STRESS CORROSION OF E-GLASS FIBER G
.
K
.
SC H M I T 2 A N D A
.
G
.
M
ETCA L
FE
,
Solar, A Dioision of Infmotionnl Horoerf,. ..~, . ..
~
~~~
Stress corrosion on glass exposed to water vapor has long been assumed to occur on existing surface defects by a continuous process. The present work on glass fila1 ments has shown that incubation and flaw growth by corrosion constitute the two stages preceding fractuire. In the incubation period, water reacts with cations in the glass leading to hydrolysis and increased concentrotion of hydroxyl ions. The incubation period continues until this conctentrotion reaches the level ntxessary for corrosion to occur; this stage occupies most of the life of the fiber. Subsequent corrosion of the surrounding silico network takes place until the , , . critical size flaw has formed, wnen manure ensues. StnE.SS corrosion was studied on E-glass fibers under constant load i in both 50 and 100% relative humidiIty. cIontrary to earlier assumptions, the growth rote of flaws was not controlled by a single exponent governing the growth equation. Stress corrosion occurs during typical tensile Itests, and restJlts of such tests were correlate d with stress corrosion data.
=TRESS
corrosion occurs in metals, ceramics, and plastics
..... ....
under carefully controlled conditions, so that uniform filament strengths could be achieved and fiber diameters could be held in close tolerance. Typical strength values \yere 450,000 to 520,000 p.s.i., with an over-all average of 430.000 p.s.i. Diameters were held within 38 to 40 X 10-j inch. determined from measurements under 500x magnification in a Leitz microscope with bifilar eyepiece. T h e draiving facilities permitted drawing of 70-inch-long monofilaments. measured between bushing and winding drum. Generally, eight to ten monofilaments were drawn during one drawing period. T h e filaments were wound on serrated frames suitable for 1-inch free gage length, and were stored in airtight boxes containins Drierite. Control of Filament Strength. One or two monofilaments from each drawing period were checked for tensile strength at 12 locations, using a multiple tensile tester at standard strain rate of 0.06 min.-’ in laboratory atmosphere of approximately 50% RH. These tests were made within 1 to 2 hours after drawing. and served to check drawing performance. T h e tensile strength of monofilaments used for static fatigue tests was obtained from five specimens from each monofilament, again tested under standard conditions. T h e strength data were used for correlation between monofilaments and as a base line for evaluation of strength of fibers surviving static fatigue tests. Sampling Procedures. Because of the importance of sampling in statistical measurements of brittle materials, much attention was paid to the sampling procedures. Three sets of five fibers each were obtained from each 70-inch monofilament. One set was used for the control and reference data mentioned above. T h e two other sets were used for the static fatigue tests. In test series I (Table 11): one set each was assigned to 50 and 100% RH, respectively, at one stress level. In test series 11, one set from each filament was tested at one applied stress, whereas the second set had been used for preliminary tests in 100yc RH at different stresses, but are not reported here. Sampling in the latter tests was entirely random, since each specimen of a group of five had been loaded to one of the programmed stresses. However. the long test duration of the low-load fibers proved to be too timeconsuming for the over-all program and the method had to be discontinued in test series 111. In these experiments. all fibers of one group were loaded to the same stress. Since nvo groups were available from one monofilament: a certain correlation could be achieved by testing the groups at two different stress levels within the programmed stress range (Table 11). T o compensate for the simultaneous usage of five fibers from the same source, a large number of tests were made at each stress level. As it turned out: the distribution of failure times of a given group of five fibers was rather wide and. consequently, the effect of biased sampling \vas minimized. Test Procedures. T h e frame-mounted fibers were inserted simultaneously into the molten sealing wax of the fiber gripping
rods. T h e work reported in this paper relates to E-glass (calcium-magnesium aluminoborosilicate) monofilaments. T h e results obtained are a t variance with present theories of stress corrosion of glasses, and are presented a t this time to inform others interested in this field. Experimental
Test Program. Four series of static fatigue tests were conducted to study stress corrosion phenomena. T h e program is summarized in Table 11, listing the applied stress levels and the number of test specimens for static fatigue tests and for control fibers tested in tension. Also listed are gage length and purpose of each test series.
[To avoid confusion, the following terminology has been adopted for this discussion: “Static fatigue” means the test; fibers tested in static fatigue are subject to ”stress corrosion” (which may only be part of the total failure mechanism). A ”stress corrosion process” includes both incubation period and corrosion period, as well as fracture. T h e term “stress corrosion mechanism,” then. denotes a specific sequence of events that occur in the course of the stress corrosion process.] Instrumentation. T h e static fatigue tester is shown in Figure 1 . This apparatus permits simultaneous testing of five fibers and can be adjusted for gage lengths from 0.1 to 7.5 inches. Levers carrying the predetermined dead weight load are oil-dampened to prevent accidental failure resulting from shock loads during both fiber loading and fiber fracture. T h e tester was contained in a humidity- and temperaturecontrolled chamber (Model Vapor-Temp, Blue M) . Actual humidities and temperatures obtained during the nominal 100% relative humidity (RH) were 96 to 9970 R H a t temperatures between 82’ and 86’ F . T h e chamber was not used for the tests at 507,RH. T h e test apparatus. as well as the chamber, was shockmounted to eliminate effects on fatigue life from vibrational sources, which are most detrimental in long-time tests a t low applied stresses. Effectiveness of the damping system was checked by placing a tray of Lvater on the tester. Only very slight ripples could be observed occasionally, and this was taken as an insurance for reliable test results. Failure times were recorded by five electrical digital clocks operated individually by small mercury switches located on the loadbearing arms of the tester (Figure 1 ) . ‘The clocks recorded the time in minutes, with the last digit indicating tenths of a minute in continuous rotation. This allowed time t o be interpolated within 2 seconds and was adequate for the purpose of this investigation. Fiber Drawing and Storage. E-glass filaments were drarvn at Solar from a conventional one-hole platinum bushing
Table II. Ajjlied Stress, Test Series
I I1 111
a
2
K.S.I. 320 320 480 440 400 360 320 280 240 200 150 320
Static Fatigue Program Summary
of Test Specimens Static Fatigue A t 50% A t 700% Controls at RH RH 50% RH 40 40 32 28
..
..
15 30 30
. .
.. ..
30
32 15 30
l&EC
PRODUCT RESEARCH A N D DEVELOPMEN1
,
1 0 ,
Remarks
Study effect of moisture, and of strength retention of fibers surviving static fatigue tests Study strength-failure time relation
30 30 30 30 20 20 15
30 30 20 20 .. 15 , . 10 (1007, R H only) Fiber washing 10 (lOO~oR H and water coating) Controlfibers tested a t standard tensile strain rate 0.06 min.-’ .. .. ..
Gage Length, Inch 0.1
1,0
Preliminary investigation of effect of excess water on corrosion mechanism
points (Figure 1) after the desired humidity in the chamber had been stabilized for some time. Alignment of the specimens was achieved by fixed positioning of the serrated frames which were machined to close tolerance. The temporarily removed Plexiglas bell j a r enclosing the tester was replaced immediately and the humidity was restored while the wax cooled. After 5 minutes, the “zero position” cam shaft was turned manually through a window in the side of the bell jar, thereby loading the fibers simultaneously in approximately second. The clock circuits were activated at the same time by a master switch. Failure times of specimens breaking immediately upon load application (mainly a t high stresses)
$
were recorded as “failed within 0 to 2 seconds” and are plotted a t 1 second on the logarithmic time scale of the respective diagrams. One of the analytical tools employed in this study required the tensile strength of unbroken test fibers to be determined after a certain exposure time to static fatigue. This was accomplished by means of single lead shot of 0.07-gram average weight being fed into the load baskets (Figure 1 shows weights instead of baskets). The shot were fed out of a small glass tube, one at a time, in a controlled succession until the fiber broke (typical failure load equaled 20 grams).
300
100
1
lo2
10 Figure 2.
lo3 lo4 TIME TO FAILURE, sec.
lo6
10”
Static fatigue data from 1-inch virgin E-glass fibers in
10
1 0 0 ~ RH o
Test series 111
Results
99
I
0
1
2
3
log T I M E ,
4
J
6
7
seta
Figure 3. Failure time probability plots of various applied stresses Test series 111
Figure 2 shows the customary plot of applied stress us. time to failure on a log scale. The number of test points per stress level is given in Table 11. Fibers that failed within 2 seconds after initial loading in the static fatigue test are plotted at the 1-second time mark. Open circles with arrows denote unbroken test specimens that were tensile-tested a t the plotted times. In Figure 3. logarithmic failure times of fibers tested a t the various stress levels are plotted on Gaussian probability paper. Best fitting curves are drawn through the data points. based on the general pattern. Deviations from actual failure times. specifically a t 360 k.s.i. (1000 p.s.i.), are assumed to have originated in variation of initial filament strength. Figure 4 shows the static fatigue curve as derived from the best fitting curves of the failure time plots in Figure 3 , a t the 50% probability level. The 507, time was adopted as a convenient method to deal with the problem of choice of representative values. One of the advantages of this method is that the effect of immediate (0- to 2-second) failures, as well as of the nonfailed (surviving) specimens. is minimized, specifically if a family of curves is available for evaluation. The tensile strength distribution of unbroken (surviving) fibers from series 111 is shown in Figure 5. The strength distribution of the control fibers is shown for comparison, but this was determined a t 50% RH and a strain rate of 0.06 min.-’ Uncertainties in the corrections to be applied to make the control strength data appropriate for 1007, RH leads to a band, as shown in Figure 5. For the average strength, the VOL. 5
NO. 1
MARCH 1966
3
0.1
I
1
I
APPLIED STRESS (ksil 320
DJUSTED 100% R H
+
280
e
0
100
-L =
"\,
1 LllCll
D = J8\lO-'inch
COXTROL FIBERS, 50%
I
0
4
correction is 0.95 X (strength at 50% R H ) . Figure 6 shows additional tensile strength distributions of surviving fibers from series I and 11. Analysis of Results
Static fatigue data obtained in 50 and 100% R H are analyzed with respect to strength loss with time, effect of humidity on the life of fiber, and strength retention of fibers after prolonged exposure to stress corrosion. Static Fatigue Curve. T h e plot of strength LIS. failure time (Figure 2) shows the usual spread over 2 to 3 decades of time for each of the stress levels. T o arrive at a meaningful static fatigue curve. the 50% probability of failure time has been used for the different applied stresses (Figure 3). Best fitting curves are drawn. taking into consideration the recognizable general pattern. T h e failure time values at 50% probability have been read from the curves to plot the fatigue curve in Figure 4. This curve is terminated at 440.000 p.s.i. because of the high rate of immediate failures a t 480,000 p.s.i. T h e slight curvature of the extrapolated section is necessary because a n upper limiting stress must be approached asymptotically. At lower stress levels. the curve shows a marked downward curvature. T h e validity of this curvature is supported by the uniformity of tensile strength data of control fibers (Table 111).
Strength of Control Fibers in Static Fatigue Tests Tensile Strength D a t a of Control L o t Coeficient of Stattc Fatigue Ar: , strength, A.0. k.s.i. oariation, % Stress, K.S.I.
Table 111.
480 440 400 360 320
280 240
200 150
1.. 5
30 30 30 30 30 20 20 15
47 5 503 495 483 505
473 475
48 6 488
7.6 12.1 8.3 10.3 8.0 10.9 12.6 9.2 7.4
Effect of Humidity. T h e effect of humidity (50ojC us. 1007, R H ) on the fatigue life of fibers has been investigated on a set of comparable sample populations of fibers of 0.1-inch gage length and an applied stress of 320,000 p.s.i. (test series I ) . Evaluation by means of failure time probability plots shows a life increase of 40y0for the lower humidity. Strength Retention after Exposure to Stress Corrosion. Fibers in static fatigue, not failed at the time of test termination, 4
l&EC PRODUCT RESEARCH A N D DEVELOPMENT
99
OL
I
200
300
I
I
I
400 500 FAILCRE S T R E S G T H , k s t
600
Figure 5. Tensile strength distribution of unbroken fibers of test series 111 r . A D J U S T E D 100% R H
FAILURE STRENGTH (Sei i e s I) A
FAILURE STRENGTH fSeries IIJ B
FAILURE S T R E S C T H ( S e r m s 111) C
Figure 6. Tensile strength distributions of unbroken fibers from static fatigue tests in 50% and 1 OOyo RH
were loaded to failure. T h e respective tensile strength data are correlated with the tensile strength of control fibers in two ways : tabulated average strengths and failure probability plots. Table I V lists the data of test series I through 111 for control fiber strength (column 1). applied stress levels (column 2), and tensile strength after termination of static fatigue tests (column 3). T h e fourth column. strength ratio (average strength after test average strength before test). shows the per cent strength retained by the fibers after exposure. T h e average of 0.93 for series I11 has been computed on the basis of the individual results. This value is not directly comparable with the respective values of series I and I1 for the follo\ving reasons: 1. T h e sampling method led to a closer approach to the 1 0 0 ~ life o of the fibers. (See Figure 2 for time of test terminations. Series I and I1 fibers were sampled closer to the 507, life-Le., approximately 1O4 seconds). 2. T h e tensile strength of control fibers was determined in 50% R H instead of 100% R H . As to item 2, a correction can be applied from available tensile test results in 50 and 1007, R H . I 1was found that the average tensile strength in 100% R H differed from 507, R H by a factor of 0.95, Application of this factor to the average
Table IV. Tensile Strength before Test AC. Jtrength, RH, % k.s.i.
Series
50
I
510
Strength Retention of Stress Corrosion Tested Fibers
Stress Corrosion Test
A p p i RH, % 50 100
50 50
I1 I11
50 100 100 100 100 100
480 505 473
50 50
47 5
50 50
486 488
stress, k.s.1.
320 320 320 320 280 240 200 150
ratio raises the value from 0.93 to 0.98. Hence, the average fiber strength is essentially unchanged by the exposures under load. ‘l’hc srcond method of presentation-i.e., failure probability plots-is shown in Figure 5 for the 100% R H fatigue data. Individual data points of the strength distribution of surviving fibers are related by different symbols to the applied stress lrvels a t which they were tested. T h e data points are equally distributed at both sides of the strength band for 100% R H . T h e low strength data could be interpreted as being obtained during the last phase of the life of the fibers when stress corrosion has progressed. Corresponding plots for the 50% R H data are shown in Figure 6. Conclusions. T h e static fatigue strengths are not linear functions of log time. A marked downward curvature occurs at lower applied stresses. Humidity has a noticeable effect on the static fatigue life of fibers; lower humidity extends the life, as might be expected. The initial tensile strength of fibers subjected to stress corrosion is retained over most of the life of the fibers, for both 50 and lOOY, R H .
Tensile Strength after StresJ Corrosion Test AL’. strength, R H , 72 Tested, % k.s.i.
Strenpth Ratio, after A\ro. of and before spec. Test 40 50 37.5 510 1 .oo 40 100 2.5 550 1.07 28 50 46.5 515 1.07 30 100 6.6 420 0.83 30 100 13.3 51 5 1 09 20 100 20.0 460 0.97 20 100 25.0 425 0.88 15 100 13.3 380 0.78 Av. of individual results in series I11 0.93 .4v. adjusted to 1007, RH of tensile strength before test 0 . 9 8
-
Among several investigators, Charles’ work has obtained wide recognition ; he first studied the corrosion process independently and then investigated the effect of stress by means of static fatigue tests. Charles has shown that the corrosion of soda-lime glass rods has a temperature dependence leading to an activation energy of 20 kcal. per mole? approximately equal to the activation energy for self-diffusion of sodium ions in the glass. This led Charles to postulate a corrosion mechanism involving terminal groups in the silica network: I
+ HzO
-+
I
A
L
f Na+
-SOH
+ OH-
I
-+
+ OH-
(2)
~
-SOH I
-t -SOI
(3)
(4) L
l
-1
Discussion
T h e results and conclusions presented so far will be examined in the light of the Charles current theory on stress corrosion (2). Charles’ work was based on soda-lime glass in the form of rods of 0.10-inch diameter, whereas this investigation is concerned with E-glass filaments 1/2ap as large. Besides the difference in glass composition (E-glass contains only trace amounts of the critical alkali), there is considerable difference in glass structure due to the different rates of cooling during drawing. T h e rod structure is close to equilibrium, and surface stresses during cool-down initiate surface cracks of detectable sizes ( 7 ) . T h e structure of filaments is “frozen” in an expanded, nonequilibrium state, and the surface is apparently free of detectable surface cracks. Soda-Lime Glass
__ E-Glass
(Corning ’Vo. 0080)
7c Si02 Na20 CaO
MgO
t
70
72 17 5
Si02 CaO
54.5
A1203
6
Bz03 MgO
14.5 8.5 4.5
17.0
1.o
Current Theory on Stress Corrosion of Glass. Stress corrosion of glass has been studied primarily for glass rods.
where Reactions 3 and 4 provide an autocatalytic mechanism for continued attack on the silica network. The first reaction must proceed to the degree necessary to create the requisite concentration of hydroxyl ions (or pH) because the second reaction can proceed only a t this requisite p H . This explains the observed incubation period in the plots of corrosion penetration us. time. Based on these reaction processes, Charles postulated a mechanism of stress corrosion tied to a stressaccelerated corrosion rate. He postulated that the corrosion rate in the x direction a t temperature 7’ is given by:
$~ ,’,
= ki(u)”
+k
(5)
where k l and k are constants, and u is the actual stress a t the root of a flaw. When the stress-activated corrosion is very much greater than ordinary corrosion, the root of an already existing flaw will be attacked and this will lead to an increase in the severity of the flaw because the stress concentration is a function of ratio (flaw depth/root radius). If the sides of the flaw are not attacked: but the depth increases, the stress at the tip of the flaw will increase. Application of this corrosion rate equation (Equation 5) to flaw growth led Charles to a relationship between the time to grow a flaw of critical size and the applied stress: - log k
log t = n log
(6)
UA
VOL. 5
NO.
1
MARCH
1966
5
where uA is the applied stress and k is a constant. This equation provides a convenient way to calculate the exponent, n, relating stress to corrosion rate. Discussion of Flaw Growth Theory. The flaw growth mechanism of Charles implies that flaws should become more severe with time under load in the static fatigue test. Table IV shows that the average tensile strength of fibers after exposure to 50 to 100% relative humidity has not decreased; exposure times were approximately 2 X lo4 seconds for series I and 11, and 1 to 3 X lo5 seconds for series 111. The respective failure strength distributions are shown in Figure 6. A few fibers in Figure 6 C have strengths less than the control fiber strength adjusted to 100% RH, but others have greater strength (Figure 5). Some additional tests conducted a t 80,000-p.s.i. applied stress showed the same result: Two out of five fibers tested were broken after 2.5 and 3.5 X 106 seconds and showed strengths of 470,000 and 480,000 p.s.i. us. 470,000p.s.i. average control fiber strength; the last fiber failed at 4.2 X 106 seconds. These unexpected results show that: 1. There is no continuous growth of existing flaws, as suggested by Charles’ theory, in times up to a t least 95% of the static fatigue life. 2. Both the average strength and strength distribution remain unchanged (Figure 6, A and B ) throughout most of the static fatigue exposure, showing that the flaws are unchanged. The lack of change in the low strength tail shows that the severe flaws are not the origin of static fatigue. The deviations observed in Figure 6 C are consistent with statement 1 because sampling took place in the last phase of the statistical life of the fibers. 3. A long incubation period precedes failure, with crack growth restricted to the period immediately preceding fracture. The foregoing results show that the present theory is not adequate because growth of flaws cannot be detected throughout most of the test duration. The deficiencies of the present theory may be related to the following : Inglis’ (7) criteria for large stress concentrations lead to the relation : ‘depth of critical flaw flaw root radius
(7)
tip (Figure 7). But if the actual mechanism does not require any corrosion until the incubation period is over, then an equation with stress raised to a high poLver represents an empirical fit of a two-stage process. Analysis of static fatigue data by the method of Charles shows that the required linear relationship between log (time to failure) and log (reciprocal applied stress) (Equation 6) is not obtained. The value of the exponent, n, varies from 20 at high applied stress to approximately 3 at low stresses, as shown in Figure 8. Data from another study (6) are included for comparison. These data were deriv-d from failure time plots, evaluated by the same method of analysis. The limits shown are those of 68y0of the data and correspond to the 1sigma limits in a Gaussian distribution. Proposed Mechanism of Stress Corrosion. I t is proposed that stress corrosion occurs where the so-called type C flaws are located at the surface. A description of the fla\vs is included in the Appendix. Interaction of water with the cations in the C fiaws leads to hydrolysis and an increase in concentration of hydroxyl ions (or of pH). The incubation period continues until the p H reaches the level necessary for corrosion to occur for 95y0(or more) of the total static fatigue process duration.
6-
5 0
111
4-
2-
d log T
r 32
d log
(L) ffA
2-
oo -5.6
-15 . 8
-5.4
4.0
-5.2
i o g l (psi-‘) VA
Figure 8. Static fatigue failure time vs. inverse applied stress for virgin RH E-glass fibers in 1 0 0 ~ o
where uf is fracture stress (taken to be 1,500,000 p.s.i.), and uA is the stress calculated from the applied load. For failure at 300,000-p.s.i. applied stress, the stress concentration must reach a value of 5. Assuming very local attack and a flaw root radius of atomic dimensions (2 A. assumed), a depth of corrosion of 12.5 A. is required. Such amounts of corrosion could occur in very short intervals once corrosion starts. Hence, corrosion cannot be occurring throughout the entire stress corrosion exposure. Calculation of the exponents for the stress-accelerated corrosion (Equation 5) has led to typical values for n between 16 and 26. Such high values give a n effective incubation period because the corrosion velocity, V,, is negligible until a flaw has been deepened or sharpened to raise the stress a t the flaw
,
,
,
I GROWTH OF F L A W DEPTH
vx 99b--”--i-
I
I
4
LOG T I M E , sec TIME
Figure 6
7. Growth of flaw depth with time
l&EC PRODUCT RESEARCH A N D DEVELOPMENT
Figure 9. Delay of corrosion in fatigue due to presence of water
static
To support this theory of the incubation, a comparison was made of static fatigue life a t 320,000 p.s.i of wet and dry fibers in 99+% relative humidity a t room temperature. O n e set of monofilaments was kept dry (adsorbed water films), and the second set was sprayed to leave water deposits on the surface. It was reasoned that the larger water reservoir in the second case would dilute the N a + and OH- ions, thereby delaying the time a t which the critical p H would be reached. Figure 9 provides preliminary evidence that this effect is occurring. although complete water coverage was not attained. Once the critical p H is reached, attack on the silica network will occur extremely rapidly by the autocatalytic mechanism postulated by Charles with resulting failure. Summary. Results from static fatigue tests have been examined against the background of Charles' theory of corrosion mechanism on glass. Disagreement was found in the stress us. failure time relation (Equation 6), which does not provide a single exponent required for the stress term in the flaw growth equation used by Charles. T h e equation is designed to govern the entire stress corrosion process from load application to failure-that is, incubation period and corrosion period. Based on these findings, a new mechanism of stress corrosion failure of glass is proposed T h e salient features of the proposed mechanism are: A prolonged incubation period, extending over approximately 9576 of the life of glass fibers. is controlled by hydroxyl ion buildup to a critical level of p H . Once the critical level is reached, chemical attack occurs rapidly (autocatalytically) until failure occurs. Chemical attack originates a t preferred sites on the fiber (structural defects or C-flaws) which may or may not coincide with the location of (much larger) surface flaws. T h e apparent rate of corrosion, if expressed as a power function of stress (Equation 5), is not determined by a single exponent of this function.
T h e damage is assumed to include the incubation and corrosion stages. so that a damage fraction of 0.1 might indicate only that the buildup of hydroxyl ions is a little over lOyo complete, but no actual corrosion had occurred. The biggest assumption in this derivation is that the event occurring during the process at each stress level is the same. But this assumption is not too critical because the major portion of the damage occurs a t the highest stresses prior to fracture. Table V shows a sample calculation for a strain rate of 0.06 min.? and a gage length of 1 inch. The stress corrosion data used are for 100% RH,so that the calculated tensile strength of 469,000 p.s.i. is very close to the tensile strength of 462,000 p.s.i. The latter figure has been obtained by applying a 5% correction to a representative tensile strength measured at 50% RH. Table VI presents typical strength data for both 0.06- and 0.006-min.-' strain rates, and the times of the actual tensile test as well as the calculated equivalent time.
Table V.
Cumulative Damage from Stress Corrosion in RH Tensile Test at 1 0 0 ~ o
(Sample calculation for standard strain rate 0.06 min.-' and gage length 1 inch) Loss f r o m Stress Corrosion Stress Range,
K.S.I. 0-100 100-200 200-300 300-400 400-420 420-440 440-460
460-470
Au. Stress Corrosion Failure T i m e , a ts, Sec. 106
3 x 106 5 X lo4 1 . 2 X lo3 5 X 10' 15 5
10 10 10 2
2 2
2.2
1
Total
Stress Corrosion in Tensile Tests
Stress corrosion must be occurring throughout the course of a tensile test. but because of the changing stress with time, it is not possible to correlate tensile results with the static fatigue data. Figure 10 illustrates the problem: What time under a load equal to the tensile strength would be required for static fatigue failure, and does this correspond to the summation of all the stress corrosion experienced under the continuously increasing loads of the tensile test? A simple assumption was made for the purpose of calculating the stress corrosion damage: that the damage fraction is equal to the fraction of the failure time for which the specimen was exposed under that load. I n other words, if the failure time under a nominal stress, S, is equal to tsn and a specimen is only exposed for a time, t,, the fractional life lost is
a
Tensile Test Time, t , Sec. 10
41
Total lost (cumulatioe)
Fraction life lost, t/ts
0.00001 0.00001 0.00004 0 00003 0.0002 0.00024 0.0083 0.00854 0.040 0.0485 0.133 0 1815 0,400 0.5815 469,000 p.s.i. at 1 . O 0.455 1 ,0365
From Figure 4.
Table VI.
Stress Corrosion Time Equivalence of Tensile Tests
(Gage length constant at 1 inch) T i m e , Sec. ( 700% R H ) EguiuaTensile lent test time* 44 2
Strain Rate, ~~Tensile Strength, K.S.I. .Min.-' 50% RH 700% RH Calcd." 469 0 06 496 (462) 42 9 425 15 0 006 465 (4401 a See Table V f o r calculation. T i m e under load ( i n static f a t i g u e ) equal to tensile strength in 100% RH atmosphere.
tn tS?l
Lt'hen
" 1
t
2
=
STRAIN R A T E T E S T S ( v s Total T e s t Time)
1.0. failure will occur
ts, TENSILE
sTREbsp---zEl
rEssiLE
____
STATIC F A T I G U E T E S T S ( vs T i m e U n d e r Load)
STATIC FATIGUE ~
TIMES
$
300
100% R H
L = 1 lnch
I I I
I
FAILURE TIME
EQUIVALENT TIME
Figure 10. Time relationship and static fatigue test
between tensile
2ooL
I
I
0
1
1 2
I
3
I 4
4 3
log T I M E
Figure 1 1 . Comparison of fatigue data
strain
rate and static
VOL. 5
NO. 1
M A R C H 1966
7
Data from variable strain rate tensile tests and from the static fatigue tests can now be compared by using these equivalent times. Figure 11 summarizes these results. T h e equivalent times bring the tensile data in conformity with the static fatigue data. Conclusions
Tensile strength measurements of E-glass fibers a t various times during static fatigue tests have shown that the initial strength did not deteriorate up to approximately 95% of the life of the fibers. Corrosion must, therefore, occur very rapidly in the last phase of the life. To explain this behavior, it is assumed that chemical processes during the prolonged incubation period must reach a critical stage before the corrosion process can GCCUT. T h e chemical processes are believed to occur on sites of structural flaws where concentration of terminal cations interrupts the silica network. Interaction of water with cations leads to hydrolysis and a n increase in concentration of hydroxyl ions-i.e., an increase in pH. Incubation continues until the p H reaches a critical level for corrosion to occur. Preliminary experiments have supported this view. Structural flaws (defined in earlier work as C-flaws) are distributed densely in the glass fibers. They may be located at the surface or below, and may coincide with much larger surface flaws (A- and B-type flaws). Corrosion originates a t C-flaws. Analysis of static fatigue data by the method of Charles shows that the required linear relationship between log (time to failure) and log (inverse applied stress) is not obtained. Some aspects of Charles‘ stress corrosion mechanism, based on this linear relationship, are subject to doubt. A mechanism of stress corrosion is proposed, based on separation of the two phases-incubation and corrosion. T h e corrosion phase occupies only a small fraction of the life of the fibers, approximately 5 to 10yG,and occurs rapidly after chemical processes during the incubation period have reached a critical phase. Stress corrosion in tensile tests becomes effective only a t stresses high enough to allow the total corrosion failure process to occur in the time available for fracture. Calculations on the basis of static fatigue data indicate that the process takes place in approximately the last 5% of the tensile test time.
Appendix.
Flaws on Glass Fibers
Prior work (8, 7 7 ) has determined the characteristics of flaws on glass fibers by tensile tests on monofilaments a t different gage lengths. T h e shapes of the distribution curves of strength data have been analyzed to determine the corresponding distribution of flaws. T h e variation of these distributions with gage length has been analyzed to determine a second characteristic of flaws-that is, the separation of flaws. I t has been concluded that three types of flaws exist on monofilaments, as revealed by tensile tests in air. For both E- and S-glasses, the flaw types are identical, but all strength values
8
l&EC PRODUCT RESEARCH A N D DEVELOPMENT
are higher for S-glass by 25y0. For E-glass the flaw characteristics are :
Flaw Type A. Severe surface flaws that arise from handling and processing, and control the strength primarily in the range below 480,000 p.s.i. These flaws have an average distance apart of 2 cm. Flaw Type B. Mild surface flaws that control strength in the range ?OO:OOO to 480,000 p.s.i. These flaws are approximately lo-* cm. apart. I t is likely from the flaw severity that these flaws are etch pits, but it is not known if these are present as drawn, grow by corrosion in the absence of stress, or grow by a stress corrosion mechanism during the tensile test. Flaw Type C. These flaws control strength above 700.000 p.s.i. and can be observed only when gage lengths of less than 0.1 cm. are tested. The proportion of failures controlled by type C flaws increases as the gage length is decreased to 0.025 cm. Extrapolation of this proportion suggests that a gage length of to cm. would have to be tested to obtain 50y‘ of fibers with strengths of 1,500,000 p.s.i. or above: indicating that this is the average separation of type C flaws. I t is believed that type C flaws are internal, structural defects arising from the heterogeneity of the glass, and thus represent concentrations of terminal cations that interrupt the silica network. These are not flaivs in the normal sense of voids, but generate a stress concentration because the weaker bonds at these regions lead to a local reduction of elastic modulus and hence to a stress concentration. These flaw characteristics have been determined by tests in standard laboratory atmosphere of 507,RH and at a standard strain rate of 0.06 min.-’ Since flaws are changing with time, humidity, and stress: a n apparent change of characteristics as revealed by failure strength data is to be expected under other test and environmental conditions. Analysis of published data has showm that these flaws are characteristic of both Eand S-glasses prepared and tested in several different laboratories. literature Cited
(1) Cameron, N. M., “Introduction to the Factors Influencing the Strength of Glass Fibers,” University of Illinois, T & A. M. Rept. 186, 1961. ( 2 ) Charles. R. J.. J . Abbl. Phvs. 29. 1549-60 119581 (3j Elliot, H.A , , Zbid.!pp. 224-5. (4) Glathart, J. L., Preston, F. IV..Ibid.: 1 7 , 189-95 (1946). (5) Griffith, .4.A , , “The Theory of Rupture,” Proceedings of International Congress on Applied Mechanics, 1924. (6) Hollineer. D. L.. Jordan: T. J.,Advanced EnPineerine and Technolzgy Dept.,’ General Electric Co., private comm;nication, 1965. (7) Inglis, C. E., Proc. Inst. S a a a i Archztects 5 5 , 19-30 (1913). (8) Metcalfe, A. G., Schmitz, G. K., “Effect of Length on the Strength of Glass Fibers,” A S T M , Proc. 64, 1075-93 (1964). (9)- Mould. R. E., Southwick, R. D.? J . A m . Ceram. Soc. 42, No. 12: 382-92 (1959). (10) Murgatroyd. J. B., J . SOC.Glass Technol. 28, N o . 130, 406-31T (1944) ; Ceram. Abstr. 1946, p. 157. (1 1) Schmitz, G. K., Metcalfe, A. G., “Characterization of Flaws on Glass Fibers,” SPI Reinforced Plastics Division. Proceedings, 1965. (12) Stuart, D. A , , Anderson, 0. L., J . A m . C u a m . SOC.36, No. 12, 41 6-24 (1953). (13) Taylor, N. IV., J . AppZ. Phys. 18, 943-35 (1947); Ceram. Abstr. 1948, p. 191g. ~~~
.
~~~
- I
\
I
I
RECEIVED for review June 7, 1965 ACCEPTED November 26, 1965 Work performed under Navy Contract Nom 3654(00) (X)A2. Society of Plastics Industry meeting, Chicago, Ill., February 1965.
