application of a neural network for an improved control of the

Transcription

APPLICATION OF A NEURAL NETWORK
FOR AN IMPROVED CONTROL OF
THE METALLURGICAL PROCESS
C.Mapelli, M. Morotti, W.Nicodemi
Dipartimento di Meccanica of Politecnico di Milano
Abstract
The factors of influence involved in many metallurgical
problems are featured by a non-linear relation, so that their
control is not an easy task. On the other hand, the quality
of the metallurgical aspects of the product depends greatly
on the knowledge of the relation that can allow a successful
management of the metallurgical process.The technological
control of the process can take advantage from the
application of non-linear numerical methods that describe
and simulate some not easy-understandable behaviours of
the metallurgical systems. The neural network can be
applied to build an automatic procedure for the definition
of the productive parameters to obtain the desired results.
Moreover, the neural networks find not only reliable
relation for the forecasting task but also to develope a
speedy and reliable classification of the different production
cases to be treated.The neural network model here shown
is devoted to implement both these issues and has been
validated on the definition of the effect of the
electromagnetic stirring on the solidification structure of
a continuous casting machine, but it can be adapted to
treat also other metallurgical processes in the foundry
operations and so on.
Riassunto
I fattori di influenza coinvolti in parecchi processi metallurgici sono caratterizzati da
relazioni non –lineari, così che il lro controllo non è facilmente realizzabile. D’altra la
qualità metallurgica finale del prodotto dipende fortemente dalla conoscenza di una
relazione che ossa permettere la corretta gestione del processo. Il controllo tecnologico
del processo può quindi trarre vantaggio dall’applicazione di modelli numerici nonlineari, che descrivono e simulano i comportamenti di sistemi metallurgici caratterizzati
da relazioni interne tra i fattori non immediatamente comprensibili. Le reti neurali possono
essere applicate per costruire procedure automatiche finalizzate ad ottenere i risultati
desiderati. Inoltre, la rete neurale non solo è in grado di fornire risultati previsionali
attendibili circa l’esito di un processo, ma può consentire una veloce classificazione dei
diversi casi trattati. La rete neurale qui presentata consente di raggiungere entrambi
questi scopi ed i suoi risultati sono stati ritenuti affidabili dopo la sua applicazione al
problema della determinazione della struttura finale di solidificazione di billette colate
su macchina di colata continua con stirring elettromagnetico. D’altra parte un tale
strumento può essere adattato pure per trattare i processi metallurgici nelle operazioni
di fonderia.
INTRODUCTION
procedure to produce a metal product which respects the required
specifications. In the studied case the billet has to be featured by a
solidification microstructure which grants an efficient plastic deformation.
Thus, the adjustment of all the productive parameters to implement an
efficient process can assure a sucessful plant management. The neural
network approach seems to satisfy these needs in metallurgical problems
hard to be treated by more classical methods.
Two main types of models can be used to understand and describe a
technological process: physical models and statistical models. The neural
networks belong to the second category, but their performance is greater
than that of a linear regression with which most scientists are familiar,
because the neural networks show at least three strong aspects that
differentiate them from the classical linear or pseudo-linear regressions
[1,2]:
• the users are not constrained to choose a previous hypothesis about
the relationship among the factors;
• the neural networks can describe very well non-linear relationships
among the inputs factor and between these ones and the outputs.
The model proposed has a significant peculiarity: the use of a classifying
algorithm which is able to distinguish a case from another on the basis of
The effect of the electromagnetic stirring has many
similiraties with other metallurgical processes, that
is the non-linear relations among the factors of
influences and between these ones and the final
microstructure of the product. Moreover, the
complete treatment of many problems by means
of a precise physical formalism implies hard and
long time spending simulation that can never be
applied to the in-line control of the process,
because the development of the industrial
operation belongs to a shorter time scale than
the time spent by the computation procedure.This
situation makes the use of the physical model very
difficult for the application in the industrial practice.
An innovative approach that involves the
interaction among the different significant
parameters of the process is needed, so that the
operators can apply a rapid and automatic
22 - Metallurgical Science and Technology
the former learning task and on the main factors of influences that feature
the case itself.The classification procedure is implemeted by a specific module
based on the neural network as well and generally known as SOFM (Self
Organizing Feature Map) [3].
Structure of the developed neural networks
A neural network model is composed of different layers of nodes. Generally,
the first layer of nodes is the input layer, while the last one is the output
layer.These extreme layers are linked by a network formed by other layers,
named hidden layers, which are linked in a complex network that permits
the information transfer from a node to the other ones (fig.1). The input
data are multiplied by weights which characterize every single linkage
between the nodes (wi) and the sum of all the product becomes the
argument of a non-linear function, that in the case of the present model is
the hyperbolic tangent (hidden function):
h = tanh( ∑ w (j1) x j + k )
i
Fig. 1: Structure of the neural network with three layers [ ]
where k is a constant value, named bias value, and xj is the input value
coming from the former layer.
The weights of the successive layer eventually present are indicated as w(n).
The core of the method consists in the definition of the weights that link
the different nodes and the procedure for this definition is the training of
the neural networks, that needs some experimental tasks.The weights that
define the interactions between the factors are adjusted on the basis of the
experimental tasks, in which the value of the chosen factors of influence
(input factors) were known and the output values were measured.Through
a drawback propagating algorithm the network is trained by the computation
of the weights. In this case a genetic algorithm [4] of back propagation was
developed on the basis of a previous model proposed by Cormier and
Raghavan[5].
The correct number of nodes and layers is fundamental for the stabilization
of the weights of the networks. In the present case, the neural network is
composed of three layers (one input layer, one hidden layer and one output
layer) and the number of the nodes are defined by
an empirical procedure [6], provided that there
are six input nodes and one output node.
The aim of the development of a statistical
numerical model based on the neural network
method permits the suitability of its results to a
wide range of products and the possibility to
support the operators by an automatic
computation of the interesting aspects which in
this case are the size of the different structure of
the inner region of the billet solidified under the
action of the electromagnetic stirring.The correct
control of the inner solidification structure can
allow to increase the performance of the
productive plant by increasing the homogeneity
of the billet structure and to improve the
coordination between two different subjects
involved in the process, the steelmaking unity and
the rolling one that has to deform a material with
a known inner structure.
The approach described here allows the definition
of a mathematical model that can represent the
measure of the different area of the microstructure
conditioned by the electromagnetic stirring action:
the chill zone, the dendritic zone and the central
equiaxic zone.
The classifying step is needed in the architecture
of the designed model for the following forecasting
step. During the learning period the weights of
the network have been defined by means of the
back propagation genetic algorithm for every
experimental case that is introduced for fixing the
correct weights to be applied to the neural
network.The basic idea is that two cases are similar
if they have similar weights relating the input
factors of influence with the output.The condition
of similarity is determined on the basis of the
eulerian distance between the set of the weights
belonging to the different cases. If the weights of
two cases are represented in this way:
{w
{w
A
1
B
1
, w2A , w3A , w4A , w5A ....wnA }
, w2B , w3B , w4B , w5B ....wnB },
the eulerian distance is:
6
δ = ∑ ( wiA − wiB ) 2
i =1
where A and B are two different cases compared
by the difference between the homologous
weights. This expression has the properties of a
distance also in a multidimensional system because
it respects the following relations:
δ ≥0
∀wi
δ AB = δ BA
δ AB ≤ δ AC + δ CA .
If a case is found that has a distance from the other
cases under a limit value, this case will represent
well the behaviour of the nearest cases and it is
possible to call it a model item for every case.
The role of the SOFM is to define the set of weights
that represents a model item.The algorithm used
to implement the SOFM is known as the Kohonen’s
algorithm [3].The idea of this procedure is to start
from randomly chosen weights that characterize
some different possible model items that have
some initial relations with the experimental cases
featured by the measured value of the input factors
of influence and the correlated value of the output
parameter. After some iterations the randomly
chosen model items converge in a position within
the multidimensional weight space and these
movements permit them to characterize well the
adjacent cases. It is possible that two or more
randomized cases converge in the same position,
thus reducing the number of the model items.
The model items are constituted by the sets of
weights that characterize the whole population
has been introduced during the learning step.
When a case characterized by a particular input
set is introduced to the numerical model, the
particular weight of a single model item is applied
to that.The choice of the more suitable and reliable
model item to be applied to the introduced case
is operated by the comparison of the inputs of
the introduced case and the inputs of all the
experimental cases that have been used for the
training of the algorithm and for the definition of
the model items. The comparison between the
input values is still performed by the computation
of the Eulerian distance. When among the real
experimental cases used for the training the
nearest one to the case to be forecast is found,
the numerical model evaluates what is the
reference model item of that experimental case.
So, the set of weights of this model item are used
for the forecasting step of the introduced case to
be forecast.The numerical model performance can
be adapted by a continuous learning task richer
and richer in experimental samples to train the
neural network model and to enrich it of other
model items.
Fig. 2: Flow chart of the implementing software
Thus, this model is organized by two cooperating software modules
developed by a C++ language code, one devoted to the forecasting and
endowed with the SOFM procedure while the training procedure is
performed by the genetic algorithm. The peculiarity in the system
architecture is represented by the presence of a SOFM algorithm that is
able to distinguish several fundamental cases to which all the cases that can
undergo the treatment of the numerical model (fig.2).can be reconducted.
Experimental Procedure and validation
The shown model is validated by the observations developed on the effect
of the electromagnetic stirring on the solidification microstructure of the
round billets solidified within a continuous casting machine equipped with
that electromagnetic device.
The studied materials are four types of resulphurised steel (tab.1)
Table I. Chemical composition
of the four studied steels
Elements %wt
1
2
3
4
0,17
0,2
0,25
0,17
1,4
1,4
1,6
1,2
Si
0,26
0,27
0,35
0,4
P
0,013
0,012
0,014
0,013
S
0,02
0,025
0,035
0,015
Al
0,025
0,024
0,025
0,025
Ca
0,001
0,001
0,001
0,001
N2
0,008
0,008
0,009
0,007
C
Mn
different four main zones of the billet solidification
structure: the chill zone, the dendritic zone, the
dendritic-equiaxic zone and the central equiaxic
zone (fig.2, fig.3).
The measurements of the area of 250 etched billets
are performed after the Baumman’s etchings by
means of an image analyzer which permits to
measure the areas of interest.
The model has been trained on 210 samples and
then validated on the other 40 samples whose
significant input data are introduced to perform
the forecasting task devoted to define the area of
the different inner microstructure of the billets.
The results of the model are very satisfactory,
provided that the number of samples introduced
to train the neural network can be considered
smaller than that often used for these trainings
(fig.3, fig.4, fig.5).
EQUIAXIC CENTRAL ZONE
49
2
measured value (mm )
47
45
43
41
39
37
35
Fig. 2: A macrographic billet structure etched by the Baumman’s etchings
35
40
45
50
2
computed value (mm )
Fig. 3: Comparison of the measured and computed chill
zone area
DENDRITIC ZONE
110
measured value (mm 2)
100
90
80
70
Fig. 3: Example of a treated image of the billet surface in which the different main
zone are recognized: chill zone, dendritic zone, dendritic-equiaxic zone, central
equiaxic zone (from the outer side to inner side)
60
60
70
80
90
100
110
expected value (mm2)
The presence of the sulphur allows the application of the Baumman’s etchings
requiring the use of a 92% H2SO4 concentrated solution that points out
the sulphur prints on photography paper and permits to well recognize the
Fig. 4: Comparison of the measured and computed
dendritic zone area
The factors of influence chosen in these tasks are
the super-heat temperature of the continuously
cast steels, the total heat extracted from the billet
by the mould and the involvement of some
different parameters related to the chemical
compositions (%Mn, %C, %S, %Cr) which can
variate and influence the solidification structure
by modifying the liquidus temperature, the solute
CHILL ZONE
25
computed value (mm 2)
24
23
22
21
20
19
18
17
16
15
15
17
19
21
expected value (m m 2)
23
25
Table II. Average and variance of the difference
between computed and real experimenal value
during the validation task
Central
equiaxic zone
Dendritic zone
Chill zone
Average
-0,36
-0,71
-0,18
Variance
1,03
2,07
0,51
partition during the solidification and the thermal conduction of the
solidifying metal. The working parameters of the electromagnetic device
are mantained constant in all the observed cases, so they have not been
included in the factors of influence used in the neural network. I
The average value and the variance of the difference between the computed
and the expected values offer a good reliability for the numerical model
(tab.2).
However, this model can represent a good tool for the understanding and
the description involved in problems related to the control of other
metallurgical processes implemented by several industrial systems, provided
that the main factors of influence are known. If one of the input factors is
related to a less important factor of influences its neural connection linking
it to the other nodes should be made unactive after several iterations of
the training steps.
Fig. 5: Comparison of the measured and computed
dendritic zone area
CONCLUSIONS
REFERENCES
The neural network approach can be used
profitably to describe and control the evolution
of some metallurgical processes governed by
several non-linear relations among the main factors
of influence and between these ones. The other
great advantage of these method is that they are
short time spending, if compared with precise
simulation physical softwares that for this aspect
cannot always find a good and speedy application
needed in the indutrial practice.
The neural network software here developed and
validated on a steel making continuous casting plant
of a steelmaking unit shows an architecture that
is endowed with a SOFM algorithm which can
permit also a classification of the case used for
the learning and more important can recognize
what are the model items which well represent
the population in which the main possible cases
of the studied phenomenon can be classified.
[1] R. CALLAN - The essence of Neural Network, Prentice Hall Europe,
London (1999), p.1-50.
[2] H.K.D.H.BHADESHIA - Neural Networks in Materials Science, ISIJ
Int., 39, 10(1999), p. 966.
[3] R. CALLAN - The essence of Neural Network, Prentice Hall Europe,
London (1999), p. 59-80.
[4] Z.MICHALEWICZ – Genetic algorithm + Data structure = Evolution
Programs, Springer-Verlag, New York (1996), p.13-120.
[5] Z.MICHALEWICZ – Genetic algorithm + Data structure = Evolution
Programs, Springer-Verlag, New York (1996), p.337.
[6] A.FERRARI - Aspetti applicativi delle reti neurali artificiali, Franco Angeli,
Milano(1996), p.10-40.
CORRECT MANAGEMENT OF
INCLUSIONAL DEFECT BASED
ON A FAILURE ANALYSIS
M. Cusolito Rodacciai S.p.A.
C. Mapelli, W. Nicodemi - Dipartimento di Meccanica, Politecnico di Milano
Abstract
The non-metallic inclusions are metallurgical defect always present within the steel
bulk. Although high cleanliness levels can be reached by the most up dated techniques,
their presence is not avoidable, so the users of steel have to learn a correct management
of such defects. This problem has become even more significant with the use of
resulphurised steels, in which the sulphur is added to cause the precipitation of an
abundant MnS population, which makes the steel more workable by the cutting tools.
However, the presence of a not proper inclusional population can forbid the
implementation of certain technological processes to be applied, i.e. welding, hot working
etc.. Moreover, the sulphide inclusions improve the workability, but they remain within
the product also after the technological transformations and they can weaken the whole
material, so their use has to be decided only after a proper choice of the technological
route.Thus, some indications about this topic have been obtained on the basis of some
significant cases on which a failure analysis has been applied.
Riassunto
Le inclusioni non metalliche sono un difetto sempre
presente all’interno degli acciai, nonostante attraverso le
tecniche siderurgiche più moderne si possano raggiungere
elevati minimi livelli di presenza. Quindi gli utilizzatori di
acciaio devono apprendere la corretta gestione di tali difetti.
Tale problema è divenuto ancor più significativo a seguito
dell’avvento degli acciai risolforati a lavorabilità migliorata,
nei quali lo zolfo è stato aggiunto per provocare la
precipitazione di una consistente popolazione di MnS, che
rende questi acciai più lavorabili agli utensili da taglio.
Comunque, la presenza di una popolazione inclusionale
dalle caratteristiche non corrette in termini di dimensione
media, distribuzione e forma, può impedire l’applicazione
di par ticolari processi tecnologici, es: saldatura,
deformazione plastica a caldo ecc.. Inoltre, i solfuri, se da
una parte migliorano la lavorabilità, dall’altra rimangono
all’interno del materiale anche a seguito del processo di
trasformazione tecnologica e possono indebolire il
materiale, quindi il loro utilizzo può essere effettuato solo
in seguito alla scelta di una corretta procedura tecnologica.
Alcune indicazioni circa questo aspetto sono state ottenute
nel presente studio mediante l’analisi eseguita su alcuni
casi di avaria particolarmente significativi.
INTRODUCTION
Today it is possible to produce “clean steels”, i.e.
steel with an extremely low inclusion content. But
the production cycles are very expensive and in
consequence they are not affordable for generic
uses.
In this paper, based also on experimental evidences
founded on the failure analysis of some critical
components, we will define the concepts of
imperfection and defect and describe the origin
of the most common defects as, in particular, the
inclusions. At the end, we will present the most
recent trends of the acceptance criteria based on
the final application of the piece.
This paper aims to clarify the problems arising from inclusional compounds
that with the other frequent defects, like segregational imperfections or
defects in the metallic matrix of a steel product represent the most usual
and not avoidable defect within the steel products.
The designers and the users think that steels are perfect, completely
homogeneous and internally sound.The possibility to have an interruption
of the matrix is normally excluded, although it certainly takes place.
The reality in industrial production is not able to avoid a certain number of
imperfections and defects, more or less detrimental for the life of the
produced piece.
DEFINITION OF DEFECT
In the steel, the internal or surface discontinuities
are geometrical irregularities or unwanted nonhomogeneous areas.
These discontinuities are classified as
imperfections when their dimensions are equal to,
or minor of, the specified limit.They are classified
as defects if their dimensions is above this limit.
It is true that some imperfections can originate
failures even if they have a dimension defined as
“acceptable” by the designer: for example they can
act as notch [innesco] to create larger and more
important defects. But it is also true that a lot of
defects exist, with a much larger dimension than
the limits defined as acceptable, and they do not
create any consequence on the piece and can be
easily tolerated.
Origin of the defects
The defects in the steel can originate in any step
of the production cycle: melting, solidification,
cooling, hot deformation, heat treatment, cold
transformation, surface finishing.
Normally the defect forms during the melting and
the solidification, while they are modified during the deformation processes.
The defects caused by the first steps are internal or on the surface, but
those generated towards the end of the production are mainly superficial;
nevertheless important exceptions exist, especially related to the cold
deformation phases.
The inclusions are probably the most known defect of the steel because
they can be found in every material. They originate from different causes:
• external pieces of foreign bodies drawn into the steel by mechanical
action or chemical erosion
• not controlled chemical reactions that should have happened and didn’t
or should not have happened and did.
• modification of the solubility of some elements in the liquid bath with
the modification of the temperature.
One of the most important and frequent defect is represented by the nonmetallic inclusions, that in some cases are also intentionally produced with
particular chemical composition, solidification and cooling pattern. An
interesting case is constituted by the resulphurised or micro-resulphurised
steel, in which the non-metallic inclusions, generally regarded as an usual
defect, are developed to improve the cutting operations of the steels. A
good calibration of this task could be an important example to understand
the management of the defect in order to improve the desired mechanical
characteristics and to avoid the detrimental effects that the non-metallic
inclusions can generate too.
EXPERIMENTAL PROCEDURE
Four examples of the role of the non-metallic
inclusions have been studied starting from the
failure analysis developed on four different types
of mechanical components which have shown a
not proper behaviour during the exercise and the
technological operations, like cutting, plastic
deformation and welding.
The failure cases have occurred repeatedly in a
great number of samples produced and worked
in different conditions:
1. rupture along the extrusion axis in a
resulphurised steel (table I) component during
the drilling along the extrusion axis itself. This
type of piece has been heated at 1200°C, hot
extruded and then annealed to remove the
residual tension and the hardening effect before
the drilling operation (fig.1);
2. rupture of two types of steel samples (table II)
produced by hot forging operations performed
between 1200-1260°C. After forging the
samples have been austenitized between
=
>
Fig. 1: Failed sample of the hot extruded steel with crack developed along the
extrusion axis. Longitudinal view (a), transverse view with the hole produced by the
drilling operation (X1)
860°C-880°C, oil quenched and tempered between 540°-560°C. The
geometrical shape of these two types of failed samples are characterized
by axial-symmetry;
3. defects on the welding borders showing significant and not acceptable
irregularities.The two steels involved in the welding are X14CrMoS 17
and X5CrNi 18 10.
Table I. Chemical composition
of the hot extruded resulphurised steel (%wt)
%C
%Mn
%Si
%Cr
%Ni
%P
%S
0.31
0.68
0.5
12.32
0.21
0.019
0.06
TableII. Chemical composition of the two hot
forged steels (%wt)
Steel
%C
%Mn
%Si
%P
%S
Ca (ppm)
A
0.28
1.01
0.62
0.001
0.5
7
B
0.31
0.9
0.38
0.001
0.1
5
On the hot extruded steel 10 tensile tests have been performed according
to ASTM E8 in the not deformed state, in the deformed state, and in the
annealed state after deformation.The specimens for these tests have been
taken with their axis parallel to the extrusion one.
The study of the failed samples has implied to perform SEM analysis aided
by EDS (Energy Dispersive Spectroscopy) for determining the chemical
compositions of the non-metallic compounds involved in the fracture
process.The optical microscope has been used to acquire the images studied
by the image analyzer system to determine the ditribution, the shape and
the volume fraction of the observed non-metallic
compounds. On each of the hot extruded and
forged samples eighty photos have been performed
with a magnitude of X200 and the statistical
determination of the specified parameters has
been developed.
For a steel with the same composition of the
forged ones five tests for the determination of KIC
have been performed according to ASTM E1820
with a classical notch compact specimen tested at
the environment temperature (25°C).
The shape ratio has been determined by the
approximation of the non-metallic inclusion with
an ellipse whose major and minor axis have been
determined.The anisotropy has been evaluated by
the ratio between the major axis and the minor
one.
In the forged sample a chemical etching has been
operated to point out the stream-line of the plastic
flow.The chemical etching has been performed by
a chemical solution of 50%HCl and 50%H2O in
which the samples have been introduced for 17
minutes at 80°C.
RESULTS AND DISCUSSION
Fig. 2: Typical string of non-metallic inclusions in the extruded samples
Fig. 3: Typical string of non-metallic inclusions in the extruded samples (X400)
In all the failed samples the non-metallic inclusions
have played an important role. The feature
landscape of the inclusions of the plastically
deformed samples is quite different.
In the hot extruded samples the inclusions have
shown a significant anisotropy, because the shape
factor defined just before has an average value of
11.2 and a standard deviation of 5.7, while in the
not deformed annealed condition these
parameters have the value of 2.4 and 0.95
respectively. The measured percentage volume
fraction has an average value of 1.2% and a standard
deviation of 0.3%. The population of the nonmetallic inclusions is prevalently composed by MnS,
nearly pure, actually only an average value of 4%wt
(st.dev.1.7%wt) of Fe has been detected.The Ca is
practically absent because it is present in an average
quantity of 0.77%wt (st.dev.0.7%wt), so it can be
concluded that there are not CaO and CaS within
the non-metallic inclusions.The inclusions are not
homogeneously distributed within the metal
matrix, because they appear to be grouped like
strips involving several aligned non-metallic
inclusions (fig.2,fig.3) and they usually assume the
form of aligned clusters grouping an average
between 3 or 4 inclusions with some significant
exceptions (fig.4).The average area of the inclusions
is 83*10-6mm2 and shows an important standard
deviation of 93*10-6mm2.
Fig. 4: Cluster with a number of non-metallic inclusions over the average value. Only a
little number of these examples has been evidenced in the extruded samples (X400).
According to the classification proposed by
Gladman [1] the inclusions can be considered in
this state as holes, because they can be regarded
as discontinuities in the metal matrix. The string
shape of the clusters is probably due to the
elongation of an original inclusion fractured during
the plastic deformation of the metal matrix. As
this behaviour has been observed, it can be
concluded that the MnS observed belongs to the
II types [2]. From several studies [3,4,5] on the
mechanism of the ductile fracture the decohesion
occurring at the interface between the nonmetallic inclusions and the metal matrix (fig.5) has
been found to be the phenomenon ruling the
fracture process. The growth of such voids once
nucleated is highly dependent on the strain state
imposed. The rate of void elongation is increased
with respect to the longitudinal strain.
The void width is not sensible to the strain in a
plane tensile test, but it can decrease under biaxial
compressions like that occurring during the hot
extrusion. On the other hand, the ultimate failure
can occur when the adjacent voids link by shear.
The strain to fracture produced by void
Table III. Properties of the steel
in the not deformed state
R(MPa)
Rp0.2(MPa)
%A
315
270
16
Table IV. Properties of the steel
in the deformed-annealed state
R(MPa)
Rp0.2(MPa)
%A
320
290
11
Fig.5 Layout of the decohesion between the non-metallic
inclusion and the metal matrix under a stress in
longitudinal and transverse direction with respect to the
major axis of the inclusion itself (a, b 1 are the
longitudinal axis and transverse one with respect to the
tensile axis respectively; b is the width of the nucleated
void longitudinal to the tensile axis, R is the frontal
radius of curvature (b2/a))
coalescence generated by inclusion-matrix decohesion depends on the
inclusion parameters:
F2 k
 2 + 2
f
r
ε t = ε 0 + 0.5
k

 1+ r 2







where
εt true strain to fracture
ε0 strain for the initial void decohesion
F constant related to critical volume fraction of voids for catastrophic
failure
f initial volume fraction of the non-metallic inclusions
k stain void factor (about 2)
r average shape factor of the inclusions
The F constant for the hot extruded steel has been determined on the
basis of the tensile strength performed on the specimen of the not deformed
and deformed-annealed samples (table III, table IV) tested on the direction
parallel to the deformation one.
On the basis of these results and under the hypothesis that the strain for
the initial void decohesion has a not significant value so that it can be
disregarded, the F value has been determined to vary between 0.01340.0135.
But if the measured values are inserted taking into account the transverse
direction (in which the shape ratio assumes the value of 0.09) then the
ductility resources of the material has nearly completely eliminated, because
is reduced to 1.07% against the 19.6% proper of the longitudinal stress.
The drilling tool during the perforation induces a strength just in the
transverse direction with respect to the elongation of the non-metallic
Table V. Average chemical composition of the nonmetallic inclusions observed in the forged
samples (steel A and steel B) (%molar)
Steel
%Mn
%S
%Fe
% other
elements
A
37.7
35.6
15.2
11.5
B
43.2
44.1
7.8
5.2
=
>
Fig. 5: Streamlines flow of the forged samples for component (A) and component (B)
1.4
1.2
%Mn*%S solubility product
1
0.8
0.6
0.4
0.2
0
700
800
900 1000 1100 1200 1300 1400 1500 1600
temperature ( C)
Fig. 6: Variation of the solubility product of MnS
inclusions and this can bring the material to a fast
fracture process, as the ductility resources of the
material have been nearly completely removed.
The chemical composition of the non-metallic
inclusions observed in the forged samples have
revealed that the population is completely
constituted by MnS (table V).
But a main difference with the previously observed
samples can be found in the shape ratio which has
an average value of 2.2 and a standard deviation of
0.8, so that the anisotropy related to the nonmetallic inclusions is not significant as in the
previous case. This is really peculiar, because the
streamlines analysis has pointed out a great plastic
deformation flow (fig.5) which should induce a
great anisotropy on MnS inclusions, actually on the
basis of the nearly completely absence of CaS any
hardening effect on the inclusion structure should
be observed.The inclusions of two samples have a
similar average size of the area: 16*10-6mm2 (st.dev.
11*10-6mm2) and 14*10-6mm2 (st.dev. 12*10-6mm2).
On the contrary, the average number of inclusions
composing a cluster is higher and in the range
between 8 and 9.
The more little size and a more round shape of
the inclusions of the forged samples can be a clue
of the occurence of another important process
related to the solubility of the MnS within the
austenite and a new precipitation during the
following coolings processes in a form of string
composed by a greater number of non-metallic
inclusions featured by a more little size. Similar
processes have been described also by other
authors [6,7] but with respect to other and not
re-sulphurised materials with an inclusion
population of a more little size than that observed
in this study, however nowadays this represents
the most conceiving explanation of the observed
phenomenon. During the re-heating and
austenitizing treatment the solubility product of
MnS increases (fig.6)
%Mn %S=1/K where K=9281/T+5.19 [8]
and so part of the precipitated MnS can dissolve
again in the form of Mn and S.
The dissolved Mn and S have not a great diffusivity
within the solid state (in the order of 10-10 m2s1
[9]) so after the cooling the atoms of Mn and S
can meet again and precipitate in the form of
different and more little inclusions with a not
significant anisotropy.
However, the effect is not less dangerous, because
these adjacent non-metallic compounds constitute
a string in which the coalescence of the nucleated
voids is very easy and fast after the first growth.
As the non-metallic inclusions are really close, the
process of growth of dimples generated by the
development of the voids is decreased. Because
the part of energy spent for the growth of dimples
is greater than that spent in the process by the
coalescence process, this mechanism can facilitate
the development of the overall process leading to
the formation of a greater and not sustainable
defect. The sustainability of a defect is clearly
related to the KIC for the material. For the treated
material the KIC tests have pointed out an average
value of this parameter equal to 57MPam1/2. Under
the hypothesis of a mean stress of 270MPa:
KIC=δβ πa
where
σ is the applied stress
β is the shape factor (for this type of inclusion is
roughly 1)
a is the defect size
the steel can sustain an overall defect of 0.014m.
Because the mean length of the string is 32.5mm
only 443 elongated inclusion clusters in which the
dimples have coarsed can be enough to produce
the unstable propagation of the fracture within
the components. As the average volume fraction
of non-metallic inclusion for the two steels is 1.8%
(st.dev.0.09) for the steel A and 1.94% (st.dev.0.3)
for the steel B, there is a great possibility that the
unstable propagation of the fracture can occur.
Actually, considering the clusters like ellipses with
the found average length in a volume of only 6*105m3, it is possible to estimate that at least 793855 string shaped clusters can be theoretically
present.This means that only half of the estimated
amount can produce the unstable propagation of
the crack. It is true that some inclusions are not
oriented in a favorable direction to interact with
the applied stresses and some inclusions could not
be present in the form of clusters. On the other
hand, the performed measures have pointed out
that 75% and 69% for component A and
component B, respectively, are the ratio of the
inclusions present in the form of string cluster
and this appears certainly as a dangerous condition,
to which can be ascribed the happened and
repeated failure events occurred to these
components.
In the welding observed the non-metallic inclusions
forbid a good heat exchanging and their removing
promoted by the flow of metal can cause an
unacceptable defect due to a great irregularity of
the welding border (fig.7). This phenomenon can
be produced mainly by very great non-metallic inclusions (with one of the
dimension >25µm), otherwise the thermal shielding developed by the nonmetallic material of the non-metallic inclusion cannot develop its insulating
effect. If the heat flow is considered:
Q=2kinc(∆T)/(∆rinc)
where
Q
the heat flow
thermal conductivity of the non-metallic inclusion
kinc
∆T thermal difference between the metal pool and the metal border
Drinc inclusion dimension
it will be possible to estimate the effect of the presence of a non-metallic
inclusion of large size. The non-metallic inclusions with a higher size can
produce a decreased heat exchange fifteen times more little than that
exchanged in the absence of the insulating inclusion.This effect is amplified
by the more little thermal conductivity of the non-metallic material that
composes these compounds, based on SiO2 and MgO, which can be
evaluated to be between 4-10Wm-1K-1 [10].Actually the evidenced defects
cannot be ascribed to the MnS which has an average size of 50µm2 (st.dev.3
µm2) with a maximum linear size of 8.9 µm. Although the decreasing of the
thermal exchange, the removal of the non-metallic inclusion operated by
the convective flow in the metal pool [11,12,13] can rapidly remove the
non-metallic inclusions, eliminate the thermal insulation and permit the
adjacent liquid to compensate the void volume left by the removal of the
inclusion itself. In presence of a great non-metallic inclusion the adjacent
liquid cannot fill the pool, because the thermal insulation can avoid the heat
input needed to melt a sufficient quantity of steel able to fill the significant
volume made free by the inclusions.These inclusions are usually originated
from exogenous factors (powder covering the tundish, pieces of entrapped
refractory etc.). On the other hand, the dimensions of such inclusions are
not acceptable also because they can constitute the originating factor of a
possible fatigue fracture. They can be avoided by a better control of the
parameters of the production process.
Fig. 7: Defects on the welding border for a not complete penetration due to the
former presence of large non-metallic inclusions whose trace can be noted in the
central upper part of the welding border
CONCLUSIONS
According to the introduction of this paper and to the results of the
experimental procedures, it is evident that the possibility to perfectly define
the permissible limits of the defects contained in the steel (in particular the
inclusion content) could be a good solution in order to avoid praecox
failures of the pieces.
On the other hand, it is almost impossible to manage the form, the dimension
and the location of each single inclusion (otherwise they would not be
regarded as defects): their distribution is regulated by the statistic laws as
well as by the process parameters.
The modern plants for steel production and transformation work more
and more automatically: the process parameters are solely controlled and
the results are used for the adjustment of the process itself. Each anomalous
modification of the parameter results in an automatic rejection of the
material produced in the “out-of-control” condition; the results of the
examination of this material are also used to improve the process.
This is not enough: even if the number of the controlled parameters is
increased and all the parameters are optimised, one cannot be sure to
detect all the potential defects introduced by the process itself and, of
course, by the previous processes.
The results of the control are not completely reliable, both because the
measure are uncertain and because the processes themselves are variable.
In the following table a list of the more common controls is presented for
some siderurgical products together with the type of defect that can be
detected.
Each of the above mentioned controls have several limitations: for example
the ultrasonic control is not normally able to detect the defects just below
the “skin” of the material, while the eddy current controls are not able to
control the bar ends (50 mm each side), to detect the cracks that have
been “welded” by the hot rolling and to detect the “short” defects (machine
Table VI. Typical control and
defects for long products
Type of control
Defects that can be detected
Product: billets
destructive control of specimen
taken in particular zones
segregations, inclusions, internal defects,
cracks
Baumann test
sulphur segregations
acid attack
internal cracks, segregations
blue fracture test
macro-inclusion condition
ultrasonic control
macro-inclusion, internal defect, internal
cracks
visual control
only the larger defects and cracks
automatic control
limited possibilities to detect the defects
Product: hot rolled or cold finished bars
cut of samples and laboratory
examination
some defects of raw material: cracks,
segregation, decarburization
ultrasonic control
macro-inclusions, internal holes
eddy current
surface defects
with rotating probes) or long defects (machine
with magnetisation coils).
The possibility to perform a “good” control on a
semi-finished product depends on the type and
dimension of the defect that has to be detected,
on the technological limits of the control machine
and on the possibility to separate the defective part
(for example it is easy for a bar, but not for a coil).
A clear definition of the acceptability threshold
allows to limit the rejections to the really defective
pieces, i.e. those originating problems for the
specific application.
Too tight and not differentiated thresholds increase
the transformation and control costs without any
advantages for the final product.
Only a few standards define the type and the
dimensions of the permitted defects. In such cases,
they refer to other standards cataloguing the
defects and their evaluation criteria.
More commonly, the standards require that the
defects “detrimental” to the final use must be
avoided: this is obviously a poor definition, generally
not able to help the producers.
Moreover, the standards do not allow any mistake
in the control: no defect above the specified
threshold is permitted. This would imply perfect
process control together with perfect control
machines: as above described, this is technologically
impossible.
Only in the last period some European standards
[14] permitted a certain percentage of pieces with
defects above the specified limits (for example,
the standard for bright bars in relation to the
surface cracks).This is an important step in order
to consider the real technological limits.
Some standards limit the inclusion content (for
example the European standard for case-hardening
steels [15]), but almost no standard specifies the
limit for macro-inclusion content.
More generally, the control of the internal
soundness is left to the agreement at the enquiry
and order; but for the bars no reference standard
exists for the ultrasonic control with automatic
machines (the most widely used).
The control of the process and the control of the
semi-finished products cannot eliminate the
defects completely: some defects are not detected
and, in consequence, cannot be eliminated.
The situation is even worse for the inclusions,
where the distribution and the evolution of the
defects in the metallic matrix, as explained in the
previous part of this paper, is difficult to evaluate
because it depends upon several process
parameters.
In order to approach the “zero defect” philosophy
it seems necessary to control, under certain
circumstances, also some peculiar characteristics
of the final products. This kind of control is extremely expensive, is not
always effective for the inclusions, and cannot be performed by the
intermediate transformers on the semi-finished products.
REFERENCES
[6] Stachura, S.: Journal of Materials Processing Technology, 53(1995), 781797.
[7] Ito, J. Nasumits, N. Matsubara, K.: Trans. ISIJ, 21 (1981), 477.
[8] Cicutti, C.E. Madias, G. Gonzales, J.C.: Ironmaking and Steelmaking,
24(1997), 2, 155-159.
[9] Nastac, J.: Acta Materialia, 47 (1999), 17, 4253- 4262.
[10] Schneider, S.J.: Ceramics and Glasses, Engineered materials handbook
vol.4 ASM International, 1991.
[11] Vedani, M. : La Metallurgia Italiana, 10 (2001), 17-24.
[12] Limmaneevichitr, C. Kou, S. : Welding Journal, 8 (2000), 231.
[13] Limmaneevichitr, C. Kou, S. : Welding Journal, 5 (2000), 126.
[14] EN 19277-1, (1999).
[15] EN 10084, (1998).
[1] Gladman, T.: Ironmaking and Steelmaking, 19
(1992), 6, 457-463.
[2] Gonzales, J.C.: Metalurgia Moderna, 2 (1985),
1, 29-46.
[3] Henr y G. and Plateau J.: La
mircrofractographie, Edition Metaux IRSID, St.
Germain, 1957
[4] Gladman, T. Holmes, B. McIvor, D.: Effects of
second phase particles on the mechanical
properties of steel, 68, 1971, London, The Iron
and Steel Institute
[5] Gurland, J.and Plateau, J. Trans. ASM, 56 (1963),
442.
DYNAMIC SOLIDIFICATION OF
SAND-CAST ALUMINIUM ALLOYS
P. Appendino *, G. Crivellone**, C. Mus**, S. Spriano*
*Politecnico di Torino - Dipartimento di Scienza dei Materiali ed Ing. Chimica - Torino
** Teksid Aluminium S.p.A. Via Umberto II, 3/5 - Carmagnola
Abstract
The effect of low-frequency mechanical vibration, applied during the solidification process,
on the microstructure and mechanical performance of aluminium sand-casting was
investigated. Both green-sand moulds and chemically bounded sand moulds were used.
The vibration was applied to the moulds along the vertical axe. The investigated
acceleration range was included between 0.1g and 15g. The effect of different sections
and cooling rate was considered, as well as the influence of using different amplitude at
constant acceleration. The microstructure and porosity of the ingots was evaluated by
optical microscopy and image analysis. The mechanical properties were investigated by
tensile tests performed both on cylindrical and different section flat specimens.
Microstructure modification of dynamically solidified castings was achieved, consisting
on a refined or a completely non-dendritic microstructure, with a globular aspect and
quite rare dendrite fragments. The effective threshold acceleration in order to modify
the microstructure of the different section considered was assessed.The ingots presenting
modified microstructure revealed interesting mechanical performance, that means mainly
higher fracture strain.
Riassunto
INTRODUCTION
Cooled and uncooled probes or magnetic fields
can be used at this purpose, but they present
technological disadvantages 3 , 4 . A wide frequency
and amplitude range, obtained by these different
equipment, were randomly explored by different
authors, starting from low mechanical vibration
(1Hz-1 cm) up to high ultrasonic waves (30000
Hz-10µm) 1. Most of the data refer to graphite,
cast iron or steel moulds 1.
In this work vertical low-frequency vibration of
sand moulds was applied to A356 alloy castings,
according to the interest on automotive
application. Sand moulds, bound by thermosetting
polymer, present a sufficient mechanical strength
for this purpose.
Several data confirm that grain refinement of pure
metals by dynamic solidification requires strong
acceleration 1, 5 . In the case of pure aluminium this
means that an acceleration of 4×104 g must be
The aim of this work was to obtain a modification of the microstructure of
sand-cast aluminium alloys by mechanical vibration. The final goal consists
in an improvement of the mechanical performance of sand castings and at
this purpose a non-dendritic microstructure is preferable. Traditional
methods utilised in order to obtain this effect are rapid cooling rate and
addition of grain refining elements. Alternatively we employed dynamic
solidification condition, by applying a mechanical vibration to the sand moulds.
Because of the expensive equipment required and size limitation, this method
was not widely used up to now. Nowadays the new development of dynamic
system technology, with low investment, could permit new interesting largescale industrial application.
The effect of dynamic condition during the solidification of castings was
investigated in literature from a long time (1868), but it was not completely
clarified due to the wide number of different equipment and materials
employed and the often non-systematic investigations and data reports
performed 1 , 2 . A dynamic state can be applied both by mould vibration
(along a vertical, horizontal or rotational axe) and just by casting vibration.
Vengono qui presentati i risultati della sperimentazione
condotta su getti in lega d’alluminio colati in forme in sabbia.
Sono stati studiati gli effetti sulla microstruttura e sulle
caratteristiche tensili di vibrazioni meccaniche a bassa
frequenza applicate durante la fase di solidificazione. Per le
forme si è usato sia terra verde che sabbia legata
chimicamente. La sollecitazione vibrazionale è stata
applicata lungo l’asse verticale della motta. Il range di
accelerazione investigato varia da 0.1g a 15g. E’ stato
valutato l’effetto della velocità di raffreddamento
considerando sezioni di diverso spessore. Si è anche valutata
l’influenza di differenti ampiezze a parità d’accelerazione.
L’esame microstrutturale e le misure di porosità sono stati
condotti con microscopio ottico ad analisi d’immagine.
Le caratteristiche meccaniche sono state misurate con
prove di trazione quasi statica sia sulle provette cilindriche
che su quelle piatte di differente spessore.
La solidificazione dinamica dei getti ha determinato
modifiche microstrutturali in generale affinando la struttura
dendritica fino, in alcuni casi, a mostrare una microstruttura
con aspetto globulare e rari frammenti detritici.
I getti con microstruttura modificata mostrano
caratteristiche meccaniche interessanti soprattutto
nell’elevata capacità di deformazione a rottura.
applied. On the other hand, suppression of
columnar growth on alloys presenting dendritic
solidification can be obtained by applying low or
medium acceleration (0.1 g - 2×103 g). Grain
refinement and non-dendritic microstructure by
dynamic solidification was explained in literature
considering a higher nucleation frequency by
fragmentation of primary dendrite arms, but the
discussion on the cause of dendrite fragmentation
is still open 7. Some mechanical effects, like bending
stress and fracture of primary arms, turbulent
liquid flow around dendrite arms and showering
mechanism due to impingement of detached
dendrites were supposed 1. Other authors
reported that the main cause was the damage of
dendrites due to gas bubbles and cavitation effects 1. This is also the main
cause of refinement in the case of pure metals, where a planar front
solidification is involved and quite high accelerations are necessary. Finally
thermodynamic aspects, like remelting at the neck of dendrites, due to
local recalescence produced by stirring, and reduction of solidification time
were also considered 6.
Discordant data were reported about the secondary effects of dynamic
solidification. Some authors underlined a decrease of pipes and shrinkage
porosity, due to a decrease of the temperature gradient and solidification
time 7. This effect can be related to a higher liquid flow around dendrite
arms and to small dendrites carried around in the liquid. At this regard
smaller feeders could be employed in the case of dynamically solidified
ingots. Furthermore the dynamic state of the liquid causes also a degassing
effect. The ingot soundness can take advantage of this issue or it can be a
source of higher gas porosity depending of feed-head design.
MATERIALS AND METHODS
The vibration equipment employed consisted of a 900kg heavy machine
(2m/s maximum speed, 60g maximum acceleration and 3000Hz maximum
frequency). The vibration was applied to the moulds along the longitudinal
axe, as a sinusoidal wave. The frequency range explored started from 5 up
to 350 Hz and the amplitude range was included between 1×10-5 and 1×103
m. The peak acceleration can be obtain by using equation (1):
(1)
a = 4 π2 f2 x
a = peak acceleration
f = frequency
x = peak to peak amplitude
It was explored the peak acceleration range included between 0 and 15g.
The vibration was applied to each ingot for 5 minutes.The melt temperature
was checked before each casting and it was of minimum 717 and maximum
733 °C.
The ingot shape can be observed in figure 1. It consisted of a conic shape
central body (maximum and minimum diameter 60-40mm) and three
rectangular ribs of different section (mm x mm): 5x20 (thin), 7x20 (medium)
and 9x20 (thick). This ingot shape was selected in order to investigate the
effect of the different vibration parameters on samples of different section
and cooling rate. One ingot, representative of each vibration condition tested,
was sectioned along the longitudinal axe and twelve metallographic
specimens were obtained. In fact for each ingot the microstructure of the
top, centre and bottom of the central conic body and of the three rectangular
ribs was observed.
The porosity of the ingots was evaluated by image analysis, performed on
low magnification microphotographs, as ratio between pore and massive area.
Tension tests were performed on three ingots for each vibration condition
employed. From each ingot one cylindrical specimen was obtained from
the conic central body and three rectangular samples (3-5-7mm thick) were
cut from the lateral ribs. All the tensile samples were prepared according
to ASTM normative B557-94 (“Standard test methods of Tension Testing
Wrought and cast Al and Mg alloy products”).The tests were performed by
controlling the cross slide speed with a nominal strain rate of 10-3 s-1. All
the ingots were thermally treated, before cutting the tensile specimens,
according to the T6 condition (solution at 540°C for 4 hours; quench in
cold water within 25 seconds; ageing at 160°C for 4 hours).
The A356.0 alloy was used and its average chemical
composition was: Si 6.2%wt, Mg 0.3%wt, Fe max
0.2%, Ti max 0.2%, Al balance. The alloy was
modified by Na and the modifier was added, to
the melt bath, hour after hour in order to avoid
that the modification effect was lost.
Both green-sand moulds and chemically bounded
sand moulds (cold box system) were used. Greensand means a simple mixture of sand-silica, clay
(calcium bentonite) and water: pressure was
applied to the sand in order to compact it firmly
against the face of the pattern. The cold box
systems used worked with phenolic resin and
isocyanate component, cured by a vaporised amine
catalyst.
Fig. 1: The photograph reproduces the shape of the
prepared ingots. They were formed by a central conic
body and three lateral ribs of different sections
J.Campbell 1 reported an extensive review about the effects of vibration on
solidifying metals, containing useful amplitude-frequency maps. By using these
figures the theoretically serviceable conditions, in order to obtain
microstructure modification, can be approximately found. The lower limit
corresponds to the lowest stress sufficient to bend roots, while the upper
limit is the threshold beyond which surface standing wave patterns, with
melt ejection, can occur. The amplitude-frequency range included between
these two limits can be effective or not, depending on material mould,
cooling rate, material and dimension of the ingot, type of the equipment
employed.
In this work we tested the conditions reported in figure 2.The range included
between 0.1-2.5g was explored in the case of green-sand castings, because
of the limited strength of these moulds, while accelerations up to 15g were
tested in the case of chemically bounded moulds. In fact this type of mould
showed good firmness up to 18g. In the case of 3 and 5g we compared the
effects of the same acceleration obtained by using two different frequencyamplitude pairs, one close to the lower and one to the upper theoretic
limit.
Considering the microstructure of the α-aluminium phase, all the castings
solidified by using acceleration lower than 2.5g presented a dendritic
microstructure quite close to the static reference, as well as all the casting
solidified by using low amplitude vibrations. In fact no significant
microstructure modification was observed on the castings solidified at 3 or
Fig.3: Microstructure of the medium thickness rib of a static ingot. Typical dendritic
aspect. - 25x
5g, obtained by using vibration conditions close to the lower theoretic limit
(Figure 2). On the contrary the castings solidified at 5g or higher acceleration,
obtained by using amplitudes close to the upper theoretic limit (Figure 2),
showed a non-dendritic microstructure both in the central conic body and
in the lateral ribs. In figures 3, 4 and 5 the microstructure of the medium
thick ribs was reported as an example. All the microphotographs referred
to the central part of the rib. It could be observed that while the static
ingot (Figure 3) showed a typical dendritic microstructure with long primary
arms, the specimen solidified at 5g (Figure 5) presented a substantially nondendritic aspect. In the case of the ingots solidified at 3g a sort of intermediate
microstructure was observed in the lateral ribs, while the central conic
shaped had a non-dendritic aspect. In the case of the medium thickness rib
solidified at 3g (Figure 4), dendrite fragments were distributed in the matrix
1000
5
3
15
7
100
2.5
5
3
Hz
RESULTS AND DISCUSSION
0.8
10
0.1
1E-5
1E-4
m
1E-3
experimental conditions tested [g]
lowest stress sufficient to bend roots
highest limit in order to avoid liquid ejection
Fig. 2: Vibration condition tested. The two lines
represent the lower and upper theoretic limits
with a higher frequency than in the specimens
submitted to higher accelerations during the
solidification. Furthermore the microstructure of
this sample had a refined aspect respect to the
static one, with a SDAS of about 40µm respect to
the average value of 28µm in the static specimen.
This gradient of microstructure modification
versus the casting thickness can be explained by
considering that thicker is the section of the ingots
and slower is the cooling rate, as consequence
longer is the time of vibration during the
solidification. So it can be expected that the same
acceleration will be more effective on thicker and
slow cooled section.The metallographic specimens
related to the ingots solidified at very high
acceleration showed a completely non-dendritic
microstructure, with a globular aspect and quite
rare dendrite fragments, as in the case of figure 6
(central conic body solidified at 15 g). In any case
the microstructure was not significantly refined
by using very high acceleration.
So as first it was assessed that the microstructure
modification can be obtained, in the investigated
cases, by using 3g or higher acceleration during
the solidification process, that means that
polymeric bounded sand moulds have to be
employed. In the second place the use of an
acceleration just higher than the assessed
threshold and obtained in condition close to the
upper theoretic limit, seams to be effective in order
to refine the microstructure in thinnest sections
and to obtain a non-dendritic microstructure in
the thickest ones. Slightly higher acceleration
allows to completely modifying the ingot
microstructure, while on the contrary the use of
quite high accelerations does not present any
significant advantage.
Considering the Si phase morphology it could be
noted that the eutectic Si presented a wellmodified aspect in all the samples. Furthermore a
slight coarser cell dimension was registered in the
specimens related to the solidification at 3g or
higher acceleration (Figure 7-8). The effect of
dynamic condition of solidification on the
morphology of the eutectic Si, in modified and unmodified alloys, is a controversial argument in
literature 1, 1 . In any case data in agreement with
our observations are reported in several papers
2, 3, 4, 5
. In these cases authors supposed that the
vibration disturbs the envelopment and poisoning
layer, that is reach in the modifier element, on the
silicon growth front. As a consequence, the silicon
diffusion to the growing silicon site is facilitated
and hence a coarser structure develops.This effect
can be avoided by short time of vibration. In any
case it seams to have a moderate effect in the
cases investigated in this research.
The porosity of the static and dynamically solidified
ingots was evaluated by quantitative image analysis
on the metallographic specimens. It was observed
that an increment in the acceleration, applied
during the solidification process, involved an
increment in the average porosity both in the
central conic body and in the lateral ribs.The effect
was more evident increasing the thickness of the
section considered. In figure 9 the data obtained
on the thinnest ribs were reported as an example.
The almost linear trend of the porosity respect
to the acceleration applied could be observed also
by comparing the porosity on the top, centre and
bottom area of the different ribs. In the most of
the cases reported on figure 9 the top showed a
higher porosity than the bottom. This trend was
not reproduced in thicker sections, where in some
cases the bottom presented a porosity peak. The
position of the pores along the longitudinal axes
can be an indication of the occurrence of a
degassing effect due to the mechanical vibration.
It seams that in thinner section it was easier to
obtain sound ingots and that the most of the pores
were confined on the top. This means that
opportune feeders could probably avoid them. By
comparing the same acceleration obtained at low
and high amplitude, it could be remarked that in
the thinner sections the samples obtained in
Fig. 4: Microstructure of the medium thickness rib of an ingot obtained by applying 3g
acceleration (high amplitude). Refined aspect - 25x
Fig. 5: Microstructure of a medium thickness rib of an ingot obtained by applying 5g
acceleration (high amplitude). Non-dendritic aspect - 25x
Fig. 6: Microstructure of the conic shape central body of an ingot solidified by applying
15 g. Non-dendritic microstructure - 32x
vibration condition close to the upper theoretic limit (high amplitude)
presented a lower porosity. This is less evident as thicker was the section
considered, in any case the highest level of average porosity was registered
in the case of the central conic body of the ingot solidified at 5g at low
amplitude (2.7%).These data are in agreement with some published results
6
where an enlargement and coagulation mechanism of the gas bubbles, due
to vibrational treatment, and the preferential location of the pores on the
top of the ingots was also reported. So it must be considered that when
un-degasified melts are dynamically solidified, feeder dimension, location
Fig. 7: The microphotograph reproduces the eutectic Si morphology in the conic
central body of a static ingot. The alloy was modified by Na - 500x
Table i. Mechanical data obtained on flat specimens
Tensile
sample
thickness
[mm]
Acceleration
applied
Y
UTS
E%
to 217 MPa and 230 MPa. In table 1 the results
obtained on the flat samples were summarised.
The average values obtained on three samples
were reported, with the indication of the standard
deviation (SD). In figure 10 two stress-strain curves
related to static and dynamically solidified
specimens (5mm thick) are reported as an
example. It can be underlined that 3mm thick
specimens solidified at 3 or 7g (high amplitude)
showed an increment of the 66% in the elongation
(E) respect to the static reference, with any
significant decreasing in the Y value and also an
increment in the UTS value. Also considering the
5 mm thick specimens a better behaviour of the
specimens obtained at 3g (high amplitude) respect
to the static reference can be observed and 52%
of the increment in the elongation (E) was reached
average
average
average
value
value
value
[MPa] St.Dev [MPa] St.Dev [%] St.Dev
3
5
7
a=0
a=3g
(low amplitude)
a=5g
(low amplitude)
a=3g
(high amplitude
a=5g
(high amplitude)
a=7g
(high amplitude)
a=0
a=3g
(low amplitude)
a=5g
(low amplitude)
a=3g
(high amplitude)
a=5g
(high amplitude)
a=7g
(high amplitude)
a=0
a=3g
(low amplitude)
a=5g
(low amplitude)
a=3g
(high amplitude)
a=5g
(high amplitude)
a=7g
(high amplitude)
252
0
284
6
2.4
0.6
246
4
281
8
2.6
1.0
228
7
267
4
2.9
0.8
248
4
300
8
4.0
0.8
250
1
282
7
1.8
0.7
243
239
4
7
293
274
7
1
4.0
2.3
0.4
0
237
6
272
8
2.6
0.5
227
5
265
7
2.7
0.6
238
10
279
13
3.5
1.0
242
2
272
0
1.8
0
233
226
3
2
269
260
3
1
2.3
2.6
0.2
0.4
232
2
262
7
2.1
0.6
215
6
250
8
2.4
0.3
223
3
264
5
2.7
0.5
224
3
253
1
1.7
0
220
5
253
2
1.8
0.1
and design have to be strictly controlled in order to confine pores in uncritical
zones.Alternatively by pouring well-degasified melts, dynamical solidification
could be effective in order to minimise shrinkage 6, 6 .
Tensile tests were performed both on cylindrical specimens cut from the
central conic body of the ingots and on flat samples sectioned from the
lateral ribs. The tensile tests executed on the cylindrical specimens didn’t
show any remarkable difference between static and dynamically solidified
ingots.The fracture occurred in any case at about 0.7% strain; the yield (Y)
and ultimate tensile strength (UTS) average values were respectively close
Fig. 8: The microphotograph reproduces the eutectic Si
morphology in the conic central body of an ingot
solidified by applying 5g (high amplitude). The alloy was
modified by Na - 500x
(Figure 10). These data could be related to the
microstructure modification observed on these
types of samples. The correlation between SDAS
of cast structures and the fracture strain is
sufficiently strong and confirmed from many
examples found in the literature 7 , 8 and the
presence of a non-dendritic microstructure seams
to produce a similar effect.
This effect is of special significance considering that
the fracture strain is the referring specification for
the mechanical designer since the safety
component, realised by using cast aluminium alloy,
has to be verified in the crash resistance. The
difference between static and dynamically solidified
specimens is less evident in the case of thicker
tension specimens (flat 7mm thick and cylindrical
samples).This can be related to the higher porosity
observed on these specimens respect to the static
ones. In any case it could be supposed that the
dynamically solidified specimens presented a higher
deformation capability, considering that in spite of
a higher porosity they showed a mechanical
behaviour closed to the static reference.
Unfortunately porosity enhancement probably hid
the effect of microstructure on the mechanical
performance also in the case of 5g (high amplitude)
thin tension samples.According to metallographic
observation the specimens solidified in condition
closed to the lower theoretic limit (3 and 5g at
low amplitude) didn’t show any remarkable
difference respect to the static reference, both in
the case of thin and thick sections.
So, considering that the main objective of this
research was to verify the effect of dynamic
solidification on the microstructure of sandcastings, it can be concluded that once assessed,
case by case, the threshold acceleration to be
overcome, dynamic solidification allows to refine
or completely modify the microstructure. The
feeder design must be opportunely modify in order
to avoid gas porosity. Furthermore the preliminary
mechanical data obtained evidenced an
improvement of the fracture strain of dynamically
solidified sand castings.
Fig. 9: Porosity data obtained on the thin lateral ribs of static and dynamically solidified
ingots. L= low amplitude; H= high amplitude
Fig. 10: Stress-strain curves related to tensile tests performed on static and
dynamically solidified (3g high amplitude) medium thickness ribs
CONCLUSIONS
Microstructure modification of aluminium sand
castings was achieved by low-frequency mechanical
vibrations. In fact, while in the static condition the
tested castings showed a typical dendritic
microstructure with an average SDAS of 40µm
(medium thickness ribs), on the contrar y
dynamically solidified samples, prepared at low
acceleration, showed a dendritic microstructure
with a refined aspect (average SDAS of 28µm).The
use of higher accelerations allowed obtaining a
completely non-dendritic microstructure, with a
globular aspect and quite rare dendrite fragments.
The threshold acceleration to be employed
depends on casting section and cooling rate, it was
closed to 3g in the case considered.
The ingots presenting modified microstructure
revealed interesting mechanical performance, that
means mainly higher fracture strain.
Thanks
REFERENCES
1
2
3
4
5
6
7
8
9
10
11
Campbell, J. Effects of vibration during solidification. Int.Met.Rev. 2 (1981), 71.
Gittus,J.H.The inoculation of solidifying iron and steel castings by means of vibration
J. of the Iron and Steel Inst. 6 (1959), 118
Fang, Q.T. and Bruno, M.J. Casting of high purity aluminum alloys using mechanical
stirring for grain refinement Light Met. 2 (1991), 851
Southgate, P.D.Action of vibration on solidifying Aluminum alloys J.Met. 4 (1957),
514
Southin,R.T.The influence of low-frequency vibration on the nucleation of solidifying
metals J. Inst.Met. 94 (1966), 401
Fisher,T.P.Effects of vibrational energy on the solidification of aluminium alloys. Br.
Foudryman 66 (1973), 71
Kocatepe, K and Burdett, C.F. Effect of low frequency vibration on macro and
microstructures of LM6 alloys. J.Mat.Sci. 35 (2000), 3327
Burbure, R.R., Hareesha, I. and Murthy, S.S. Influence of low frequency vibrations
on aluminium eutectics. Br. Foudryman 72 (1979), 34
Pillai, N.R. Effect of low frequency mechanical vibration on structure of modified
aluminum-silicon eutectic. Metall.Trans.3 (1972), 1313
Abd-El-Azim,A.N. Proc. 7th. Inter. Light Metals Congres (Vienna, 1981), p.118
Pandel,U., Sharma,A. and Rajan, T.V. Effect of vibrations on some cast Al-Si-Cu
alloys Proc. Indo-USWorkshops (Hyderabad, India)TransTech Publications , vol 2,
p.769
Flemings, M. Solidification Processing, McGraw-Hill, Inc 1974
Campbell, J.“Castings”, Butterworth Ltd, 1995.
The authors thank Mr. Ezio Merlo and Mr. Stefano
Plano from Centro Ricerche Fiat for their 12
cooperation during the trials and results evaluation. 13
FATIGUE PROPERTIES OF
A CAST ALUMINIUM ALLOY FOR
RIMS OF CAR WHEELS
C. Bosi, G.L. Garagnani, R. Tovo
Dipartimento di Ingegneria, Università degli Studi di Ferrara (Italy)
Abstract
Thanks to the potential weight saving, aluminium castings are good candidates for
automotive, electronic, aeronautic, sport equipments and other high performance
products. At present, one of the main limits to a wide use of aluminium alloys for these
applications is a lack of complete understanding of their fatigue behaviour and of the
relationships to microstructural features, particularly as far as casting alloys are
concerned.
In this paper, the rotating bending fatigue behaviour of a cast Al-10Si-0.6Cu alloy has
been investigated. The specimens for both the fatigue tests and the microstructural
analyses have been drawn directly from the rims of car wheels. The wheel design
influences the microstructure of the alloy, conditioning the cooling rate during
solidification and, of consequence, may have important effects on the wheel fatigue
performances.
Measurements of the sizes of the microstructural constituents, such as secondary
dendrite arm spacing and of porosity, have been also carried out by means of optical
microscopy, supported by an image analysis software.
Rotating bending fatigue tests have been performed on specimens with different types
of notches. In this way, after the fatigue tests, it was possible to study the effect of the
wheel design and so, of stress concentrations, on the fatigue life.
Riassunto
Le leghe di alluminio da fonderia sono ampiamente
utilizzate per molteplici applicazioni nell’industria
automobilistica ed aeronautica, in articoli sportivi,
nell’elettronica ed in impieghi dove sono richiesti
materiali ad elevate prestazioni. Uno dei fattori che
impediscono un più vasto uso dell’alluminio in questi
settori industriali è la limitata disponibilità di dati sul
comportamento a fatica ed in particolare sulle correlazioni
tra questo tipo di sollecitazione e le caratteristiche
microstrutturali e geometriche del componente.
In questo lavoro è stato studiato il comportamento a
fatica a flessione rotante di una lega da fonderia Al-10Si0.6Cu, impiegata per la produzione di cerchioni per
autovetture. I provini per le prove meccaniche e per le
analisi metallografiche sono stati ricavati direttamente dai
componenti. È stato dimostrato che la geometria ed il
disegno dei cerchioni, ottenuti per fusione in conchiglia,
hanno una forte influenza sulla microstruttura della lega
nelle varie parti del pezzo, in quanto condizionano la
velocità di solidificazione e di raffreddamento del materiale
e di conseguenza hanno significativi effetti sul
comportamento a fatica in esercizio.
Sono state eseguite osservazioni al microscopio ottico
ed analisi di immagine con misure della spaziatura
dendritica secondaria (SDAS) e delle dimensioni delle
microporosità residue.
Le prove di fatica a flessione rotante sono state condotte su
provini con differenti tipologie di intaglio, eseguendo
successivamente le osservazioni al SEM delle superfici di frattura.
In tal modo è stato possibile studiare l’effetto localizzato dei
difetti, della concentrazione localizzata delle tensioni e della
geometria dei cerchioni sulla resistenza a fatica.
INTRODUCTION
correlating them with microstructure and, as a
consequence, with the cast processing
parameters, since the 1940s [1].
Cast pores are preferential crack initiation sites
in these materials and, so, they have been
considered as the main parameter to study
because of their influence on the mechanical
properties. Secondary dendrite arm spacing
(SDAS), inclusions and grain size are also
considered to be important microstructural
factors to understand the mechanical behaviour
of cast alloys.
Recently, many studies have been carried out on
the influence of microstructure and microporosity
on the mechanical properties of aluminium cast
alloys. In some cases, the influence of casting pores
and of secondary dendrite arm spacing on the
fatigue crack initiation and propagation in cast
The application of aluminium alloy castings in many mechanical components,
especially for cars and rail vehicles, has gradually increased in the last years,
thanks to the great potential of these materials as replacements for ferrous
alloys.
In particular, for those applications in which the necessity of high mechanical
properties is combined with the need of a substantial weight saving,
aluminium castings seem to be extremely interesting solutions.
Moreover, the opportunity of producing cast components in a finished or
semi-finished shape permits a high reduction of the production costs.
The lower mechanical properties and reliability of the aluminium cast
alloys can be principally caused by the presence of defects and
inhomogeneities, which could be preferential fatigue initiation sites.
Also, a lack of complete understanding of the fatigue behaviour of aluminium
cast alloys, does not allow a full exploitation of the potential weight and
cost savings distinctive of these materials.
The mechanical properties of aluminium castings have been studied
aluminium alloys has been evaluated [1-6].
It is well known that, controlling the cooling rate
during solidification of cast structures, it is possible
to control the microstructural constituent sizes,
in particular secondary dendrite arm spacing.
Flemings [7] found that the SDAS of cast
structures usually has a stronger influence on
mechanical properties than the grain size has.
Increasing the cooling rate during solidification
SDAS decreases and, generally, mechanical
properties increase.
B. Zhang et al. [1], found that fatigue cracks
initiated from porosity in the material solidified
at slow cooling rates, while, as cooling rate
increased, the fatigue cracks initiated from near-
MATERIAL AND
EXPERIMENTAL PROCEDURE
The material investigated was a Al-10Si-0.6Cu cast
aluminium alloy. Samples have been directly drawn
from the rims of two cast wheels (named “A”
and “B” respectively); the two shapes were
different, especially concerning the rim section:
the minimum thickness changed, in fact, from
about 25 mm for rim of wheel A to 15 mm for
rim of B (see Fig. 1, a-b). The wheels had been
cast in a partially cooled metallic shell and were
used in the as-cast condition.
The mechanical properties of the alloy, evaluated
by the wheels supplier in previous tests, have been
reported in Tab. I; in Tab. II the chemical
composition has been summarised.
Microstructural characterisation of the material
has been carried out by standard metallographic
grinding and polishing; measurements of secondary
dendrite arm spacing have been also carried out
on samples cut from different parts of the two
wheels. In order to estimate the maximum
occurring defect size, the pore size has been
studied by means of a software for image analysis;
the results have been then elaborated.
Rotating bending fatigue tests, at a frequency of
200 Hz, have been then performed. Specimens
with three different notches and un-notched have
been tested, in order to evaluate the material
behaviour in presence of stress concentration
effects comparable to those in the fillets between
two rims.
Table II. Chemical composition
of the aluminium cast alloy
Chemical element
Si
Cu
Mn
Wt, %
10.7
0.6
0.35
surface eutectic microconstituents.
Porosity has been observed to affect fatigue life of cast alloys, in particular
at an high number of cycles; eutectic microconstituents, instead, affect
fatigue life particularly at a low number of cycles (high stresses) [4].
This research has been carried out in order to obtain a better
understanding of the relationship between cast design, which develops
continuously, microstructural characteristics and mechanical properties
of cast aluminium alloys. In particular, the fatigue behaviour of a cast
aluminium-silicon alloy has been investigated. As the material came from
the rims of two car wheels with different shapes, the effect of the rims
dimensions on microstructure has been studied. Rotating bending fatigue
tests have been performed on specimens un-notched and with different
notches, in order to evaluate also the stress concentration effect of
different notch tip radii. The fatigue strength reduction factors have been
investigated according to notch theory as presented in [8].
=
>
Fig. 1: Different design of the two cast aluminium wheels: a) wheel A); b) wheel B
Table I. Mechanical properties
of the cast alloy
Value
Ultimate
Strength
(MPa)
Yield
Strength
(MPa)
Elongation
(%)
Youngs
Modulus
(GPa)
220
160
15.0
72
The stress concentration factors calculated for these three different geometries
(named, respectively, R1, R2, R3) have been summarised in Tab. III [9].
Finally, in order to understand the fatigue nucleation and propagation
mechanisms, fractographic analyses have been performed on selected
fatigue fractured specimens.
Table III. Stress concentration factors of the
three different specimen geometries
Specimen
Notch tip radius (mm)
Kt
R1
R2
R3
10
4
1.5
1.11
1.26
1.68
RESULT S
Microstructure
Representative optical micrographs have been gathered in Fig. 2. In Tab. IV,
the values of secondary dendrite arm spacing (SDAS), measured for the
two different rims, and different parts of wheel “A”, have been reported
(medium values obtained from twenty optical microscopy measurements).
>
100 µ m
=
100 µm
Table IV. Mean values of
secondary dendrite arm
spacing measured in different
parts of the two wheels
SDAS (µm)
Wheel A
Hub
Rim
Wheel B
Rim
74.8
46.5
36.0
The SDAS measured for the rim of wheel B is
clearly lower than that of rim of wheel A, due to
rim B lower thickness, which induces higher
cooling rates during solidification.
In Fig. 3, has been reported a diagram of the
measured number of pores observed in the
material drawn from the two different rims. It is
important to note that porosity had the typical
shape of solidification shrinkages. The rims of
wheel A clearly present a higher porosity level.
Fig. 2: Optical micrographs of the dendritic structure of the cast Al-Si alloy: a) rim
drawn from wheel A; b) rim from wheel B
Rotating bending fatigue tests
4
Measured number of pores (mm-2 )
Rim A
Rim B
Rotating bending fatigue experiments (R = -1)
have been conducted at 200 Hz; for the rims of
wheel A un-notched specimens have been used.
The specimen geometries used for the fatigue
tests have been shown in Fig. 4.
3
2
1
0
100
1000
Pore area (µm2)
10000
100000
Fig. 3: Measured number of pores (per square mm) as a function of pore area in the
material drawn from the rims of the wheels A and B)
=
Fig. 4: Specimen geometries used in the fatigue tests (all
dimensions in mm); a) un-notched specimen, b) notched
specimens, respectively: R1, R2, R3
>
R
R1: R = 10 mm
R2: R = 4 mm
R3: R = 1.5 mm
The results of this first serie of tests have been
elaborated in the Wöhler curve of Fig. 5.
Another serie of tests have been carried out on
the notched specimens drawn from wheel B (see
Fig. 4 -b). The three Wöhler curves, drawn by
these further tests, have been reported in Fig. 6.
As presented in [10], in order to estimate the
reduction of fatigue resistance induced by the
presence of a notch, the following equation can
be used:
kt
kf=
Amp [MPa]
Wheel “A”
(1)
(1+ br )
N. cycles
Hence the fatigue strength of notched
components (as a function of nominal stress) turns
out to be:
σ
(notched) =
A, 2 106
(
σA, 2 10
Amp [MPa]
(un-notched)
kt
6
)
(1+ br )
σpeak
(1+ br )
Wheel “B”
(2)
Where r is the notch tip radius and b is a
coefficient characteristic of the material.
A best fitting of the experimental data permitted
us to find optimal values of the fatigue limit of
the material (un-notched specimens, N = 2.106)
and of b using the eq. (2).The results of this analysis
have been reported in Tables V-VI.
According to these data it is possible to evaluate
the “notch sensitivity” of this material, i. e. the
ratio kf/kt as a function of the notch tip radius.
In Fig. 7, the trend of the kf/kt as a function of the
notch tip radius has been plotted.
Unfortunately, in geometrically complex
components (as the car wheels), the nominal
stress values cannot be defined, and only local
stresses are available from numerical or
experimental investigations. As a consequence, in
order to take into account the reduced fatigue
notch sensitivity, the local stresses should be
reduced by a function of notch tip radius and
material parameter b. For a proper fatigue
strength prediction it is possible to compute the
“effective value” (σeff) of stress at notch. σeff is
function of (and lower than) peak stress value,
according to eq. (3):
σeff =
Fig. 5: Stress amplitude versus fatigue life for wheel “A”; un-notched specimen
(3)
In Fig. 8 the Wöhler curves as a function of the
effective stress value for the three series of
specimen have been reported.
N. cycles
Fig. 6: Stress amplitude versus fatigue life for the three notched specimens drawn
from wheel B
Table V. Experimental and estimated fatigue
strength of notched specimen at 2 . 106 cycles
r
(mm)
Kt
σA, 2·106 (MPa)
(experimental
σA, 2·106
(MPa)
(eq. 2)
A
10
1.11
76.1
76.6
B
4
1.26
72.7
71.8
C
1.5
1.68
61.5
61.9
Specimen
Table VI. Material parameters for fatigue
notch behaviour prediction
σA, 2·106 (MPa)
(eq. 2)
b (mm)
81.25
0.96
kf/kt
1
0,95
0,9
0,85
0,8
0,75
0,7
0,65
0,6
0,55
0,5
0
5
10
15
20
25
30
Notch tip radius
35
40
45
50
Fig. 7: Evaluation of kf/kt as a function of the notch tip radius
Amp. (eff. stress) [MPa]
As can be clearly seen from Fig. 8, the Wöhler
curves tend to superimpose one upon another,
forming a unique curve, when σeff is concerned,
(while it is not so when peak values are
considered); this curve represents the fatigue
resistance of the material correspondent to the
wheel B. It follows that eq. 3 (instead of eq. 2)
can be used when peak local values are available
(instead of nominal stress values).
The fatigue resistance of the material drawn from
wheel B is clearly higher than that of the material
from wheel A, in particular at a high number of
cycles.
Fractography
A SEM micrograph showing a typical fracture
surface generated during the fatigue tests, has
been showed in Fig. 9-a. These fracture
morphology has been frequently observed both
from specimens from wheel A and from wheel B.
Crack initiation has been usually observed in
correspondence of surfacial defects as porosities
and scratches due to specimens machining.
SEM fractographs at different steps of the crack
propagation have been reported in Fig. 9 – (b,c,d).
As can be seen in Fig. 9-c another typical crack
initiation site has been observed in near-surface
eutectic microconstituents.
Wheel “B”
N. cycles
Fig. 8: Wöhler curve as a function of the effective stress value for the material drawn
from wheel B
=
>
?
@
Fig. 9: SEM topographies of the fatigue fracture surfaces; a) a typical fracture surface;
b)-c) fatigue nucleation sites: b) scratch caused by machining, c) near surface eutectic
microconstituents; d) region in the first stage of growth, with typical fatigue
striations
DISCUSSION
The experimental results confirm obviously that
the fatigue resistance decreases as the notch tip
radius increases. Anyway, the lower fatigue
resistance observed in the material drawn from
wheel B could not be determined simply by the
coefficient kt , because the peak stress value does
not always control the failure mechanisms. In
order to understand in which measure the fatigue
strength is reduced by stress concentration
effects, it has been necessary to study the fatigue
behaviour of specimens with three different
notch tip radii, comparable to those existent in
wheel B.
Worth noting is the high dispersion of the asobtained results, which is probably caused by
microstructural defects and inhomogeneities,
characteristic of cast alloys.
After the experimental tests it was possible to
evaluate both the fatigue resistance of the material
and the fatigue strength reduction as a function
of the notch tip radius and microstructure
parameters.
As a consequence, after that it has been obtained
with a FEM analysis the peak stress value, it is
possible to calculate also the effective stress value,
which controls the fatigue behaviour.
Regarding the fracture surface analysis, two
potential crack nucleation sites have been
observed: surfacial porosities and near-surface
eutectic microconstituents. Also scratches and
microscopic notches caused by the tool during
specimen machining should be considered.
From a comparison between Fig. 5 and Fig. 8 it is
possible to note that samples drawn from wheel
A show a lower fatigue resistance than the ones
drawn from wheel B, in particular at lower
stresses (high number of cycles); we had no
information on the fatigue behaviour of the
material from wheel A at high stresses (higher
than 100 MPa), due to a low bending stiffness
caused by the geometry of this first serie of
specimens (see Fig. 4 -a).
It’s important to note that the wheel A was characterised by a different
geometry, with thicker rims than wheel B; the consequent lower cooling
rate during solidification caused a slight coarser microstructure (larger
secondary dendrite arm spacing).
Secondary dendrite arm spacing (SDAS) is a parameter that characterises
the medium dimensions of dendrites in cast alloys; in literature it is possible
to find several data [1,4,5] regarding the effect of SDAS on mechanical
properties.
Dendritic structure controls the precipitates and inclusions morphology
in the cast alloys; higher cooling rates, reducing secondary dendrite arm
spacing, reduce also the formation of interdendritic shrinkages and, so, of
the consequent porosities. Also for this reason, as SDAS decrease,
mechanical properties increase. Obviously, the best conditions are insured
by extremely fine dendritic structure and absence of porosity.
Our results, obtained both from microstructural analyses and from fatigue
tests, substantially confirmed what it has been found in literature; the
lower fatigue properties of the material from wheel A have been due to
its different geometry that, causing lower cooling rates during solidification,
caused the formation of a coarser dendritic microstructure, which affects
mechanical properties of cast alloys.
CONCLUSIONS
2) Samples drawn from wheel A showed a lower fatigue resistance than
the ones drawn from wheel B, in particular at lower stresses (high
number of cycles).
3) The lower fatigue properties of the material from wheel A have been
related to the different geometry of this wheel that, causing lower
cooling rates during solidification, caused the formation of a coarser
dendritic microstructure (larger secondary dendrite arm spacing), which
mainly affects mechanical properties of cast alloys.
4) The cast alloy showed low notch sensitivity. Fatigue stress reduction
was lower than stress concentration factor, and, so, a notch theory has
been used.
5) This research has confirmed that the marketing necessity of a continuous
development of the cast design should be always accompanied by an
accurate analysis of the consequent changes in cooling rates and
solidification mechanisms, that strongly influence microstructural
parameters, and of geometrical influence on stress distribution, valuable
by the stress concentration effects.
The fatigue behaviour of a cast aluminium/silicon
alloy for car wheels has been investigated. In
particular, the relation between rim thickness,
material microstructure and fatigue properties
has been evaluated. Rotating bending fatigue tests
have been performed on specimens un-notched
and with three different notches, in order to
evaluate also the fatigue strength reduction caused
by different notch tip radii. The investigation
allowed to draw the following conclusions:
1) The aluminium cast alloy of wheel “A” showed
a coarser dendritic microstructure, with larger
secondary dendrite arm spacing. The porosity
level in rims of wheel “A” was also higher.
REFERENCES
1)
2)
3)
4)
5)
6)
S. BERETTA and P. CLERICI, Conf. Proc.“La Fatica nelle leghe di Alluminio”,
Padova (1997), AIM, Milano (1997).
7) M. C. FLEMINGS, Solidification Processing, McGraw-Hill, NewYork (1974),
p 341.
8) B.ATZORI, P. LAZZARIN, and R.TOVO, J. of Strain Analysis 34, (1999), p
347.
9) R. E. PETERSON, Stress Concentration Factors, John Wiley & Sons, New
York (1974).
10) P. LAZZARIN, R. TOVO, G. MENEGHETTI, Int. J. Fatigue 19, (1997), p
647.
B. ZHANG,W. CHEN and R. POIRIER, Fatigue
Fract. Eng. Mater Struct. 23, (2000), p 417.
S. E. STANZL-TSCHEGG, H. R. MAYER, A.
BESTE and S. KROLL, Int. J. Fatigue. 17, (1995),
p 1995.
S. BERETTA and Y. MURAKAMI, Fatigue &
Fracture of Eng. Mat. and Struct. 21, (1998), p
1049.
R. TOVO, F. BONOLLO and G. MUFFATTO,
Conf. Proc.“La Fatica nelle leghe di Alluminio”,
Padova (1997), AIM, Milano (1997).
M. ROSSO, G. SCAVINO, M.ALBERTAZZI and ACKNOWLEDGEMENTS
A. ZENADA, Conf. Proc. “La Fatica nelle leghe
The research has been carried out with the financial support of Fondmetal
di Alluminio”, Padova (1997), AIM, Milano
Technologies S.r.l. (Casumaro, FE).
(1997).
MECHANICAL AND MICROSTRUCTURAL
CHARACTERISATION OF AN ALUMINUM
FRICTION STIR-WELDED BUTT JOINT
M. Di Paola, A. Falchero - Innovazione Tecnologica Alenia Aeronautica, Torino, Italy
M. Cabibbo, E.Evangelista, E. Meccia
and S. Spigarelli - INFM/Department of Mechanics, University of Ancona, Ancona, Italy
Abstract
The microstructure and the mechanical properties of a 6056 aluminium alloy Friction
Stir-Welded (FSW) joint were investigated in the present study.The structure was analysed
using light, scanning and transmission electron microscopy.The change in microstructure
across the welded joint was found to correspond to significant variation in hardness.As
in most FSW joints, the structure was characterised by the presence of a region of
severely deformed grains in proximity of the weld nugget, i.e. of a region of fine
recrystallised grains. Tensile tests showed that the joint material exhibited a rupture
strength similar to the parent material, even though the former was significantly less
ductile. This difference resulted in a reduction in ductility of the welded sheets. A T6
treatment increased tensile strength, but further reduced joint ductility. Nevertheless,
the strength of the welded sheet was found to be very close (80-90%) to that of the
base alloy.
Riassunto
Il presente studio ha preso in esame la microstruttura e le
proprietà meccaniche di un giunto in lega di alluminio 6056
prodotto per Friction Stir Welding (FSW). La
microstruttura è stata analizzata tramite microscopia ottica
ed elettronica in scansione e trasmissione. Si è osservato
che la variazione della microstruttura trasversalmente alla
saldatura corrispondeva a significativi cambiamenti del
valore della microdurezza. La microstruttura, come del
resto usuale in giunti prodotti per FSW, era caratterizzata
dalla presenza, in prossimità della zona indicata come “weld
nugget”, cioè della zona di grani ricristallizzati, di una regione
in cui i grani erano severamente deformati. Le prove di
trazione hanno mostrato che il materiale del giunto
possedeva una resistenza a rottura abbastanza simile a
quelle della lega madre, anche se era significativamente
meno duttile. Un tale risultato si traduceva quindi in una
complessiva riduzione della duttilità dell’insieme delle
lamiere saldate. Un trattamento T6 aumentava la resistenza
meccanica, a spese di ulteriori riduzioni della duttilità.
Nondimeno, la resistenza delle lamiere saldate rimaneva
molto prossima (89-90%) a quella del materiale base.
INTRODUCTION
top to the bottom of the joint, a change that
corresponds to a variation in recrystallised grain size.
The present study aims at characterising a friction
stir-welded joint from the microstructural and
mechanical points of view.To do this, specimens in
as-welded, welded and aged conditions were
investigated using different testing methods.
Aluminum alloys are widely used to produce aerospace components with
high specific strength. On the other hand, when traditional welding processes
are applied to these alloys, they often entail disadvantage that have sometimes
discouraged the use of welded components. Friction stir welding (FSW) is
a recent method of joining materials patented in 1991 by TWI [1-13]; this
process was developed from the classic friction welding methods and has
the advantage of operating in solid state. In conventional friction welding
the plasticized material is constrained in two dimensions (vertical and
horizontal), while in FSW the layer of plasticized material is constrained in
three dimensions [14]. FSW assures the absence of porosity, distortion and
residual stresses, which are typical defects of the fusion welding processes,
and the possibility to operate in all positions without protective gas.Alloys
difficult to weld like the 2000 and 7000 series have successfully been welded
using FSW. The process uses a shouldered rotating tool with a profiled pin
that penetrates the clamped parts to be joined; the tool then starts to
move along the join line (Fig.1). The heat produced by friction softens the
alloy and the pin stirs the material of the joint until the sheets are joined.
The stirring of one material into the other is associated with a solid-state
flow, i.e. with a very high deformation, and involves recrystallisation of a
portion of the joint (weld nugget).These extremely fine recrystallised grains
are known to slide one over the other [8], leading to a superplastic flow
that accommodates the FSW process. FSW temperature decreases from the
EXPERIMENTAL DETAILS
The joined sheets measured 200x470x4 mm; the
material, a 6056-T4 alloy, had the following chemical
composition (wt.%): 07-1.3 Si, 0.3-1.0 Mn, 06-1.2
Mg, 0.25 Cr, Cu=0.5-1.1, Zn=0.1-0.7.(Ti+Zr)=0.2,
Fe=0.5, Al=bal.
Scanning Electron Microscopy (SEM) was used to
analyse the friction stir-welded surface, i.e. the
upper surface in Figure 1; upper and cross-sections
were grinded, polished, etched with Keller’s reagent
(20 sec), and examined by light microscopy (LM).
Vickers microhardness was measured along the
weld on the same surface.
Samples for Transmission Electron Microscopy
(TEM) were cut from the central region of the
weld to observe the microstructure of the severely
deformed zone and the weld nugget.Thin foils were
mechanically thinned to a mean thickness of 5060 µm and treated with an electropolishing double
jet with a solution of 25% HNO3 in methanol at
18 V and –35°C.TEM micrographs were taken with
a Philips CM200 microscope equipped with
double-tilting specimen holder. Tensile tests were
carried out on specimens of different dimensions
drawn from different locations (Fig. 2) to measure
the mechanical properties of base, as-welded and
T6-treated materials.The grey specimens in Figure
2 were subjected to ageing treatment, whereas
the white ones were tested as-welded. Small
samples were used to estimate the mechanical
properties of the joint and large ones to evaluate
the overall mechanical behaviour of the welded
sheets.
Post-weld heat treatment consisted in solutioning
at 530°C for 4 h and ageing. The base material
was first aged for different times at 160°C and
175°C to identify the peak ageing condition.
Fig. 1: Schematic representation of Friction Stir Welding process. The head of the
rotating pin is in black. The grey zone represents the weld
Results and discussions
Figure 3 shows a SEM image of the upper surface
of the joint, where tool and sheet were in contact.
The roughness of the surface after welding can
easily be distinguished (recent studies mention the
possibility of using radially located cutters on the
welding tool to reduce this effect [2]).
Figure 4 shows the macrostructure of the
transverse surface across the welding zone. The
cross-section has the typical aspect of a friction
stir-welded zone, whose most prominent feature
is the presence of a “weld nugget zone”
characterised by fine equiaxed recrystallised grains.
As mentioned above, the presence of this zone is
due to the recrystallisation occurred during the
welding. Figure 5 shows the microstructure of the
weld.The difference in grain shape and dimensions
between the weld nugget (Fig.5b) and the parent
material (Fig.5c) is quite evident: the mean value
of the grain size as measured by LM is 11 µm in
the weld nugget and 65 µm in the parent alloy.
Microstructure variations are also evident on the
upper surface, where fine grains along the welding
line are easily distinguished.
Microhardness tests along the welded joint (Fig.6)
showed the relationship between microstucture
and mechanical properties. For the cross-section,
the hardness values are closely connected with
the nature of the process experienced by the
Fig. 2: Identification of tensile test samples: small specimens had a gauge length
of 25 mm, and a 5x4 mm section; large samples had a 55 mm gauge length
and a 20x4 mm section
Fig. 3: SEM micrograph of the FS-welded zone
zones. In FSW aluminium joints four zones exist:
the recrystallised central region (Fig.5b)
surrounded by the thermomechanically affected
zone (TMAZ), a heat affected and deformed region
(Fig. 5a) and the heat affected zone (HAZ) on either
side of the joint [14]. FSW creates a softened
region around the weld center in precipitationhardened Al-alloys [15]; the analysis of hardness
distribution in the weld led indeed some authors
[16] to conclude that the hardness profile in
particle-hardened materials is mainly governed by
the distribution of fine precipitates. The lowest
hardness value in Fig.6 is found in an area of great
grain distortion, indicated as ‘’A’’ in Figure 4. A
peak in hardness is observed in the central part
of the weld-nugget; then, HV gradually decreases,
reaches a minimum value in the heat-affected zone,
and gradually increases up to the value typical of
the parent alloy.
Figure 7 shows the microstructure of the region
of transition between weld nugget and nonrecrystallised stirred material (Fig.7a) and the
typical microstructure of the weld nugget interior.
A dramatic difference in dislocation density
between recrystallised and non-recrystallised
grains is apparent. The grain interior is heavily
Fig. 4: Structure of the thermo-mechanical affected zone
decorated with intragranular Mg2Si precipitates, which are extremely
effective in reducing dislocation mobility, as clearly indicated by the high
fraction of dislocations pinned on these precipitates. The low dislocation
density in the recrystallised grains is well documented in Figure 7b.The fine
grain size in the weld nugget can easily be appreciated; in particular, TEM
analysis demonstrates that the average grain size obtained with LM
measurements is overestimated. The size and distribution of intragranular
precipitates do not change appreciably in the recrystallised region.
Fig. 5: Microstructure corresponding to the A (a) and B (b) zones in TMAZ. In (c) the structure of the parent material is shown
140
center of
the weld
HV
120
605 6
100
X= 0
80
500 m
60
-2
0
2
4
6
8
position [mm]
10
12
14
Fig. 6: a) Hardness profile along the weld (transverse section); in b) the microstructure and the x = 0 position are clearly shown
>
=
2
2
m
m
Fig. 7: Microstructure of the joint (TEM): a) transition between recrystallized
and non-recrystallized zones; the pinning effect of precipitates is apparent; b) typical
morphology of recrystallized grains; the fine grain size and the relatively high fraction
of precipitates can be easily appreciated
150
6056
140
130
HV
Ageing curves at 160°C and 175°C are shown in
Figure 8. Analysis of this figure allowed to identify
the conditions for T6 treatment (solution
treatment at 530°C for 4 h and artificial ageing at
175°C for 24 h). Figure 9 shows the typical
microstructure of the weld nugget after T6
treatment.The grain growth that took place during
solution treatment is evident in Figure 9a. Figure
9b permits to appreciate the distribution of Mg2Si
precipitates after ageing.
Figure 10 shows some of the stress-strain curves
obtained in this study. The parent material is very
ductile even in the T6 condition; by contrast, the
small samples machined from the joint are
characterised by a low ductility. The T6 condition
enhances this brittleness, with minor advantages
in terms of tensile strength.The strain to fracture
of the large transverse samples is thus lower than
that of the parent material, an obvious effect of
the relatively poor ductility of the joint.
Nevertheless, the mechanical response of the
welded sheets in terms of tensile strength and
yielding is 80-90% of that of the parent material.
Figure 11 summarised the averaged results of
tensile tests. The variation in strength illustrated
in Figure 11 are fully consistent with other
literature data on 6000 alloys [17].Also in the case
of a 7075 alloy, the as-welded strength and ductility
were found to be lower than those of the base
material, that was characterised by the presence
of precipitates of various sizes [18]; a post-weld
heat treatment further reduced ultimate strength
and ductility. This observation, and the fact that
the tensile samples fractured in the region of the
HAZ that experienced temperatures between 300
and 350°C during the welding process, led the
authors to conclude that also in this alloy the
mechanical response of FSW joint was controlled
by the size and distribution of precipitates.
The study of the fracture location showed that in
as-received specimens the failure began from the
transition zone of the cross-section (Fig.5a), a
result that again is fully consistent with the available
data [14] that indicate that fracture do not occur
in the weld nugget or in the joint centerline. By
contrast, in T6 samples sometimes the fracture
originated from the joint line, probably due to a
greater concentration of precipitates in that zone
after heat treatment, even if a recent study [19]
suggested that, in the case of a solution treated
and aged joint, the fracture should occur in the
region with the minimum average Taylor factor, i.e.
the fracture location depends on crystallographicorientation distribution and strain tensor of
imposed deformation.
120
110
1 60 C
1 75 C
100
90
80
0
1000
104
105
106
time [s]
Fig. 8: Ageing curves obtained on parent alloy after solution
treatment at 530°C for 4h
>
=
2 µm
2 µm
Fig. 9: Microstructure of weld-nugget after T6 treatment (TEM); a) effect of grain
growth during solution treatment; b) distribution of precipitated Mg2Si
350
300
x 11
4 x
5x
x
300
200
stress [MPa]
250
stress [MPa]
400
x
x
6056-T4
weld (sample 11)
weld (sample 8)
base (samp le 10)
150
100
200
x
10
605 6
T4 (samples 10-11)
T6 (samples 4-5)
100
50
0
0
0
5
10
15
20
25
0
5
strain [%]
10
15
strain [%]
20
25
Fig. 10: Stress-strain curves for weld and parent material
FSW was found to produce microstructural
variations across the weld, due to the
thermomechanical treatment involved in the
process. Such variations result in local changes in
mechanical properties. Nevertheless, the tensile
strength of the welded joint is close to 90%, and
its yield strength is almost equivalent to those of
the parent alloy, an extremely satisfactory result
that demonstrates the high quality of the process.
After a T6 heat treatment, tensile strength rises
to 95% of that of the parent alloy. The only
disadvantage with FSW welding is a reduction of
the material ductility, a disadvantage is enhanced
bt subsequent T6 treatment.
stress [MPa]
400
R
Rp02
6056
25
A
20
300
15
10
200
5
100
base
base T6
weld
weld T6
strain to fracture [%]
CONCLUSIONS
0
Fig. 11: Histogram showing the mechanical response (rupture strength, R, yield
strength, Rp02, and ductility, A) for the FSW material; samples 4 and 10 were used to
estimate the properties of base material, and samples 5,6,11 and 12 for the weld
REFERENCES
1] C.J.Dawes, Friction Stir welding, in “Training in Aluminium Application
Technologies – TALAT”, European Aluminum Association, in CD-ROM (1999),
lecture 4410.
2] O.T.Midling, E.J.Morley, A.Sandvik, “Friction stir Welding”, European patent
Application 959 078 88.
3] G. Liu , L.E.Murr, C.-S. Niou, J.C.McLure and F.R.Vega, Scrita mater., 37 (1997) 355.
4] K.V. Jata and S.L.Semiatin, Scripta mater., 43 (2000) 743.
5] T.J.Linenert, R.J.Grylls, J.E.Gould and H.L.Fraser, Deformation Microstructures
in friction Stir Welds on 6061-T651, in Proc. of the TMS Fall Meeting, The
Minerals, Metals and Materials Soc.,Warrendale, PA, USA, 1998, 159.
6] H. G. Salem, A. P. Reynolds, J. S. Lyons, Scripta Materialia 46 (2002) 337.
7] Y.S. Sato, H. Kokawa, K. Ikeda, M. Enomoto, S. Jogan andT. Hashimoto, Metallurgical
and Material Transactions A, 32A, 45 (2001) 109.
8] S. Benavides,Y. Li, L.E. Murr , D.Brown and J.C.McClure, Scripta Materialia, 41
(1999) 809.
9] Y. Li, L.E. Murr, D. Brown and J.C. McClure, Material Science and Engineering
A271 (1999)
10] R.A. Prado, L.E. Murr, D.J. Shindo, and K.F. Soto,
Scripta Materialia 45 (2001) 75.
11] P.S. Pao, S.J. Gill, C.R. Feng, and K.K. Sankaran, Scripta
Materialia 45 (2001) 605.
12] K.N. Krishnan, Material Science and Engineering
A327 (2002), 246.
13] Ying Li, L.E. Murr and J.C. McClure, Scripta
materialia 40 (1999) 1041.
14] R.Fenn, W.M.Thomas, Light Metal Age, 59 (2001)
28.
15] Y.S.Sato, H.Kokawa, M.Enomoto, S.Jogan,
T.Hashimoto, Metall.Mater.Trans., 30A (1999) 3125.
16] Y.S.Sato, S.H.C.Park, H.Kokawa, Metall.Mater.Trans.,
32A (2001) 3033.
17] S.Totori, Ligh Metals Age, 59 (2001) 18.
18] M.W.Mahoney, C.G.Rhodes, J.G.Flintoff,
R.A.Spurling and W.H.Bingel, Metall.Mater.Trans.,
29A (1998) 1955.
19] Y.S.Sato, H.Kokawa, Metall.Mater.Trans., 32A (2001)
3023.

application of a neural network for an improved control of the

Transcription

Similar documents

Lithuanian Radio Sport Federation Kaunas University of

How to Tie a Woggle

lnstallation - ConservationMart.com

Turkey - EOPOWER

Key - Scioly.org

Correlation of MicrostructEire vvith WIAZ Ernbrittlement

Hotondo February Newsletter

Lot 17 Altivole Drive, Edenbrook Estate, Norville

Intro Takin scann proba betw direc regio cingu Meth The s More The

RÉSoNANSS: a regional contribution to