance in paints. Some properties acquire
new attributes when used in a particular application. Thus, dm pick-up is associated with appearance,
but might also be a timction of porosity, p33, or of surface roughness, P34. Cathodic disbandment is a
property which is generally found only in coatings technology.

When the elicitation process is carried out for Rule 2 in the context of the Application, the relative
importance of each property is evahrated in terms of each possible diagnosis. The same filtering process
which reduced the diagnoses from 24 to 7 operates on the properties, reducing them from 39 to 13.

STAGE HE OPTIMISATION

In a separate work [9], property vahres were sought for coatings which would illustrate the KBS
tiurction. It is noted that many sources of such data do not quote specific numeric vrdues but deal in
uncertain quantities. Thus, for example, the hardness of a specific paint might be quoted as very good
or moderate, whilst abrasion resistance might be given as fair. These uncertain terms need to be
translated into CF index vahres which compare with those obtained from the elicitation process. Thus a

322/6
paint with an excellent hardness might be expected to almost certainly not fail by the appropriate
diagnoai~ i.e. CF = -0.8, whilst a paint with a moderate hardness would be considered to maybe not
fail: CF = -0,4. Table 6 lists en evaluation of CF vrdues for properties of two possible marine matings:
chlorinated robber and coal tar epoxy. To decide which coating will give the best performance, the
certainty factors are recalculated using the rule of incrementally acquired evidence. Table 7 lists results
for two diagnoses, d8 (abrasive load Ieadiig to wearkrosion) and dl 6 (temperature). The data teUs us
that the coating will almost certainly fail by diagnosis d8, and almost certainly not fad by d16. It also tells
us that in most cases the coal tar epoxy should out-perform the chlorinated rubber paint, except in except
for tbe property of service temperature. However, the specification of the Application excluded
temperature as a stimulus. Thus, the computer is clearly able to discriminate between the two coatings for
seawater petiorrnance, something which the human expert already knew.
Chwersion to I%babilities
Humans at the present time are naturally aceptical that computers will ever perform in the way that is
claimed. However, it should be remembered that a good KBS will relate well with a hurnaq showing only
the type of details a human will want to see or understand, and it must provide answers at least as good
as the human could. Humans are much more concerned with probabilities for use in lifetime calculations
than with MBs and CFS so the final interface will show only plain language end a probability when
necessary. In principle, probabilities, Pr can be calculated using the equation, Pf = (1 + CF) / 2, but this
step is still by no means as clear as it might seem at thk early stage.
CONCLUSIONS

An embryonic model which builds upon the work of experts has been presented which shows how a KBS
for materials selection could be constmcted using Certainty Theory. For a given application it uses
knowledge elicited from the experts, combining it with a knowledge structure for materials, their failure
mechanisms and their properties. Although this work has focused on selection of a marine coating, the
model is generic and applicable to the entire spectrum of materials and applications.

REFERENCES
1. Trethewey, K. R. and P. R. Roberge, 1995, Corrosion Management in the 21st century, Br. Corr. .L
30(3), 192-197.
2. Trethewey, K. R. and P. R. Roberge, 1994, LlfetirnePrediction in Engineering Systems: The Irrtluence
of People, Materials and Design, 15(5): 275-285.
3. Ashby, M. F., 1995, Materials Selection in Mechanical Design, Butterwotih-Heinemamr, London.
4. Watson, M, Cebon, D., Ashby, M, Charlton, C., Chongj W. T., Cambridge Materials Selector, version
2.02 (1994), Omrrta Design Ltd, Cambridge, UK.
5. Tretbewey,K., Y. Puget, R.J.K. Wood, and Roberge, P., Development of a Knowldege-Based System
for the selection of seawater-resistant polymeric coatings. Materials and Design, submitted for
publication.
6. Durkin, J., 1994, &pert Systems: Design and Development, MacMillan Publishing Company, New
York, NY.
7. Trethewey, K. R. and Roberge, P. R., Design and Structure of Knowledge-Baaed Systems for Improved
Materials Performance, Corrosion Reviews, 1996, in press.
8. Munger C., Corrosion Prevention by Protective Coatings, NACE International, Houston TX, 1984.
9. De Rerrzo, D. J. (cd), Handbook of Corrosion Resistant Coatings, Noyes Data Corporation, p~k
Ridge NJ, 1986.
TABLE 1: Summary of Environments, Irrihrences and Diagnoses for System Performance

Environment A@cent I@aence L4rfrel Diagnosis - I Diagnosis -2 Lhagnasis-3


frase
P%tid .
Chemical Compowtron d, Chemical Attack Corrosion Envirnmnental
damage
Liquid Composition ChemicalAttack Corrosion Environmental
damage
Gas Composition ChemicalAttack Corrosion Environmental
darnage
Mechanical Solid Impactbad Fracture Deformation Impact Failure
ShearInad Fracture Deformation Shear Failure
AbrasiveLoad Wear Erosion
FatigueLoad Fatigue

Liquid AbrasiveLoad Erosion Wear Surface Darnage


ImpactLead Fracture Deformation Impact Failure
Gaa Impactf..oad Fracture Deformation Impact Failure
Physical Solid Magnetism MagrreticDamage
Radiation RadiationDamage
Temperature Creep Deformation Overheating
Pressure Explosion Implosion Fracture
Electricity ElectricalBreakdown

Liquid Temperature Creep Deformation Overheating


Radiation RadiationDamage
pressure Explosion Implosion Fracture

Gas Creep Deformation Overheating


Rsdi;tion RadiationDamage
Pressure Explosion Implosion
Biological Solid Comfroaition Biofording Environmental
darnage
Liquid Composition Blofouling Environmental
damage
Gas Composition Blofouliig Environmental

TABLE 2: Certainty Factors (atler Durkin [6]) TABLE 3: Certainty Factors from a Knowledge
Elicitation Interview with an Expert.
Plain language Certain~ di= diagnosis number (see Table 1).
descriptor Factor Diagnosis Certain@ Factor
D@ritelv not -1.0
d2 0,4
Almost ixtairdy not -0.8
probably not -0.6 dS 0,s
Maybe not -0.4 d9 0.s
unknown -0,2 to to.z d16 -0.4
Maybe 0.4
d17 -0.8
Probably 0,6
Almost certainly 0.8 d18 0.2
Detlritely 1.0 d23 0,6

322/8
TABLE 4
Classification of Materisls Properties

Cfrrss VW number Property porn Ad@ [4J) Property (r?omMwrger [8])

fntrtilc Physical pl atomic vohmre (average)


Jmrinaic Physical p2 energy content
Intridc physical p3 density
Intrinsic Mechanical@ hardness hardness
hrrrMc Mechanicalp5 tonghness direct impact resistance
hrtrirraic Mechanicalp6 Youngsmcdulus flexibility
Irstilc Mechanicalp7 ahear moduhra flexibility
Jsltrirric Mechanicalp8 yield strength
Inwlc Mechanicalp9 ductility
IntrirL4c MechanicalP1O endurance tinrit
ImrisrSc Mechanicalpl 1 tensile strength
fntilc MechanicalP12 compressive strength
hrtrirMc Mechanicalp13 Poissms ratio
hrtrirr.ic Mechanicalp14 brdk modulus
Intrinsic Mechanicalp15 10sscoefficient
Intilc Mechanicalp16 modulus of rupture
Intrinsic Thermal p17 specificheat
IrrtriMc Thermal p18 thermal expmraion thermal expansion
hstrinaic llrerrrrd p19 thermal conductivity
Inwlc Thermal p20 glass temperature
fntrinsic lfsermal p21 melting point
frrtrtilc rhcrmal p22 latent heat of iirsion
fntrirrsic Thermal p23 nraximnnrservice temperature maximum dry heat
fntrwlc Thermal p24 tilmum service temperature
fntrtilc Electrical p25 breakdownpntential
hrtrinsic Electrical p26 dielecuic constant dielectric constant
fntrinaic Electrical p27 resistivity
fntrtilc Electrical p28 power factor
Application Wear P29 wear resistmrce friction, abrasion resistance
Application Chemical p30 chemical resistance resistance to water, chemicals,
atmosphericenvironment (wearbering,
ageing, oxidation)
Application Geometrical p31 dimensions
Application Geometrical P32 shape
Application Fabrication P33 porosity
Application Fabrication P34 surface roughness
Application Fabrication p35 bund strength adhesion, cathodic disbandment
Application Economic p% cost
Application Economic P37 recycle fraction
Application Biological P38 biofouling resistance biofotding resistance
Application Cosmetic p39 aPPCSMItce,COIOW, dirt pick-up

32219
TABLE 5
Certainty Factors from a Knowledge Elicitation Interview with an Expert.
di = diagnosis number (Table 1); pk = property number (Table 4)

enwroarnenral
abrasive irnpaceternperarureradiation pressure biofording
attack knldlaad
Label Properly d2 d81i9 d16 d17 d18 d23
@ hardness -0.6 0.8 0.8 -0.8 0.0 0.7 -0.8
ps toughness -0.6 0.5 0.6 -0.8 0.0 0.7 -0.8
@ Youngsmoduhrs -0.6 0.5 0.6 -0.8 0.0 0.6 -0.8
p7 shear modulus -0.6 0.5 0.5 -0.8 0.0 0.6 -0.8
p23 maximum service temperature -0.6 4.6 -0.6 0.8 0.3 -0.8 -0.2
p29 wear resistance -0.6 0.8 0.2 4.8 0.0 -0.8 -0.8
p30 environmental resistance 0.8 -0.6 -0.6 0.6 0.0 -0.8 0.2
p31 dimensions -0.8 0.4 0.5 0.1 0.1 0.7 -0.6
P32 SIISW -0.8 0.4 0.5 0.1 0.2 0.7 -0.6
p33 porosity 0.2 0.5 0.5 0.1 0.0 0.6 0.2
p34 surface roughness -0.6 0.6 0.2 -0.6 0.0 -0.6 0.2
p35 inmd strength -0.6 0.7 0.5 0.5 0.0 -0.6 -0.6
P38 Lriofotimgresistance 0.1 0.0 -0.6 0.2 0.0 -0.6 0.8

TABLE 6
Certainty Factors that a coating will give good performance,
as determined by its property, for two different coatings systems.
Data derived from performance data contained in reference [9].

Lubel Property chlorinated coal tar


rubber epoxy
p4 hardness -0.2 -0.6
ps toughness -0.2 -0.6
p6 Youngs modulus -0.2 -0.4
p7 shear modulus -0.2 -0.4
p23 maximum service temperature -0.8 0.4
p29 wear resistance -0.2 -0.4
p30 chemical resistance -0.6 -0.8
p31 dimensions -0.4 -0.8
p32 shape 0.0 0.0
p33 porosity 0.0 0.0
p34 surface roughness 0.0 0.0
p35 bond strength 0.0 0.0
~38 blofouling resistance 0.0 0.0

322110
TABLE 7
Certainty Factors for coating performance considering two diagnoses.
The columns Iabelled elicit are incrementally acquired CFS from Tables 3 and 5.
The columns for the coatings are incrementally acquired CFSusing Table 6.

G%- abrasive load d16 - temperature

Label Property Elicit ChlonMted Coal Tar Elicit Chlorinated Coal Tar
rahter EPOXY rabber EPOXY
p4 hardness o.% 0.95 0.90 -0.88 -0.90 -0.95
ps to@mess 0.92 0.90 0.30 -0.88 -0.90 -0.95
p6 Youngs modulus 0.92 0.90 0.87 -0.88 -0.90 -0.93
p7 shear medrdns 0.92 0.90 0.87 -0.88 -0.90 -0.93
p23 maximum service temperature 0.03 -0.80 0.40 -0.88 -0.98 -0.80
p29 wesr resistance o.% 0.95 0.93 -0.88 -0.90 -0.93
p30 chemical resistance 0.00 -0.60 -0.80 0.33 -0.40 -0.70
p31 dimensions 0.84 0.73 0.20 -0.25 -0.55 -0.85
P32 shsp 0.84 0.84 0.84 -0.25 -0.25 -0.25
P33 porosity 0.88 0.88 0.88 -0.25 -0.25 -0.25
p34 snrfsce roughness 0.92 0.92 0.92 -0.88 -0.88 -0.88
P35 bond strength o.% o.% o.% 0.00 0.011 0.00
p38 biofouhg resistance 0.75 0.75 0.75 -0.25 -0.25 -0.25

322/11
Clad

3
Dipped
Electrodeposited
Sprayed
Ano d

Fatigue
on

4 Fracture
)

FIGURE ,1-

V Creep
Fatigue
Fracture
Deformation
Environmental Interaction
3 Surface Damage
/ Physical
roperties
u

FIGURE 2- Knowledge tree showing the types of properties which influence the wear and erosion
failure mode.

322/12

S-ar putea să vă placă și