Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

International Journal of Cast Metals Research

ISSN: 1364-0461 (Print) 1743-1336 (Online) Journal homepage: http://www.tandfonline.com/loi/ycmr20

Uncertainty analysis of metal-casting porosity


measurements using Archimedes' principle

R. P. Taylor, S. T. McClain & J. T. Berry

To cite this article: R. P. Taylor, S. T. McClain & J. T. Berry (1999) Uncertainty analysis of metal-
casting porosity measurements using Archimedes' principle, International Journal of Cast Metals
Research, 11:4, 247-257, DOI: 10.1080/13640461.1999.11819281

To link to this article: http://dx.doi.org/10.1080/13640461.1999.11819281

Published online: 29 Nov 2016.

Submit your article to this journal

Article views: 24

View related articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=ycmr20

Download by: [The UC San Diego Library] Date: 30 May 2017, At: 00:18
Uncertainty analysis of metal-casting porosity
measurements using Archimedes' principle

R. P. Taylor*, S. T. McCiain and J. T. Berry


Materials and Mechanics Laboratory, Department of Mechanical Engineering, Mississippi State University,
Mississippi State, MS 39762

A detailed uncertainty-analysis model is developed sensitivity coefficients in the uncertainty-propagation


for metal-casting porosity measurement using equation and also to investigate ways to reduce the
Archimedes' principle. The model is used to magnification of uncertainties in the weight measure-
investigate the influence of liquid density on the ments and estimates of the liquid and theoretical
sensitivity coefficients in the uncertainty densities. A numerical example is presented and dis-
propagation equation. The sensitivity coefficients cussed to demonstrate the process of estimating the
are found to be very large, resulting in significant measurement uncertainties and typical values for the
magnification of uncertainties in weight uncertainties that can be expected in metal-casting
measurements and liquid and theoretical density porosity measurements.
estimates. Increasing the density of the liquid
results in a decrease in sensitivity coefficients and Pyknometry
an improvement in porosity accuracy. A numerical Pyknometry is a process of comparing the relative
example is given to demonstrate the process of densities of a solid and a liquid. If the density of the
estimating the measurement uncertainties and liquid is known, the density of the solid can be
typical uncertainties that can be expected in metal- computed. The process is described schematically in
casting porosity measurements. Fig. 1. Three weight measurements are made: Ws =
dry-weight sample measurement, Wsb =buoyant-
Keywords: porosity, uncertainty, Archimedes, pyknometry weight sample plus basket measurement, and Wb =
buoyant-weight basket measurement. In Fig. 1, the
Introduction buoyant-weight measurements are made by suspend-
Pyknometry is a simple, effective, and inexpensive ing the sample using a wire basket in a cup of liquid
procedure to estimate the porosity of cast-metal sam- that is supported over the balance pan using a suitable
ples. The procedure consists of using Archimedes' bridge. As shown in Appendix I, the porosity m
principle to determine the sample density by weighing percent can be computed using the formula:
the sample in air and then in a buoyant liquid (usually
distilled water). This density is then compared with the P=l00(1-PL Ws ) (1)
theoretical density of the fully-dense, porosity-free Pth Ws- (Wsb- Wb)
alloy. Examples of the use of this technique abound; where PL is the density of the liquid and Pth the
typical results have been reported by Lee et al, Suri et theoretical value of the porosity-free density of the
al, and Villar. 1•2 •3 Power has presented an uncertainty- alloy. Equation (1) is the usual data reduction equation
analysis of a similar technique to determine the density for the porosity determination given by the stan-
of small aquatic animals using two fluids of different dards. 5•6 Sources of uncertainty in porosity are the
densities. 4 Standards describe test methods to deter-
mine material densities. 5•6 However, the reproducibil-
ity interval quoted by the ASTM Standard B311-93 is
0.03 gjcm 3 for specimens of the order of 5 g. This
corresponds to a porosity of approximately 0.5%
which is the same order as the porosity that is often
desired to be measured in metal castings; a careful
analysis of the uncertainties involved in the porosity
measurement process is required if the porosity results
are to be meaningful.
In the following, brief background sections on
pyknometry and uncertainty analysis are followed by
a detailed uncertainty-analysis model of the porosity
determination. The model is used to investigate the
Dry-W'eight Buoyant-W'eight Buoyo.nt-W' eight
SaMple Basket SaMple IS. Basket
*Author for correspondence
e-mail: taylor@me.msstate.edu Fig. 1 Pyknometry schematic

Int. J. Cast Metals Res., 1999, 11, 247-257 247


Metal-casting porosity measurements Taylor, McClain and Berry

R is computed using the data reduction equation and


measured and reference values Xi.
(3)
Two types of uncertainties exist in the individual
measured quantities, Xi. The first is precision uncer-
tainty which reflects the repeatability of the measure-
ments. The second is bias, which is a fixed or systematic
uncertainty. A bias could result, for example, from an
unnoticed zero shift on an instrument. In careful
experiments, biases are minimised by calibration.
Fig. 2 Liquid displacement However, they can never be totally eliminated. Cali-
bration standards always contain some small uncer-
tainty, which results in a bias uncertainty in the
measurements uncertainties in the weights and in the calibrated instrument. Reference values that are deter-
estimation of the densities. mined from a handbook or other source also contain
There is an additional conceptual error in Equation some uncertainty, and this uncertainty appears as a
(1) that is usually overlooked. When the tared-weight bias uncertainty.
wb is determined, the height of the liquid is slightly The precision uncertainty can be determined by
lower than when the buoyant weight of the sample plus repeating the measurement and determining the stan-
basket, Wsb, is determined. This is shown in Fig. 2. dard deviation. The standard deviation of the result
Therefore, the buoyant force on the basket is slightly can be estimated from the standard deviations of the
more when measuring Wsb than when measuring Wb. measurements:
This can be corrected for easily. Appendix I also shows
the derivation of the correction. The corrected formula
is: (sR)
R
2
t iil,
= (s x,)
2
X,
(4)

l
i=l

p = 100 [1 - PL Ws- (Wsb-


Pth
Ws
Wb)
2ND~NDb2)
(1 + De- The terms
coefficients:
Bx, are the normalized sensitivity

(2) - X 8R
Bx-.....!._ (5)
' - R ax.l
Comparing Equations (1) and (2) shows the addition
of a correction factor in terms of the cup diameter, De, The bias in the result is computed in a similar manner
the basket wire diameter, Db, and the number of wires, except correlations between biases must be accounted
N, suspending the basket. for:
To show the importance of the correction factor,
consider the example based on typical data from
our laboratory. A porosity-free cast-aluminum alloy
sample with a theoretical density, Pth = 2.685 gjml,
yields Ws = 15.8078 g, and Wsb- Wb = 9.9228 g using
water with PL = 0.9985 gjml. Substituting into where bx,xk is the covariance and 8ik is the Kronecker
Equation (1) delta:

8· - { 1 i= k (7)
P = 100 [1.o _ (o.oo75) zk- 0 i =/= k
app 2.685
The 95% level of confidence uncertainty is then given
X ( 15.8078 )] _ O 1l as:
15.8078- 9.9228 - .
Neglecting the added buoyancy results in an apparent
porosity of 0.11% for this porosity-free example.
Substituting into Equation (2) with, N = 2, De =
6.426cm, and Db= 0.1524cm gives Papp = -0.004%, where t95 is the appropriate multiplier based on the
which is much closer to the actual value of 0%. student-t distribution. As pointed out by Coleman and
Therefore, for porosities of the order of 1%, a signifi- Steele, Equation (8) can be approximated in most
cant error will result if the added buoyancy is engineering applications by taking t 95 = 2, which
neglected. For all cases in this paper, Equation (2) gives:
will be used.

Uncertainty analysis
The detailed uncertainty-analysis model is based on where PR and BR are the 95% level of confidence
that of Coleman and Steele. 7 The experimental result estimates of precision and bias limit.

248 Int. J. Cast Metals Res., 1999, 11, 247-257


Taylor, McClain and Berry Metal-casting porosity measurements

Uncertainty-analysis model uncertainty in a measurement is very small, but its


For the uncertainty analysis of the pyknometry mea- sensitivity coefficient is very large, the influence of that
surements, the uncertainty procedures outlined in the small uncertainty on the uncertainty of the porosity
Equations (3)-(9) are applied to the data reduction can be significant. On the other hand, if the sensitivity
equation (2). The result is the porosity, P, and the coefficient is much less than one, even large measure-
individual measurements are Ws, Wsb· Wb, PL· Pth• De, ment uncertainties may have negligible effects on the
and Db. Here PL and Pth are considered measurements computed porosity. Of course, the worst case is to have
for the purpose of the uncertainty analysis. Even though a large sensitivity coefficient and a large measurement
they are determined from reference sources and are not uncertainty. It is enlightening to study the sensitivity
directly measured, they do contain uncertainties. coefficients even before the uncertainties in the mea-
Applying Equation (4) but using precision, surements are known. Terms with small sensitivity
Px; = 2Sx;, instead of standard deviation yields the coefficients (much less than one) will not need accurate
equation for the precision uncertainty: measurements, while those with large sensitivity coeffi-
cients must be measured with great care.
2 2 2
PP) = (j2Ws (Pws) +B2 (Pwsb) The sensitivity coefficients are developed for the
( p Ws Wsb Wsb pyknometry problem in Appendix II. Looking at the
appendix reveals that the sensitivity coefficients, for the
weights, Bw,, iiw,b, and iiwb' are all strong functions of
the porosity of the sample and ofthe relative density of
the liquid PLI Pth·
It is interesting to investigate the influence of PL! Pth
and P on the sensitivity coefficients, since the density of
Also, applying Equation (6) gives for the bias the liquid is one of the few parameters that one can
uncertainty: control. Using typical numbers (Pb = 3.28 gjml, Pth =
2 2 2 2.713 g/ml, N = 2, Db= 0.1524cm, De= 6.426cm, and
Bp) = 02 (Bws) +B2 (Bwsb) Wdb = 0.2 Ws), the absolute values of iiw,, iiw,b, and iiwb
( p Ws W Wsb W
s sb are plotted versus PLI Pth in Figs. 3, 4, and 5 for P = 0.1,
0.5, 1, 5, and 10%. The figures show that the normal-
ized sensitivity coefficients for the weights can be very
large. For water and aluminum, PLI Pth ~ 0.37 and
iiw, ~ 300, iiw,b ~ 400, and iiwb ~ 80 with 0.5% poros-
ity. For porosity less than the order 0.5%, the sensitivity
coefficients can be huge. It would be unlikely that
porosity of the order 0.1% can be measured with
fidelity using pyknomotry. The large values of sensi-

+ 2BwsbBwb ( ~:) c:::) tivity coefficient mean that uncertainties in the weight
measurements are magnified, perhaps by several
orders of magnitude, when porosity is computed.
+2BwsBwb(~J (~:) When porosity of the order of 1% needs to be deter-
mined, great care must be taken in the weight measure-
ments. These sensitivities are greatly reduced at higher
+ 20neeDb ( i:e) (£:) (11) values of porosity.
Increasing the relative density of the liquid strongly
Here the only correlated bias uncertainties are those decreases the sensitivity coefficients. Using a liquid that
between the three weights and the one between the cup has a specific gravity of 1.5 (ethyl phthalate as reported
and basket wire diameters. Correlation of the uncer- by Villar) increases PL! Pth to 0.55. 3 This results in
tainties occur when the uncertainties in two measure- ~ensitivity coefficients of iiw, ~ 150, iiw,b ~ 200, and
ments arise in whole or in part from the same elemental Bwb ~ 40, which is a reduction by a factor of 2. _
sources. Here the same mass balance was used for the Fig. 6 shows a plot of the absolute value of ePth
weights, hence, the bias uncertainty in each weight is versus P. For porosity less than 1%, the sensitivity
the bias uncertainty in the mass balance, and all increases rapidly. For a porosity of 0.1 %, the magni-
uncertainties in the weights are perfectly correlated. tude is 1000. The sensitivity coefficient for PL has
The same is true for the diameters, since they were identical magnitude. Therefore, for porosities of the
measured with the same set of calipers. A more order 1% or less, the values for PL and Pth must be
detailed discussion of correlated biases can be found known with very good accuracy. Fig. 7 shows the
in Brown et al. 8 absolute value of iin, versus P. The curve has a similar
Inspecting Equations (10) and (11) shows that not shape to the one in Fig. 6, but the magnitude is much
only do the levels of the uncertainties in each measure- less. The sensitivity coefficient for Db has an identical
ment matter, but the amount that this is magnified by magnitude. For P greater than 0.2%, the sensitivity is
the sensitivity coefficients is also very important. If the less than 1. Uncertainties in the diameters are expected

Int. J. Cast Metals Res., 1999, 11, 247-257 249


Metal-casting porosity measurements Taylor, McClain and Berry

300
flws

200

100

0
0.2 0.4 0.6 0.8 1.0

PJPth
Fig. 3 Sensitivity coefficients with respect to W8

to have negligible influence on the uncertainty in water. The weight of the sample and basket in water
porosity when compared with the influence of the was recorded. The sample was removed from the
other measurements. basket, and the basket was replaced in the balance
and weighed in water. The temperature of the water
was then recorded. The density of the water was
Numerical example determined from the table in ASTM B311-93. Equa-
A numerical example of porosity measurement uncer- tion (2) was then used to calculate the percentage
tainty is presented for four samples of cast-aluminum porosity of the sample. The procedure was repeated
alloy 356 ranging in size from 3 to 15 g. The equipment for all of the samples.
used for the measurements was a Mettler H748AR The bias and random uncertainty of each measure-
mechanical balance, digital calipers, a 200ml beaker, a ment was determined in a variety of ways. A common
basket made from insulated copper wire, and a small way of determining the random uncertainty of a
Archimedean bench made of balsa wood. Water was measurement is to determine the standard deviation
used as the reference fluid. of a number of measurements. The precision limit, the
The particular procedure used was very similar to 95% level of confidence estimate of precision, is then
the procedure presented in ASTM B311-93. Distilled the product ofthe student-t value based on the sample
water was obtained and about 0.5 g of detergent was size and the standard deviation of the measurement. If
added per gallon of water. Both the water and samples a large sample size is assumed, the precision limit is
were allowed to reach the ambient temperature in the simply twice the standard deviation limit of the mea-
laboratory. The sample was then weighed in air. The surements. The bias limit, the 95% level of confidence
Archimedean bench and the beaker were placed in the estimate of bias, is an estimate of the fixed or systematic
balance, and the sample and basket were hung from error of the measuring device. In some cases, the bias
the balance so that the sample was submerged in the limit is estimated based on the available information

400

300
~sb i
. I
200 ·'-; r-- -- --- --- ---
~--

·l.

100

0.2 0.4 0.6 0.8 1.0

Fig. 4 Sensitivity coefficients with respect to Wsb

250 Int. J. Cast Metals Res., 1999, 11, 247-257


Taylor, McClain and Berry Metal-casting porosity measurements

80

..... 60
~

20

0.2 0.4 0.6 0.8 1.0

PJPth
Fig. 5 Sensitivity coefficients with respect to Wb

1000
I
I
800 -1-·---··--·-
I

600
o;th
400

0 2 3 4 5
p

Fig. 6 Sensitivity coefficients with respect to Pth

I
I
2.0 ---t
I
I
_ _l __ _
I I
I I
I I
1.0 --1 --- -- r- --
1 I
I I
0.5 ---+ -- - - + -- f-- ---
1 I I
I I I
0.0 L--~_:===+=====+=======i==~
0 2 3 4 5
p

Fig. 7 Sensitivity coefficients with respect to De

and best engineering judgement. Guidance on estimat- times each. The standard deviations of both the mea-
ing bias limits can be found in Coleman and Steele. 9 surements were 0.0001 g. Since the standard deviations
The precision limit of the measurement of the of the measurements were the same, the standard
sample weights in air was determined by measuring deviations of the all of the dry weight measurements
the weights of a 5 g sample and a 15 g sample fifteen were taken to be 0.0001 g. By measuring calibrated

Int. J. Cast Metals Res., 1999, 11, 247-257 251


Metal-casting porosity measurements Taylor, McClain and Berry

Table 1 Summary of the component diameter and precision limits were determined to be
uncertainties 0.150 cm and 0.004 cm for the wire basket and
6.433cm and 0.106cm for the beaker. The biases of
Measurement Bias Limit. B Precision Limit. P
the diameter measurements were estimated to be the
resolution of the calipers, 0.001 cm.
Ws 0.0002 g 0.0001 g The final component uncertainties to be estimated
Wsb 0.0002 g 0.0070 g are the uncertainties of the theoretical density of the
wb 0.0002 g 0.0048 g alloy. Unfortunately, this is the most difficult uncer-
Pw 0.0001 g/cm 3 0.0001 g/cm 3 tainty to estimate. Slight differences in alloy composi-
De 0.001 cm 0.106cm
0.001 cm 0.004cm
tion can change the theoretical density. Sigworth et al.
Dw
Pth 0.0006 g/cm 3 0.0015 g/cm 3 noted that Sr-modification of aluminum A356
increased the theoretical density from 2.6672 gfcm 3 to
2.6775 g/cm 3 . 10 The localised effects of macrosegrega-
tion of elements, especially those of atomic weights
standards and using best engineering judgement, the which are greater than those of the principle alloying
bias limit of all measurements made with the mechan- element (for example, copper or iron in aluminum
ical balance was estimated to be 0.0002 g. Because all alloys) have previously been cited 11 and are difficult
weight measurements were made with the same balance, to overcome unless knowledge of the element's con-
all weight measurement biases are correlated. centration profile is known. Higher iron contents in
There are many sources of random uncertainty in the similar alloys, such as aluminum alloy 356, can
wet measurements Wsb and Wb. During the measure- increase the theoretical density to over 2. 7 g/cm3 . 12
ments in water, bubbles can attach to the sample or the The theoretical density and the bias and precision
basket, water can seep into any surface pits on the uncertainty limits of the samples used here, Sr-modified
sample, small variations in water level can be caused and grain refined aluminum alloy 356, were determined
by losing water that adheres to the sample or basket, for the sample created from the same melt with the
or convective currents may cause variations in the highest measured density. The porosity was measured
measured weight of the sample. The standard devia- using metallography and image analysis, and the the-
tion of the weights of the sample and basket in water oretical density of the sample was determined from:
and the standard deviation of the weight of the basket
in water were determined by repeating the measure- lOOps
Pth =lOO_ p (12)
ments thirty-one times for an 18 g bar of wrought
3
aluminum alloy 6061-T6. The standard deviation of The value for the density was 2.7251 g/cm , with
3 3
Wsb was 0.0035 g. The standard deviation of Wb was Bprh = 0.0006gfcm and Pprh = 0.0015gfcm . Table I
0.0024 g. Thus, the precision limits of Wsb and Wb were presents a summary of the component uncertainties.
0.0070 g and 0.0048 g, respectively. The densities of four samples of cast-aluminum alloy
The density of the water was determined by measur- 356 were determined. The dry weights of the samples 1,
ing the temperature of the water and reading the 2, 3, and 4 were 3.0779 g, 7.2684 g, 11.6569 g, and
density of the water based on the table presented in 15.3101 g, respectively. Table 2 summarises the infor-
ASTM B311-93. The uncertainties associated with mation gathered for each of the samples. These values
density of the water are caused by the uncertainties along with the uncertainty estimates in Table 1 were
in the water temperature measurement. The thermo- substituted into Equations (10) and (11) to compute
meter used had a bias limit of 0.5 oc and a precision the precision and bias uncertainties. The overall uncer-
limit of 0.5 ac. From the table in ASTM B311-93, an tainty, Up, was determined using Equation (9). Table 3
error of0.5 ac correlates to an error of0.0001 g/cm 3 in presents the porosity of each sample and the uncer-
the water density. Hence, the bias limit and the preci- tainty of each of the measurements.
sion limit of the density of the water are 0.0001 g/cm 3 . Table 3 shows that the uncertainty in the porosity
The diameters of the wire basket and the diameter of measurement was unacceptable for the smaller samples.
the 200 ml beaker were determined with digital cali- The porosity of sample 1 was found to be 1.8143 ±
pers. Each diameter was measured fifteen times. The 0.7264 or 40% of the porosity measurement. The

Table 2 The pyknometry data

Sample 1 Sample 2 Sample 3 Sample 4

3.0779g 7.2684g 11.9569g 15.3101 g


4.2048g 6.8338g 9.7759g 11.8855 g
2.2763g 2.2746 g 2.2731 g 2.2696g
21 oc 21 oc 21 oc 21 oc
0.9980 g/cm 3 0.9980g/cm 3 0.9980 g/cm 3 0.9980 g/cm 3

252 lnt. J. Cast Metals Res., 1999, 11, 247-257


Taylor, McClain and Berry Metal-casting porosity measurements

Table 3 The sample porosities and uncertainties for a precision component uncertainty, and
c2 = R2e .e Bx;BxJ (16)
Sample Porosity% Bp Pp Up Xz X} XX
I J

Sample 1 1.82 0.026 0.73 0.73 for a correlated bias term between the component
Sample 2 1.64 0.024 0.31 0.31 biases of X; and J0. By the definition of the NPC,
Sample 3 1.58 0.024 0.19 0.19 the sum of the NPCs for all component uncertainties
Sample 4 1.43 0.024 0.15 0.15 must equal unity. Table 4 presents the NPCs for all
component uncertainties of the four porosity measure-
ments. The NPCs are listed in decreasing order of
uncertainty of the porosity measurement of sample 2 contribution.
was better, but it was still unacceptable at 0.3114. The Table 4 reveals that the precision errors in the
porosity measurements of samples 3 and 4 were accept- measurements of Wsb and Wb dominate the total
able in that their uncertainties are about 10% of the uncertainty of a porosity measurement. Pwsb and
porosity measurements. Table 3 also shows the domi- Pwb contribute more than 99% of the total uncertainty
nance of precision errors in the uncertainty of the in porosity of sample 1 to 91 % of the total uncertainty
porosity measurements. The precision components in the porosity of sample 4. The only other measure-
were all at least seven times the bias uncertainties. ments which contribute more than 0.1% to the total
While Table 3 shows that the total uncertainty of uncertainty in any one porosity measurement are the
each porosity measurement is dominated by precision bias and precision of the theoretical density of the
errors, a normalized criterion will reveal which com- alloy, the bias and precision of the density ofthe water,
ponent uncertainty causes the most uncertainty in the and the precision of the diameter of the basket wire.
resulting porosity measurement. A common normal- The maximum contributions of Pp 1h, Bp 1h, PPL' BpL,
ized criterion is the normalized percentage contribu- and PDb are 5.5%, 2.0%, 0.41 %, 0.41 %, and 0.14%,
tion (NPC), presented in Equation (13) in terms of respectively, in the porosity measurement of sample 4.
Equations (4) and (6). Table 4 also demonstrates that when one weight
2
NPC = £_ (13) measurement is subtracted from another, such as
u~ ( Wb- Wsb) in Equation (2), the correlated bias term
where: Bwsb, Wb cancels the contributions of both of the bias

c2 = R28L ( i;;Y (14)


terms of each measurement, Bwsb and Bwb· The biases
of Wb and Wsb each are also correlated to the bias of
W 5 • Table 4 shows that the contributions of Bws.Wb
for a bias component uncertainty,
and Bws.Wsb are equal and opposite. Thus, when one
c2 = R28I;(~i)
2

(15) measurement is directly subtracted from another mea-


surement and their biases are equal and completely

Table 4 The normalized percentage contributions

Normalized Percentage Contribution


Component
Uncertainty Sample 1 Sample 2 Sample 3 Sample 4

Pwsb 0.68 0.67 0.64 0.62


Pwb 0.32 0.31 0.3 0.29
Ppth 0.0025 0.013 0.035 0.055
Bpth 0.00089 0.0048 0.013 0.02
BpL 0.00018 0.001 0.0026 0.0041
PpL 0.00018 0.001 0.0026 0.0041
Pob 0.000062 0.00034 0.00088 0.0014
Poc 0.000024 0.00013 0.00034 0.00053
Bwsb 0.00056 0.00054 0.00053 0.00051
Bwb 0.00056 0.00054 0.00053 0.00051
Bws 0.00022 0.00021 0.00021 0.0002
Bob xxxxxxx 0.000021 0.000055 0.000086
Pws 0.000054 0.000054 0.000052 0.00005
Boc xxxxxxx xxxxxxx xxxxxxx xxxxxxx
Boc,Db xxxxxxx XXX X XXX xxxxxxx xxxxxxx
Bws.Wb 0.00069 0.00068 0.00066 0.00064
Bws,Wsb -0.00069 -0.00068 -0.00066 -0.00064
Bwsb,Wb -0.0011 -0.0011 -0.0011 -0.001

XXXXXXX -values less than 1.0 X 10-5

Int. J. Cast Metals Res., 1999, 11, 247-257 253


Metal-casting porosity measurements Taylor, McClain and Berry

1.0 .------,-1---~~--~~---

1 I I
0.8 1-----+----+------+----
~
0 I I I
c:
1-----L ___ j__________L___ _
~<D 0.6
0
c:
0
I I I
:::J o I I I
0.4 1------0-9---T---~----

~
t
o 0

0.2 - - - - ~- __: __: ~ ~0~ -~ ~ ~ ~o___c


I I I
I - - - - 1I - - - - - +
0.0 + - - - - - + I -----l
0 5 10 15 20
Number of Measurements, J

Fig. 8 Effect of using the mean of J measurements of Wb and Wsb on the total uncertainty in porosity

correlated, neither of the biases contributes any uncer- water (u = 0.073 N/m). 14 The lower surface tension
tainty to the reduced data. should reduce the random uncertainty of Wsb and
The domination of precision in the uncertainty of a W6 due to bubble formation. Bubble formation is a
porosity measurement leads to an effective method of source of random uncertainty in any weight measure-
reducing the total uncertainty of a porosity measure- ments in water. By using a denser fluid with a lower
ment. If the average of several measurements is found surface tension, bubble formation should be greatly
and used to compute the porosity, the uncertainty is reduced.
reduced from the uncertainty with one measurement.
If Wsb and W 6 are each measured J times and the mean
of each value is used in Equation (2), the appropriate
precision limits areY Conclusions
The example demonstrated that the main source of
and P-wb
Pw uncertainty in the porosity measurement is the preci-
=--b
/] (17)
sion of Wsb and Wb. The precision in Wsb and W 6
Fig. 8 shows the effect of the number of measurements contributed over 90% of the uncertainty in the poros-
of Wsb and W 6 on the total uncertainty of sample 1. ity measurements of samples of 15 g or less. In samples
Averaging four readings of Wsb and W 6 decreases the less than 10 g, the precision in W.,6 and W6 caused
uncertainty by about one half. Using more than 10 the uncertainty in the porosity measurement to be
readings adds little to the fidelity of the porosity unacceptable. By making repeated measurements of
determination. Table 5 shows the recomputed uncer- Wsb and Wb, the precision limits were reduced, and the
tainties when three measurements of Wsb and W 6 are uncertainty of a porosity measurement was signifi-
used. The reduction in uncertainty of samples 1 and 2 cantly lowered.
is seen to be significant. The uncertainty analysis model revealed that the
Fluids denser than water that are used to lower the fluid used in the porosity determination greatly affects
normalized sensitivity coefficients may produce an extra the normalized sensitivity coefficients of W., W56 , and
benefit. Tetrachloroethylene (s.g. = 1.6, u = 0.032 Njm) Wb. While water is normally used for porosity deter-
may also lower the uncertainty of a porosity measure- mination, choosing a fluid such as ethyl phthalate
ment because its surface tension is about half that of (s.g. = 1.5) or tetrachloroethylene (s.g. = 1.6) can
reduce the sensitivity coefficients of w., w.b, and wb
by a factor of 2 for aluminum alloys. A reduction in the
Table 5 The sample porosities and uncertainties normalized sensitivity coefficients will reduce the
with the average of three weighings in uncertainty in the porosity measurement.
water Thin section castings and adequately degassed and
fed castings can contain low levels of porosity. To
measure the porosity with fidelity, the simple applica-
Sample Porosity% Bp Pp Up
tion of Archimedes' principle using water and single
Sample 1 1.82 0.026 0.42 0.42
measurements may need to be altered. By choosing a
Sample 2 1.64 0.024 0.18 0.18 suitable liquid and by making repeated measurements,
Sample 3 1.58 0.024 0.11 0.12 Archimedes' principle is useful in determining the
Sample 4 1.43 0.024 0.093 0.096 porosity of small cast specimens with low levels of
porosity.

254 Int. J. Cast Metals Res., 1999, 11, 247-257


Taylor, McClain and Berry Metal-casting porosity measurements

Acknowledgments bR - bias in the result at the standard deviation level


The authors thank the U.S. Army Research office for Db - basket-wire diameter
partial financial support of this work under the DOD De - liquid-cup diameter
g - gravitational acceleration
EPSCoR program. Ms. Anne Sculthorpe's assistance
h - height
preparing the figures is greatfully acknowledged. N - number of wires suspending basket
P - porosity
p app - apparent porosity
References PR - precision limit of the result
1. Y. W. Lee, E. Chang, and C. F. Chieu, "Modeling of R - result of the experiment
Feeding Behavior of Solidifying Al-7Si-0.3Mg Alloy SR - standard deviation of the result
Plate Casting", Metallurgical Transactions, 1990, 21B, S x 1 - standard deviation of measurement X 1
715-722. t95 - student-t multiplier for 95% level of confidence
2. V. K. Suri, A. J. Paul, N. El-Kaddah, and J. T. Berry, Vs - volume of sample
"Determination of Correlation Factors for Prediction of W - weight
Shrinkage in Castings-Part 1: Prediction of Micropor- wb - buoyant-weight of the basket
osity in Castings, A Generalized Criterion", AFS Trans- wdb - dry weight of the basket
actions, 1994, 21B, 861-867. Ws - dry-weight of the sample
3. P. C. Villar, Study of Neural Networks for Porosity Wsb - buoyant-weight of the sample plus basket
Prediction in Aluminum Alloy A356 Castings, MS Thesis,
Mechanical Engineering Department, Northwestern Greek symbols
University, Evanston, IL, 1997. PL - liquid density
4. J. H. Power, "Errors Associated With Using Archi- Ps - sample density
medes' Principle to Determine Mass and Volume of p,, - theoretical porosity-free density
Small Aquatic Organisms", Hydrobiologia, 1996, 335, {j - sensitivity coefficient
141-145.
5. ASTM Standard B311-93, Test Method for Density
Determination for Powder Metallurgy ( P/ M) Materials
Appendix I
Containing Less Than Two Percent Porosity, American
Society for Testing and Materials, Philadelphia, PA. Derivation of pyknometry formulas
6. ISO Standard 3369, Impermeable Sintered Metal Mate- The relative density of the solid is computed using Archi-
rials and Hard Metals-Determination of Density, Amer- medes' principle as shown in Fig. 9. The weight of the sample
ican National Standards Institute, New York, NY. W is the vector sum of the apparent weight, W', plus the
7. H. W. Coleman and W. G. Steele, "Engineering Appli- buoyant force, B:
cation of Experimental Uncertainty Analysis", AIAA W=W'+B=W'+pLVsg (1-1)
Journal, 33, No. 10, 1888-1896.
8. K. K. Brown, H. W. Coleman, W. G. Steele, and R. P. The sample weight is approximated by the dry weight Ws
Taylor, "Evaluation of Correlated Bias Approximations since the buoyant force in air is negligible. The sample
in Experimental Uncertainty Analysis". AIAA Journal, volume can be expressed in terms of the sample weight and
1996, 34, No. 5, 1013-1018. density as:
9. H. W. Coleman and W. G. Steele, Experimentation and V-s-
Ws (1-2)
Uncertainty Analysis for Engineers (John Wiley & Sons, Psg
New York, 1989). The apparent buoyant weight can be computed using the
10. G. K. Sigworth, C. Wang, H. Huang, and J. T. Berry, tared-weight:
"Porosity Formation in Modified and Unmodified Al-Si (1-3)
Alloy Castings", AFS Transactions, 1994, 102,245-261.
Substituting Equations (I-2) and (1-3) into (1-1) and solving
11. J. T. Berry, "Linking Solidification Conditions and
give the expression for the solid density:
Mechanical Behavior in Aluminum Castings, A Quarter
Century of Evolving Evidence", AFS Transactions, 1995,
(I-4)
103, 837-847.
12. S. T. McClain, A Study of Porosity Quantification Tech-
niques and Pore Morphology in Aluminum Alloy Castings,
MS Thesis, Department of Mechanical Engineering, w'
Mississippi State University, Mississippi State, MS, 1997.
13. W. G. Steele, R. P. Taylor, R. E. Burrell, and H. W.
Coleman, "Use of Previous Experience to Estimate
Precision Uncertainty of Small Sample Experiments",
AIAA Journal, 1993, 31, No. 10, 1891-1896.
14. J. A. Dean, Lange's Handbook of Chemistry (McGraw-
Hill, New York, 1992).

Nomenclature
B - buoyant force
BR - biaslimitoftheresult Fig. 9 Archimedes' principle

Int. J. Cast Metals Res., 1999, 11, 247-257 255


Metal-casting porosity measurements Taylor, McClain and Berry

The porosity is then computed by comparing the sample


density to the theoretical value of porosity-free density for
the alloy:
P = 1oo(1- _&_) (1-5)
Pth (Ws- (Wsb- Wb))(Ws- (Wsb- Wb)- AWs)
Substituting: (11-3)
P = 100(1- PL ws ) (1-6)
Pth Ws- (Wsb- Wb)
Equation (1-6) is the data reduction equation for the porosity
determination. Sources of uncertainty in porosity are the
measurement errors in the three weights and errors in the (Ws- (Wsb- Wb))(Ws- (Wsb- Wb)- AU/,)
estimation of the two densities. (11-4)
There is an additional conceptual error in Equation (1-6)
that is usually overlooked. The buoyant tared-weight given B = PL 8P = D~
by Equation (1-3) is not strictly true. When the tare-weight PL p 8pL D~ -ND1
wb is determined, the height of the liquid is slightly lower _PL W
than when the buoyant weight of the sample plus basket is s
X Pth (11-5)
determined, as demonstrated by Fig. 2. Therefore, the buoy- (Ws- (Wsb- Wb)- AW,)
ant force on the basket is slightly more when measuring Wsb
than when measuring Wh; this can be corrected for easily. B = Pth .!!!__ = D~
The displacement height is determined by equating volumes: p,, p 8p,h D~- ND~

h= 4Vs 4Ws (l- 7) PL w


1r(D~- ND~) 1rPs(D~- ND1) X Pth s = _(j ( 6)
(Ws- (Wsb- Wb)- AWs) PL ll-
where Db is the diameter of the suspension wires, De is the
diameter of the cup, and Nis the number of suspension wires. B - Db 8P - 2ND~D1
The appropriate correction to Equation (1-3) is: D• - P 8Db - (D~ - ND1) 2
_PL W
1
W = Wsh-
( 7rD14WsPL
Wb- N 4 1rPs(D~ _ ND~)
) (1-8) X Pth
s
(11-7)
(Ws- (Wsh- Wb)- AWs)
Substituting into Equation (1-4) and solving for Ps gives:
B _ De 8P _ 2ND~D~

DZ~D~D~)
D, - P 8Dc - (D~ - ND~) 2
PLws(1 +
PL ws
Ps = Ws- (Wsh- Wb) (l- 9 )
X Pth = -iJD (11-8)
(Ws-(Wsb-Wb)-AWs) '
Using Equation (1-5) the corrected porosity formula is:
The one parameter that can be controlled to try

_
100 1
PL) W
_ ( Pth s
(l
+ ND1 ))
D~ - ND~
(1-10)
to minimise the sensitivity coefficients is the relative density
of the liquid PLI Pth· To study this, it is convenient to convert
p - ws - ( wsb - wh) the coefficients into functions of porosity and relative density
( using the relationships:
Pth ) =Ws ( 1- 100
fV,h-Wb=Ws ( 1-p;A _pA ) (11-9)
100

Appendix 11
Derivation of the pyknometry sensitivity
wb = wdb(1- ;~) (11-10)

coefficients Substituting into the sensitivity expressions and rearranging


Equation (2) is rewritten as: gives:
100- P ( 1 100 )
P=100(1- WsA
Ws- (Wsb- Wb)
) (11-1)
B - ----wo--IOO=P A
(11-11)
with w,- A(~-) 1
100- p
A = PL
Pth
(1 + 2ND~2)
De- NDh 100 -P(l-~A)+ Wdb( 1 - Pth PL)
B - - 100 100 - p Ws Pb Pth
The normalized sensitivity coefficients are readily evaluated w,.- ( 100 )
by taking the partial derivatives: A 100-P- 1

B - ws 8P (11-12)
w,- p aws
Wdb 100- P (l _ Pth PL)
(j _ Ws100 Ph Pth
(Ws- (Wsb- Wb))(Ws- (Wsb- Wb)- AWs) (11-13)
w,- ( 100 )
A 100-P- 1
(11-2)

256 lnt. J. Cast Metals Res., 1999, 11, 247-257


Taylor, McClain and Berry Metal-casting porosity measurements

O--D; 1 (11-16)
2 2 2
PL- Dc - ND b ( 1 N Db ) ( 100
- - - - 1)
+ D~ - ND~ 100- P
(11-14) (11-17)
(11-15)
The first three coefficients are strong functions of both P and
(j _ _ 2ND;D~ pL!Pth. The last four are not dependent on PLI Pth but are
D• - (D~ - ND~)2 strong functions of P.

(Received 10 August 1997, accepted 30 July 1998)

lnt. J. Cast Metals Res., 1999, 11, 247-257 257

You might also like