Search Documents

By J. F. Breedis

The morphology and crystallography of marten -site formed during quenclzing were examined by transmission electron microscopy in alloys whose compositions lie between Fe-19 wt pct Cr-11 wt pet Ni and Fe-33 wt pet Ni. Depending upon composition, four types of transformation products were observed in this portion of the Fe-Cr-Ni system. Three of these products are denoted as lath, plate, and surface martensite. The hcp structure found in association with martensite it? certain higlz-clzromium stairzless steels does not influence the transformation crystallogyaphy and cannot he Considered to act as a transitional structure in the transformation from austenite to martensite. DISTINCT changes in the characteristics of mar-tensitic transformations are produced by the partial replacement of nickel by chromium in austenitic iron alloys. Large, discrete martensite plates form in Fe-30 pct Ni, while aggregates of small crystals form in sheets parallel with (111)A austenite planes in stainless steels containing in the vicinity of 18 pct Cr and 8 pct Ni. These transformation structures are termed plate and lath martensite, respectively. In addition, an hcp structure (E) has been observed in certain high-chromium stainless steels in association with martensite. It has been suggested that the E structure acts as an intermediate step in the nucleation of martensitel-' and influences the morphology and crystallography of the final transformation product. However, this structure may be the consequence of the deformation of austenite arising from the shape distortion accompanying the transformation to bcc martensite.' An hcp stacking of atoms is produced in an fcc lattice by the introduction of an intrinsic stacking fault on alternate (111)A planes. At present, the arguments behind each explanation for the E structure are based on circumstantial evidence. This investigation seeks to determine the importance of the hcp structure to the martensitic transformation, and to examine the crystallography and morphology of the structures produced in bulk samples by quenching in alloys whose compositions lie between Fe-19 wt pct Cr-11 wt pct Ni and Fe-33 wt pct Ni. These aspects of the martensitic transformations occurring in these alloys have been studied by electron metallography and selected-area diffraction of foils examined in transmission in the electron microscope. EXPERIMENTAL METHODS The alloys have been prepared from thoroughly mixed powders of pure iron (99.6+ pct, 0.06 max pct C, 0.15 max pct N), chromium (99.9+ pct), and nickel (99.9+ pct). After sintering in an Ar-H2 atmosphere, the mixtures were melted by induction heating in recrystallized alumina boats under purified argon. The ingots were cold-rolled to 0.4-mm sheets (100 pct reduction in thickness) after the removal of at least 1.5 mm from their surfaces by grinding. Three-millimeter discs were punched from the cold-rolled sheet and annealed at 1150°C for approximately 24 hr in evacuated quartz capsules. The average grain size was 40 µ Thin foils for examination in transmission in the electron microscope were prepared from discs quenched to liquid-nitrogen temperature. A depression was first jet-machined into the center of a disc and a thin area (less than approximately 3000A in thickness) was produced at the base of the depression by electropolishing. Electron-microscope studies utilized the RCA EMU-3 and the Philips EM-100B microscopes operating at 100 kv. X-ray examination of the polycrystalline alloys was performed either in a diffractometer or in a

Jan 1, 1964

By T. Saburi, C. M. Wayman

The massive and martensitic transformations in ß Cu-Ga alloys were studied by optical microscopy and by transmission electron microscopy and diffraction. These types of transformation are distinct and do not merge into one another. An internal fine structure due to stacking faults was found in both the massive phase (m) and the martensite, but the origin of the faults is believed to be different in each case. The faults in the martensite are due to the lattice invariant or "second" shear of the phenom-enological theory, while those in the are probably growth faults and deformation faults arising from the transformation volume change. Evidence is presented for ordering in the ß phase prior to transformation into martensite, and the crystal structures of and ß' and ßi martensites are discussed. THE results from some previous investigations1,2 have led to the suggestion that martensitic and massive transformations have certain common characteristics. It presently seems established that massive transformations like martensitic transformations are essentially diffusionless,1,2 but unlike martensitic transformations, when a new phase forms massively there is no shape change or surface relief.= If the massive-type transformation is crystallographically similar to martensitic transformations, one would expect a fine structure as a result of the lattice invariant shear4 which is a part of the transformation mechanism. The existence of such a fine structure (i.e., slip, twinning, or faulting), if any, can be verified by transmission electron microscopy, and thus the present investigation was undertaken to compare the internal structure of massively and martensitically formed phases in Cu-Ga alloys which were derived from a parent of the same composition. The transformation characteristics of ß Cu-Ga alloys were also studied. The bcc ß phase in the Cu-Ga system, Fig. 1, is stable above 616oc.' Depending upon the cooling rate, ß Cu-Ga alloys can undergo three different types of phase transformations. The ß phase of eutectoid composition (23.7 at. pct Ga) decomposes into the ? phase (hcp) and the phase (complex cubic) by means of a eutectoid reaction. When the cooling rate is increased so that the decomposition of 0 into [ and ? is suppressed, the ß phase transforms into a nonequilibrium hcp phase, herein denoted ?m by a process of very rapid growth called a "massive" transformation. At even faster quenching rates the massive transformation can be suppressed and the ß transforms martensitically. Massive transformations, apparently similar to those in the Cu-Ga system, have also been studied in the Cu-Zn2,6-8 and Ag-Cd systems.9 Definitive characteristics if the Massive Transformations in 0 Cu-Ga Alloys1 1) The most remarkable feature of the massive transformation in Cu-Ga is that there is no bulk composition change on transformation. According to Massalski,1 the resulting [, phase is a supersaturated extension of the equilibrium [ phase. The lattice parameters continue the trend exhibited

Jan 1, 1965

By I. A. Lesk

The various materials utilized in the construction of integrated circuits, and the resultant materials interfaces, are discussed with emphasis on a materials system that is compatible with all types of integrated circuits and minimizes the number of different interfaces. Problems with interfaces may originate in the silicon, on its surface, and between other materials within the integrated-circuit package independent of the silicon. ThE difference between integrated-circuit theory and practice may generally be spelled out in terms of the difficulties in fabricating a desired structure. There are, of course, problems in forming the requisite materials, of the necessary composition and geometry, where needed. A distinct class of problem, however, is associated with interfaces— generally between different materials, but occasionally between regions of the same material. We require that each interfacial system perform its proper electrical, mechanical, and thermal function, and do so in an unchanging manner with time under the most severe operating conditions. In a practical sense, we deal in nonideal systems, such that it is generally not possible to form an interface exactly the way it was intended, even if we allow for a knowledge of the practical limitations of the fabrication system. Another way of looking at this situation is to realize that, when we form a group of interfaces under apparently the same experimental conditions, there are often differences of important consequence between them. Very high-magnification pictures, such as are obtained by an electron microscope, reveal fine structure that often differs over the area of a given sample, as well as from sample to sample. Low-magnification pictures look better, but don't eliminate the problem. The trend towards integrated circuits, and away from circuits fabricated utilizing separate components, is very strong in many areas of electronics. A typical silicon monolithic integrated circuit, still in place in the wafer on which it was fabricated, is shown in Fig. 1. The increased reliability which is being experienced through the use of integrated circuits can, to a large extent, be credited to the reduced complexity of the interface system that is present in the electronic system. A glance at Fig. 1, however, with all of its epitaxial and diffused p-n junctions, and interconnect system, shows that interfaces have been far from eliminated. Fig. 2, depicting a schematic cross-sectional view of a (partial) typical integrated circuit, illustrates in detail the numerous interfaces that occur in this type of structure. Besides the individual circuit elements, there are interfaces for isolation, and, Fig. 3, even a crossover requires a series of interfaces. What has been accomplished, however, irrespective of whether or not the total number of interfaces in the circuit is increased or decreased, is that the variety of interfaces is very much reduced. This is because there are fewer materials involved. Since the expense incurred in an R & D program to improve reliability is pretty much directly proportional to the variety of interfaces involved, it is economically advisable to work with integrated circuits for highest reliability. That is not to say that circuits built using standard components could not be made as reliable as integrated circuits, but that the expense involved because of the more complex over-all interface system would be much greater. Although electronic performance utilizing integrated circuits with all silicon elements (i.e., transistors, diodes, resistors, capacitors) is adequate for many applications, there are numerous instances where thin-film passive elements must

Jan 1, 1965

By C. B. Smith, A. G. Guy

The power and convenience of tensor analysis are employed in deriving the equations that describe three-dimensional diffusion in an n-component system plus vacancies. A "Kirkendall coordinate system, '' defined in principle by Kirkendall markers, is used in defining the intrinsic diffusion coefficients that are of primary interest. The defining equations are then transformed to local cartesian coordinates and to fixed cartesian coordinates, the latter being convenient for experimental measurements. Finally, it is shown how the intrinsic diffusion coefficients can be calculated from the experimentally determined quantities. It is generally recognized that any treatment of the essential diffusion phenomena in metallic systems—the jumping of atoms from one lattice position to another—must be based on movement of the atoms past Kirkendall interfaces.' A logical extension of this idea is the proposal that, fundamentally, diffusion must be described mathematically in terms of "Kirkendall coordinates." The basic equations thus obtained may then be transformed into other coordinate systems, when such transformation leads to simplification of either the experimental or mathematical treatment of diffusion data. Tensor analysis is clearly the appropriate mathematical tool for this purpose, since it provides methods both for determining general equations that are valid for all coordinate systems and for conveniently transforming the equations from one coordinate system to another system related to the first in some known fashion. For the present purpose three coordinate systems are used, and it is convenient to begin with a description of the relations among these three systems. FICK LAW FOR KIRKENDALL COORDINATES The reference coordinate system, and the one that is useful for most experimental measurements, is a cartesian coordinate system, xi (i = 1, 2, 3), fixed with respect to some point in the specimen, such as an end of the alloy bar undergoing diffusion. A second set of coordinates, the Kirkendall coordinate system, is one that in principle can be described by the location of inert markers (Kirkendall interfaces). A convenient mathematical relation between Kirkendall coordinates, x'i, and the reference cartesian coordinates, ii, is obtained by specifying that the Kirkendall coordinate system coincide with the cartesian system at time, t = 0. After diffusion has occurred for some time, t = t, the Kirkendall coordinate system will differ from the reference system in the manner shown schematically for the one-dimensional case in Fig. 1. The third coordinate system is Darken's1 local cartesian-coordinate system, xi, which is identical with ii in orientation and units, but which moves with the point under consideration in the Kirkendall coordinate system. Even though most diffusion problems are essentially one-dimensional, there is an advantage in first carrying out the analysis of the three-dimensional problem to gain a clearer insight into the assumptions that are made in the usual one-dimen-

Jan 1, 1962

By H. Reiss

The zone-melting process in which redistribution of solute in a solid bar is effected by the passage of a molten zone is considered mathematically. Simple approximate techniques are developed for computing the manner in which redistribution occurs as molten zones continue to pass. The introduction of the "zone-melting flux" provides valuable insight into the nature of the phenomenon, as well as a central mathematical theme in terms of which the process can be discussed. Curves are presented for typical and general cases. AN extremely efficient purification technique was described recently by Pfann.1 Known as zone refining, it consisted of the slow passage of a series of molten zones, produced by external heaters, through a long solid ingot of the substance to be purified. Equations were given for the maximum separation attainable, and a numerical computation method was described for obtaining the concentration in the ingot after a given number of zone passes. The present paper offers an interpretation of the physical nature of zone melting in terms of a transport process in which formal diffusive and convec-tive flows occur. This interpretation, in addition to providing insight into the physical nature of the process, also provides a basis for its mathematical treatment. From this basis, useful equations are developed for the solute concentration in the ingot as a function of the number of zone passes. These are applicable particularly where large numbers of passes are necessary and where numerical computational methods might be laborious. All considerations will deal with the limiting form of the process, described by Pfann,' in which: 1—Diffusion in the solid is considered negligible. 2—Mixing in the liquid is supposed, complete. 3—The distribution coefficient, k, measuring the ratio (at equilibrium) of solute concentration in the solid to that in the liquid is supposed, independent of composition. In general, this paper will be concerned with a bar of uniform (actually unit) cross section, ex- tending in the x-direction from x = 0 to x = L, where L may be infinite. Zone melting is a transport process, and as such must admit of interpretation in terms of a flux of transported quantity. However, the progress of operation is not measured in units of time but in terms of the number of completed zone passes. Consequently, the zone-melting flux will be a rate, per unit pass, rather than per unit time. The description of this flux and the examination of its properties is given below. Zone-Melting Flux Fig. 1 represents an intermediate stage of the nth pass. Concentration C is plotted against distance, x. The molten zone with its back at the point x and front at x + I moves in the direction indicated by the arrow. As it moves, it leaves behind the nth

Jan 1, 1955

By Anton Sawatzky, Erich Vogt

By means of mathematical solutions to the appropriate diffusion equations, we describe the kinetics of the thermal diffusion of hydrogen in Zircaloy-2 for the various temperatures and concentrations encountered in a heavy water moderated reactor. When the hydrogen concentration is below terminal solid solubility only the a Phase is present. Redistributions are then described in terms of the characteristic functions of the difhsion equation. For higher concentrations both the a and 6 phases are present. We assume the two phases to be always in equilibrium. For moderately small hydrogen concentrations exact solutions of the two-phase equation approach the the approximate solutions derived by Sawatzky and for all concentrations the exact solutions exhibit the qualitative features of his result: the two-phase concentration increases with time, everywhere; in the absence of a hydrogen current at the hot end of the sample an a-phase region always exists there; the interface of the a + 6 , a-phase boundary moves toward the cold md of the sample and the hydrogen concentration is discontinuous at the interface. Simultaneous solulions of the a and a + 6 hydrogen distributims and of the concomitant interface motion are obtained and compared to the observations of Sawatzky and Markowitz. The kinetics of the hydrogen diffusion process are shown to lead to m apparent heat of transport of the a phase which is lower than the actual value (even for samples with long anneals) thus resolving at least partially the disparity between experimental measurements of this quantity. A number of recent papers1"4 have reported measurements on the diffusion and redistribution of hydrogen in Zircaloy-2 under temperature and concentration gradients. These studies were instigated by problems arising from increasing use of Zircaloy-2 as a fuel element cladding material in pressurized-water power-reactors. The Zircaloy-2 picks up hydrogen during the operation of the reactor: the consequent precipitation of zirconium hydride in the Zircaloy-2 has pronounced effects on its mechanical properties. The Purpose of the present Paper is to describe the kinetics of the thermal diffusion of hydrogen in Zircaloy-2 for the various hydrogen concentrations and temperatures likely to be encountered in reactors. When the hydrogen in the Zircaloy-2 is entirely in the solid solution phase (a phase), the differential equation for thermal diffusion is well known and the redistribution can be described by standard mathematical methods some of which are given in Section 11 below. The treatment of the a + 6 (hydride) region differs from the earlier treatment of sawatzky3 by taking into account several modifications of the two-phase diffusion equations as suggested to us by Markowitz4 and Kidson.5 Even with the modifications the results obtained in our more complete treatment are substantially the same as those found by Sawatzky. As we show, the difference between the results of sawatzky3 and kIarkowitz4 is largely due to the difference in the geometry of their experiments. In the next section we derive the modified two-phase equation for arbitrary geometry. The exact solution of this equation is given in Section III for linear and cylindrical geometry. It is shown that Sawatzky's approximate solution is quite accurate for almost all the temperatures and hydrogen concentrations which are actually encountered. In Sawatzky's approximate theory the hydrogen concentration everywhere in the two-phase region increases continuously. As his paper pointed out, this result, together with hydrogen conservation, implies that in a sample with no hydrogen current flowing into the hot end a two-phase region is always accompanied by a single-phase region at the hot end of the sample. The net hydrogen gain in the two-phase region is supplied by the decrease of hydrogen in the single-phase region and by the movement of the (a, a + 6) boundary toward the cold end of the sample. Hydrogen conservation at the (a, a + 6) boundary leads to a discontinuity in the concentration and its derivative there. In Section IV it is shown how these qualitative features of the thermal diffusion kinetics arise from the simultaneous solution of the a-phase differential equation and a + 6 phase equation. Methods derived by us previously for the solution of the redistribution in both phases and the accompanying motion of the boundary are used to obtain approximate solutions valid for large times. The solutions account well for the observed redistributions of Sawatzky and Markowitz. It is shown how the continuing motion of the (a, a + 6) boundary leads to measured values of the heat of transport lower than the actual value, even for specimens with relatively long anneals.

Jan 1, 1963

By N. J. Grant, Hans Brunner

ThE materials used in this investigation were furnished by the Aluminum Co. of America and consisted of high-purity alurhinum (99.995 pct) and two aluminum-magnesium alloys containing approximately 2 and 5 pct Mg. The compositions of these alloys are listed in Table I. The materials were received as hot rolled 3/8-in. diam rods and were made from the same heats as the materials which were previously used in this laboratory.'j3 The rods were cold swaged to 1/4-in. diam by 1 in. long, and then had two parallel flats machined over the gage length 0.085 in. apart. The test bars were annealed for 10 min at 700° F, and strained 1 pct in tension at room temperature. They were electropolished (80 pct acetic acid; 20 pct perchloric acid, 70 pct concentration) at a temperature of 60°F at approximately 20 v. A grid, transverse or square, spaced 0.2 mm apart, was impressed by rolling a fine threaded screw (128 threads per in.) across the flat surfaces. The depth of the grooves was less than 5 p. For final annealing and grain size control the specimens were enclosed in aluminum foil and sealed in Vycor tubes. Annealing conditions varied from 1 hr at 700° F to 5 days at 1050° F, depending on the alloy content, to give for most of the specimens a final grain size of about 1 mm. This treatment was followed by electropol-ishing as described above. For the test series presented in Figs. 4 and 5 round specimens were used instead of flat ones, and fine machining marks served as marker lines instead of the impressed grid lines. Electropolishing was omitted. The major advantage is that the spacing of the machining marks is only about 1 u as compared to the 200 u of the impressed grid lines, permitting higher precision in strain measurements. Unfortunately the strain annealing necessary for grain size control also resulted in some preferred orientation of the grains. The preferred orientation of three arbitrary specimens was measured by X-ray diffraction methods and the averaged results are shown in Fig. 1. The possible effects of the preferred orientation on the test results will be discussed later. The testing temperatures were 410°, 515", 725", and 940° F, with one exception in the case of the 5.1 pct Mg alloy. Since the solubility limit of this alloy is 520°F, a test temperature of 545°F was cho-se sen instead of 515°F. The selection of the test temperature was based on two considerations: 1) at 900°F and above, extensive grain boundary migration made it more difficult to measure grain boundary sliding; 2) at 400°F and below, there was so little grain boundary sliding that it was again difficult to make precise measurements. The stress was chosen to give a creep rate of approximately 2.7 pct per hr in connection with Figs. 2

Jan 1, 1961

By Charles A. Stickels, Edward E. Hucke

The dihedral angle, formed by the intersection of three interfaces, is often determined by measuring a sample of apparent angles of intersection seen on a random plane of polish, then comparing the exerimental distribution of angles to the distribution calculated by Harker and Parker. Statistical tests show that the latter distribution does not satisfactorily describe the population from which samples are drawn. It is concluded that the major reason for this discrepancy is the presence of a range of dihedral angles in all specimens. When reasonable care is used experimentally, random measuring error is unimportant. It is shown that measurements made on dispersed second-phase inclusions are subject to a systematic error depending on inclusion size, which can be minimized by using adequate magnification. A dihedral angle, calculated from a sample of observed angles as though a unique dihedral angle were present, is termed an "effective dihedral angle". Experience has shown that "effective dihedral angles'' are reproducible quantities, useful for characterizing samples. Methods of determining an "effective dihedral angle", establishing its reliability, and making comparisons between samples are discussed. THE authors have recently reported measurements of the dihedral angle formed by liquid-lead inclusions lying in nickel grain boundaries.' Di- hedral angles were measured by the statistical method first employed by Harker and parker2 and used subsequently by many other workers.3"8 Since a relatively high degree of accuracy was necessary in the authors' experiments, the reliability of this method of dihedral-angle measurement was examined. Briefly, this method of measuring dihedral angles is usually performed as follows. A multiphase alloy is prepared, equilibrated under conditions which allow sufficient interface mobility to establish equilibrium, quenched to room temperature, then prepared for metallographic examination. A large number of observed angles, formed by the traces of grain boundaries and/or interphase boundaries on the plane of polish, are measured at high magnification. If all orientations of interfaces in the specimen are equally probable, Harker and parker2 showed that the population of observed angles is described by a distribution function containing a single parameter, the dihedral angle, 8. certain properties of the population, e.g., the mode or median, have been calculated from this distribution function in terms of the dihedral angle.3, 9 Thus, if the mode or median of a sample is identified with the corresponding property of the population, a unique value for the dihedral angle is determined. One defect in this procedure lies in identifying the properties of a sample of observed angles with the properties of the population from which they are drawn. If one were to proceed in a more rigorous fashion, it would be necessary to find a function of sample properties which is an estimator of the dihedral angle. Then by determining the distribution function for the estimator one could set limits on the value of the dihedral angle. The most apparent obstacle to this straightforward approach is the complexity of the distribution function assumed to describe the population. In analytical form, the distribution function may be written:2, 10

Jan 1, 1964

By R. W. Cahn, C. D. Graham

The rate of growth of a single grain growing into a strained aluminum single crystal, measured by the conventional heat-cool-etch technique, is shown to decrease with time at temperature. The growth rate measured by means of X-rays, with the sample continuously at temperature, is found to be linear. Evidence is presented to show that the decreasing growth rate obtained by the conventional technique is due to abnormal recovery in the strained grain, and that the recovery in turn results from the heating and cooling cycles, rather than from the effects of etching. The experimental apparatus is described in an appendix. FREQUENTLY, the rate of grain boundary motion during recrystallization is determined by heating a sample to a known temperature for a given time, cooling to room temperature, and etching to reveal the boundary position. The cycle of operations is repeated as many times as necessary to establish the rate of boundary motion. Although this heat-cool-etch technique has been used by a number of workers over a period of many years,'-!' other investigators have encountered difficulty with the method."'. " They have attributed the erratic results to the effects of repeated heating and cooling, or of etching, or both, and have adopted an alternative experimental procedure in which a number of identical samples are employed, each sample being cooled and etched only once. This alternative method ove1.comes some difficulties, but introduces others. In particular, the incubation periods of the observed grains are unknown: any effect of crystal orientation on growth rate cannot be determined easily. A preferable method would be to measure the motion of a recrystallizing grain boundary with the sample at temperature without etching," but such measurements have in the past been restricted to nonmetals. This paper describes the construction and operation of a device, the goniometer furnace, with which the position of a moving grain boundary between a recrystallizing grain and the strained single crystal into which it grows can be measured by means of X-rays, with the sample maintained continuously at temperature. Goniometer Furnace The operating principle of the goniometer furnace is shown in Fig. 1. The sample, in the form of a bicrystal strip, is surrounded by a furnace, not shown in the sketch, and can be rotated about the two axes A and B which intersect at the point where the X-ray beam strikes the strip. The strip is oriented so that one of the two grains gives a strong Bragg reflection, which is detected by a suitably placed Geiger-Miiller counter coupled to a counting-rate meter When the strip is moved parallel to its own length in the furnace, arrow C, so that the X-ray beam strikes the second grain, there is in general no strong reflection. By noting the position of the strip at which the drop in Geiger counter reading occurs, it is possible to determine the position of the boundary. Measuring the change in boundary position with time at temperature gives the rate of growth of the recrystallizing grain. Details of the construction and operation of the goniometer furnace are given in the Appendix. Measurements showed that the temperature in the furnace was uniform within ±10°C over its working length of about 28 cm. The temperature at the measuring position, i.e., at the point where the X-ray beam struck the sample, was estimated to be constant within ±2oC. Using the goniometer furnace, the position of a stationary boundary could be measured with a reproducibility of ±0.3 mm. At elevated temperatures, 500" to 625"C, with the boundary in motion and the X-ray reflection less sharp, the estimated accuracy of each measured boundary position dropped to about ±1 mm. Experimental Results The test pieces were single crystals of 99.6 pet A1 —principal impurities 0.19 pet Fe and 0.12 pet Si— 1 mm by 1 cm in cross section and 6 to 10 cm in length, grown by the strain-anneal method. This relatively low purity material was used because of the difficulty encountered in producing strain-anneal single crystals of very high purity—99.996 pet—

Jan 1, 1957

By R. L. Fullman

IN the investigation of metallurgical transformations and the relationships between microstructure and properties of metals, it frequently is desirable to obtain a measurement of the relative amounts of the various phases present and of the mean size of particles into which each phase is dispersed. The relative amounts of the phases can be measured by the classical methods of area, lineal, and point analysis,1-5 in accordance with the principle that the volume fraction of a phase, the fraction of a polished cross section occupied by the phase, the fraction of a random line occupied by the phase, and the fraction of randomly arrayed points occupied by the phase are all equal. The validity of this relationship depends only on the attainment of a truly random sample of area, length, or points, and not on the size, shape, or distribution of the particles constituting the phase. Smith and Guttman8 have derived a relationship between the interface area per unit volume S, and the measurable quantities L., the interface length per unit area on a cross section, and NL, the number of interfaces per unit length intersected by a random line. Their equation, Sv = — L8 = 2NL is also valid regardless of the distribution of particle sizes and shapes. In contrast to the situation concerning measurement of relative fractions of phases and of interface area, the measurement of particle sizes in opaque samples has not been subjected to a complete analysis. It has been common to measure some lineal or area dimension of particles on a polished cross section and to use the mean value as a qualitative measure of particle size. In the present paper, quantitative relationships are established among the various mean dimensions on a polished cross section and the actual dimensions of the particles present. Particles of Uniform Size Spheres: If a metal sample contains particles of a phase a dispersed in the form of spheres of uniform size, a polished cross section through the sample will reveal circular areas of phase a with radii from 0 to ?, the radius of the spheres. Consider a cube of unit dimensions to be cut from the sample. If a cross section parallel to one of the cube faces is examined, the average number of particles per unit area (N,) equals the number of particles per unit volume (Nv) times the probability p1 that the plane would intersect a single sphere positioned at random within the unit cube. Since, of the various possible positions for the cross-sectional plane over the unit length from top to bottom of the cube, only those positions existing over the length 2r would lead to the plane intersecting the sphere, the probability of intersecting a single sphere is just 2r. N8= Nvp1 = Nd-2r [1] Applying the equality of area and volume fractions, the relationship is found between sphere size and average area s of uniform spheres intersected by a random cross section, 4 - f = NV V = Nr . — pra = N s = Nd . 2rs S = —pr2 [2] A similar analysis reveals the average traverse length across spheres of uniform size when random lines are passed through the sample. If a randomly oriented unit cube is cut from the sample and a randomly positioned line is passed through the cube parallel to a cube edge, the number of spheres intersected by the line (Nl) equals the number of spheres per unit volume times the probability p1 of the line hitting a single randomly placed sphere in the cube. Since possible positions of the line occupy unit area, and possible positions for which it will pass through the sphere occupy an area of pr2, the probability of the line hitting a randomly placed single sphere is pr2. NL = Nv p1 = Nvpr2 [3] Combining this relationship with the equality of volume and lineal fraction, the desired relationship is obtained between radius and mean lineal traverse length -i, for spheres of uniform size. 4 - - 3 l=4/3r [4] Circular Plates: Consider a sample containing particles of a phase a in the form of circular plates of uniform radius r and thickness t, where r >> t. If the plates are randomly oriented, as in a sufficiently large sample of a fine grained polycrystalline material, area and lineal analysis may be carried out with parallel cross-sectional planes and lineal traverses. If the plates are not randomly oriented, it is necessary to randomize the orientation of the cross-sectional planes and traverse directions. Let a unit cube be cut from the sample, and a cross-section plane be passed through the cube parallel to one of the cube faces. The number of plates cut by the cross-sectional plane per unit area is equal to the number of plates per unit volume times the probability of a plate intersecting a single randomly positioned and randomly oriented plate in the cube. If J is the component of the plate diameter in the direction normal to the cross-sectional planes, the probability of a plane cutting a single randomly oriented plate is equal to J, the mean value of J for all possible orientations of the plate. Let 4 be the dihedral angle between a plate and the cross-sectional plane, and let p?, d? be the probability that a plate makes an angle between 4 and ? + d? with the cross-sectional plane. Then for ran-

Jan 1, 1954

By C. G. Dunn, F. W. Daniels, M. J. Bolton

IN recent papers on interface energies in metals¹,² the concept of an equilibrium of forces has been used in the measurement of interfacial tensions. Mathematically the equilibrium of three forces is given by the relations: where the 2's are the interfacial tensions (interface energies per unit area of interface) of three phases or grains meeting at angles 21 22 and 23 (212 relates to the 1-2 boundary formed by grains 1 and 2;223 to the 2-3 boundary, etc.). If these relationships are valid, one can calculate relative values of the interface energies from measured boundary angles on specimens which have been brought to equilibrium. Since Eq 1 may be derived through a mathematical process of minimizing the grain boundary energy under the restriction that the ?'s be independent of boundary orientation: the concepts of surface tensions and equilibrium of forces are not particularly essential. To deal with the general case involving dependence of boundary energy on boundary orientation, it becomes desirable to drop certain concepts formerly held (namely that energy per unit area is identical with force per unit length or surface tension) and to use the energy concept alone. It is appropriate, therefore, to use E in place of ? in deriving and using the more general equations that express the anisotropic nature of the grain boundary energy. Herring3# ' recently has derived the required general equations using the principle of minimization of energy. Specifying the boundary orientations in terms of angles (4) measured with reference to crystal axes within the grains, Herring's equations take the form E23 + E21 cos?8 + E12 cos?2 — sin ?12 + sin?2-?E12/?12 =O [2] where the partial derivatives are measured in the direction of counterclockwise rotation of the boundaries. Eq 2 implies that an infinitesimal displacement of the triple point along the 2-3 boundary produces no change in the boundary energy. Similar equations apply for infinitesimal displacements along each of the other two boundaries. From such equations Herring obtains the following general relationship (1 + E1 E2) sin?8 + (E2 — E1) cos?3 ________ E38______________ = (1+ E2 E8) sin ?1+ (E3— E2) cos ?1 __________^5____________[3] (1 + E sin ?2, + (E1 — E8,) COS ?2 1 E28 where E1 = 1/E23- E25 and PC and c8 have similar expressions. It is apparent that Eq 3 reduces to Eq 1 when the c's vanish. Eq 3 also reduces to Eq 1 when the E'S are all equal. Shockley and Read' have pointed out one limitation to these equations. When E plotted versus 4 appears as an energy cusp, as their theory requires, for example, for the energy near a twin boundary position, then Eq 3 (and therefore Eq 1 also) no longer applies if one boundary is in the minimum energy position. One of the equations of form Eq 2, however, applies if the infinitesimal displacement is along the boundary associated with the energy cusp. From the foregoing treatment it will be apparent that equilibrium angles alone are not sufficient in general for calculating relative interface energies.

Jan 1, 1951

By C. G. Dunn, F. W. Daniels, M. J. Bolton

J. P. Nielsen—The data that Dr. Dunn and his associates have been obtaining are welcome checks on the theoretical aspects of grain boundary energies. With reference to the comments on the validity of the inter-facial tension concept, I should like to say that I have been having considerable difficulty myself in untangling the confusion between "interfacial tension" and "interfacial energy." The confusion however resolved itself for me when I learned that "interfacial tension" for solids, or at least crystals, is used with two distinct meanings. One meaning12. ' "hat is implied for "interfacial tension" is the work necessary to produce or increase the surface of the solid in question by a unit area, temperature and pressure held constant. This comes from the somewhat surprising but simple fact that surface tension and surface energy in the case of liquids are identical quantities. This synonymity has carried over to crystals, where it should be kept in mind that the tension connotation has meaning only by analogy with liquids and that no real stresses are being referred to. The only reason that there is apparently a set of stresses acting at a grain boundary junction is the optical illusion that liquid interfaces are being observed for the grain boundaries. The other meaning3,14,15 for the term has to do with actual states of stress in the surface of a material as would be measured by a system of forces applied to the surface to equate the surface stress state with that in the body of the crystal. J. B. Hess—Since the authors have proposed that the surface tension concept should be discarded in order to deal with the general case of interface energies in solids, their attention should be called to a recent paper by Gurney." The latter has shown, from a thermodynamic argument, that surface tension is a valid concept in solids. In fact, Gurney has rejected the idea that surface tensions are merely mathematical fiction and, instead, has given an interpretation of their physical reality. Finally, he has concluded that, as long as appreciable atomic migration takes place, specific surface energies and surface tensions must be equivalent in solids as well as in liquids. Accordingly, the authors' eqs 2 and 3 can be written with equal accuracy in terms of interface tensions, simply by replacing each E,! with its numerically and dimension-ally equivalent Interfacial tension ?,,. If the interfacial tensions (or specific energies) are independent of interfacial orientation, as is true in liquids, then the general equations that define the geometry at equilibrium of the common junction of three interfaces reduce to eq 1. However, neglect of the effect of boundary orientation is never completely justified theoretically' in dealing with crystalline solids; hence, use of eq 1 in such cases must always be considered as an approximation, though clearly a reasonable one for many purposes.1,2 At the same time, one should not be surprised that the assumption that interface tensions in solids are free from orientation

Jan 1, 1951

By G. A. Alers, F. Borik

Rolled sheet products are elastically anisotropic and a large number of measurements of Young's modulus and rigidity modulus are necessary to completely specify their elastic properties. This paper describes and uses a simple measurement technique to obtain the nine elements of the elastic modulus tensor from which all the elastic The elastic properties of a polycrystalline material are usually specified by giving the Young's modulus E and the rigidity modulus G. Unfortunately, these two moduli by themselves are insufficient to completely specify the elastic properties because the rolling process introduces a preferred orientation which makes the material properties can be deduced. In particular, the variation of Young's modulus and the rigidity modulus in the principal planes of a sheet of ausformed steel were deduced. Furthermore, the variation of Young's modulus in the rolling plane was compared with direct measurements. anisotropic. This anisotropy makes E and G functions of the orientation of the test specimen in the sheet so that a complete description of the elasticity requires many measurements on a large number of specimens. A more fundamental approach to the problem is to consider the sheet as a single crystal and measure the elements of its elastic modulus tensor. Although this appears formidable, there have been developed recently high-frequency ultrasonic techniques which yield these tensor elements directly. The symmetry of a rolled sheet is such that it behaves elastically as if it were an orthorhombic

Jan 1, 1965

By N. A. Gjostein

The self-diffusion coefficient D, for a surface near the (100) plane in copper was determined by means of the Mullins theory of thermal grooving, and was found to obey the Arrhenius relationship, and Do = 6.5 x 10' sq cm per sec. Suppression of surface diffusion by chemisorbed impurities encounters difliculties as a satisfactory explanation for the large Q, and Do. Instead, a model is presented which indicates that Q,, as determined by a mass-transport technique, may involve both the energy needed to form an adsorbed atom from a ledge in the surface as well as that for the movement of the adsorbed atom.; LITERATURE reviews1-3 indicate that the mechanism of self-diffusion on metal surfaces is not clearly understood. The lack of experimental data on the temperature variation of the surface self-diffusion coefficient DJ has contributed greatly to this situation. Radioactive tracer techniques4-rn have been used to measure Ds, but for many years only Nicker-son and Parker8 had determined (for polycrystalline silver) an activation energy B, and a frequency factor Do for the surface self-diffusion process. Even in this case, a mechanism could not be clearly inferred because it was suspeted7 that a contribution from volume diffusion was not taken into account by the investigators. Some later work by Li and parkere on polycrystalline gold, in which they obtained essentially the same results as for silver and in which volume diffusion could be accounted for, seems to invalidate this criticism. In 1957, Mullins10'11 developed the theory for the thermal grooving process. This advance has opened a new and promising technique for measuring the surface self-diffusion coefficient Ds. Recognizing that a thermal groove may form by several different mass transport processes, Mullins first consideredlo the cases where a) the groove forms by evaporation and condensation or b) by diffusion along the groove surface, in either case, in a system where a metal surface is in contact with only its dilute vapor phase. In a later paper," he extended the treatment to cases where the groove forms by volume diffusion into c) the solid phase or d) by mass transport through a passive gaseous phase at 1 atm pressure. The time dependence of the groove dimensions and the shape of the groove profiles predicted by Mullins' treat-

Jan 1, 1962

By D. F. Stoneburner, G. Derge, Ling Yang

PREVIOUS investigations in this laboratory have shown the existence of electronic conduction in several molten metal sulfides and the specific conductance of a number of these has been measured as a function of temperature.' In order to better understand the electronic nature of these melts, the thermoelectric power of a molten FeS-solid tungsten thermocouple was measured and a value of approximately 55 µv per "C was obtained in the temperature range from 1200" to 1400c2 These studies have been continued and the thermoelectric power of several other molten metal sulfide-solid tungsten thermocouples have been obtained. The experimental arrangement is shown in Fig. 1. A Kanthal furnace was used, doubly wound to prevent any undesirable effects of an external field. A 1/2-in. McDanel tube was fitted with two vertical 5/8-in. McDanel tubes as shown. These tubes were about 2 in. apart. A graphite boat was used to hold a re-crystallized alumina boat in which the prefused sulfide was placed. Two probes were made using

Jan 1, 1960

By R. T. DeHoff, L. J. Buteau, J. Kronsbein

Network topology is introduced and used to describe sinter bodies at various stages of the sintering process. The matrices of incidence, loop-branch. tree-branch, and cut-branch are constructed and their relationship to the three-dimensional void space and the common two-dimensional bounding surface is established. The sintering body is assunzed to remain connected but this is not so with the void space. The topological invariants known as Betti numbers are related to this observed fact and provide parameters which measure the progress of the sintering process. They are extracted from the matrices by means of known algebraic operations. Experimental procedures are given which permit actual construction of the various matrices and determination of the Betti numbers by serial sectioning of the sinter bodies. A sinter body is produced by exposing a large number of metallic particles (0.00001 to 0.3 in. diam) to a temperature that is near but below their melting point. When these particles are maintained at such a temperature for a sufficient length of time, it is found that the interparticle contacts become welded. As time increases, these welds, often referred to as necks, grow. All of the space between the particles constitutes a single connected void; an ultimate consequence of sintering is the breaking up of the void into separated parts. A mathematical apparatus for describing the geometric characteristics of a sinter body at each stage of the process was first proposed by F. N. Rhinesl when he introduced the application of topology to this problem. He described the sinter body in terms of one of its topological invariants, the genus. The most important advantage of a topological description is invariance with respect to continuous deformations. This means that particle sizes and shapes need not be considered, resulting in considerable simplification and generalization of the geometric description. In the case of a sinter body, wherein the internal geometry is particularly complex, the simplification provided by topological concepts is absolutely essential. The important parameters in the Rhines model are the number of particles and the interparticle contacts. The invariant considered called genus is given by Euler's formula for the connectivity of two-dimensional surface of the sinter body. This surface includes the pore-solid interface. The Euler formula is G = C - P + 1 [1] where C = number of interparticle contacts, P = number of particles, and G = genus of the surface bounding the sinter body. In topology the term genus may be used to mean the number of self re-entrant cuts that can be made on a closed surface without dissecting it into more than one connected surface. Its use is restricted to the surface and does not apply to the three-dimensional volume enclosed or excluded by it. In the case of sintering, the volume enclosed by the surface is defined as the sinter body; the volume excluded will be referred to as the void space. There is an intrinsic reason for using genus as a characteristic number associated with a sinter body. The genus is related to the Gaussian curvature by the integral of the product of the principal curvatures taken over the entire surface of the sinter body, as given by

Jan 1, 1965

By J. J. Pye, J. B. Drew

Mzasurements of the surface diffusion coefficients of metals have been made. Diffusion profiles for the Ag-Ag system were obtained by means of a radioactive point source and a precision auto-radiographic technique. The activation energy for silver self diffusion (=8.1 kcal per mole) is lower than that previously reported (-10 kcal per mole) on poly crystalline wire by Nickerson and Parker. The bresent data indicate an effect due to parasitic volume diffusion at temperatures above 500°C. RELATIVELY few measurements have been made of the surface self-diffusion coefficients of metals. Nickerson and arker' measured the diffusion of silver over the surface of poly crystalline wires and estimated that the activation energy was 10.3 kcal per mole. Winegard and chalmers2 carried out measurements on both polycrystalline and single crystal surfaces but did not report a value of the activation energy. They found, however, that at temperatures between 250" and 400°C the diffusion coefficients were on the order of lo-' sq cm per sec and that there was an acceleration of the migration of silver on the polycrystalline sample when a change of surface shape occurred. Winegard and Chalmers used an autoradiographic technique, hereafter designated ARG, and Nickerson and Parker used a surface scanning geiger counter in order to determine the diffusion profiles. More recently, Hackerman and simpson3 measured the surface self-diffusion coefficient of copper at a single temperature (750°C), and the value of the diffusivity (- 10-5 sq cm per sec) is in agreement with that given by jostein from his thermal grooving measurements. This paper reports the results of an investigation of the surface self-diffusion coefficients of silver over a large temperature range and describes the adaptation of autoradiographic (ARG) techniques for the determination of diffusion profiles obtained from a radioactive point source. EXPERIMENTAL PROCEDURE The experimental procedure is a modification of the method employed by Hackerman and simpson3 in their measurements on copper. A brief description of their technique is as follows: A radioactive needle which sinters to the surface during the diffusion an- neal serves as the source of diffusing atoms. After the diffusion run the needle is removed and the surface is scanned with a shielded counting arrangement. The diffusion profiles reported in this paper were obtained by a modification of the above procedure which employs a precision ARG technique. Previous investigations in this laboratory and elsewhere51B have shown that under carefully controlled developing conditions and by the use of calibration sources a linear relation exists between the concentration of the isotope and the photographic density for values below unity. The use of ARG under these conditions has advantages over the counter scanning method in that cumbersome shielding and requirements for great mechanical precision of the scanner are eliminated. Also the ARG gives a complete picture of the surface which is advantageous in studies of anisotropic diffusion. A recording microdensitometer having a 0.1 p wide slit was employed. At low temperatures the disturbing effects of subsurface radiations are negligible. The diffusion anneals are carried out in the cell shown in Fig. 1. The needle is formed by grinding down a 1.0 mm rod of high-purity silver until a tip of 0.2 mm radius or smaller is formed. This tip is plated withA"' which becomes the source of the diffusing atoms that are detected by ARG. The needle carrier and the crystal holder, Fig. 1 are constructed of quartz and ports are provided in the holder pedestal which allow free vapor circulation ((2.0 oz) and the carrier apron fits snugly over the crystal holder cap, insuring that the needle does not move and scratch the surface. Temperatures are provided by a stabilized tubular furnace which can be quickly positioned around the cell, thus bringing the crystal up to temperature in a time that is short compared to the diffusing times. The diffusion anneals range from 2 hr for the high-temperature samples to about 25 hr for those at the lowest temperature. The possibility of vapor transport of the radioactive metal as a contributing factor in the diffusion profile was investigated in two ways. One method was to suspend the needle directly over a dummy sample, raise the temperature, for periods of time equal to the diffusion times, and then take an auto-radiograph of the surface. Negligible radioactivity appeared. In the second method a thin slot in the crystal face on one side of the source provided a "cong path" for surface diffusion. If evaporation was the primary source of surface atoms the region of radioactivity around the source would be symmetrical. This was not the case. The profile dipped abruptly at the edge of the slot but on the other side of the source the usual diffusion profile appeared.

Jan 1, 1963

By R. W. Hertzberg, F. D. Lemkey, J. A. Ford

The technique of unidirectional solidification has been applied to the A1-AI3Ni and A1-CuAl2 ezltectic alloy systems; the controlled microstructure of A1-A3Ni consists of parallel A13Ni whiskers emhedded in an aluminum matrix while the Al-CuAl2, system solidifies as parallel alternate lamellae of aluminm and CuAl2. Mechanical tests (tension and flexure) have shown that the particle-matrix interface bond formed during unidirectional solidification allowes for efficient load transfer from the matrix to the reinforcing phase. Strength and fracture characterstics of the alloys were studied as a function of phase orientation; the tensile strength of controlled Al-Al3Ni samples was 43,000 psi (three times greater than that of as-cast specime~s). The strength of the Al-CIA2, cutectic was studied as a function of lamellae-flexure stress-axis orientation using a microbend apparatus. A twofold increase in outer fiber stress was observed as the lamellae were rotated from a position of no reinforcing (perpendicular to the stress axis) to one of maximum reinforcing (parallel to the stress axis), The brittle reinforcing constituent in each alloy (A13Ni and CuAl2) was observed to he the nucleating phase for fracture. It is to be concluded from this investigation that unidirectionally solidified eutectic alloys have a potential use as reinforcing coniposite) materials. In a recent review article, Chadwick1 has described the work of many investigators in the area of controlled solidification of binary eutectic alloys. Thus far, more than twenty eutectic alloys have been unidirectionally solidified to produce structures consisting of two phases in the form of alternating lamellae or as whiskers embedded within a continuous matrix; in either case, the phases are aligned parallel to the growth direction. In particular, Kraft and Albright2 have shown that the Al-CuA1, eutectic alloy exhibits a lamellar microstruc- ture as a result of unidirectional solidification while Lemkey, Hertzberg, and Ford3 have studied the A13Ni whisker-aluminum matrix morphology in the A1-A13Ni system. Whether these alloys could exhibit reinforcing behavior would have to depend upon the strength and volume fraction of the reinforcing phase (AI3Ni and CuA12) and the nature of the bond between the reinforcing phase and the matrix. Hertzberg and Kraft4 have shown that the interface bond between chromium whiskers and a copper matrix in unidirectionally solidified Cu-Cr is particularly good while Lemkey and Kraft5 have reported chromium-whisker tensile strengths in excess of 1,000,000 psi. However, with the chromium volume fraction being less than 2 pct, no reinforcing behavior was displayed by the Cu-Cr alloy.6 In view of the encouraging results concerning the good interfacial bond and high strength of the reinforcing phase in the Cu-Cr system and preliminary mechanical test data reported by Lemkey et al.3 for the A1-A13Ni system, the A1-CuA1, and A1-A13Ni eutectic alloys containing 50 and 10 vol pct, respectively, of the reinforcing phase give promise of possessing mechanical reinforcing behavior. Based on this premise, the objective of the investigation was to study the strength and fracture behavior exhibited by these anisotropic microstructures. EXPERIMENTAL PROCEDURE Controlled Solidification Procedure. An extensive description of the technique of unidirectional solidification has been presented elsewhere.' However, the pertinent details for each eutectic system studied are discussed below. Al-A13Ni System. Several master heats of this alloy were prepared by vacuum-induction melting in an A12O3crucible and casting in SiO2 mold or by induction melting in alumina crucibles under argon. The initial heats were prepared from 99.99 pct A1 and 99.99+ pct Ni, while later heats used 99.999+ pct A1 and 99.99+ pct Ni. Specimens of these master heats were unidirectionally solidified using induction and resistance heating sources and argon atmospheres at growth rates from 1.1 to 28.7 cm per hr. The thermal gradients in the liquid were controlled in the range of 26" to 36°C per cm for these runs. Al-CuA12 System. The master heat and controlled

Jan 1, 1965

By J. B. Morgan

The mechanical behavior of MgCu2 from 20 o to 725°C has been determined by "brittle-ring" tensite-test techniques, axial compression, and bending experiments. Compressive ductility begins at 450°C (0.65 T/Tm) and slowly increases with increasing temperature white the tensile ductility is characterized by an abrupt brittle-ductile transition at 600°C (0.8T/Tm). The compressive strength increases eightfold from room temperature to 400°C with a "normal" decrease as the temperature is further increased. The tensile strength goes through a rather abrupt threefold increase above 600°C with a subsequent decrease at higher temperatures. The slip system is (111)[110] and the twinning system is (111)[112]. At high temperatures MgCu2 is quite rate-sensitive, and dynamic yield points are observed that are extremely orientation-dependent. The high-temperature creep behavior under constant load is characterized by an increasing strain-rate with time. The general characteristics of the mechanical behavior make it possible to classify MgCu2 with those materials in which initial mobile dislocation density and the dislocation velocity-stress dependence are significant in determining mechanical response. Mgcu2 is the C15 prototype of the cubic Laves phase. Its range of homogeneity extends from 32.8 to 35.5 at. pct Mg at 500°C and narrows at lower temperatures. The lattice parameter is reported to vary from 7.02 to 7.05A,1 and the compound is completely ordered to the melting point of 819°C. MgCu2 has a brassy metallic appearance, is extremely brittle, and many of its physical properties are intermediate between those of the constituent elements: the electrical resistivity is 3.5 x 10-6 ohm-cm, Young's modulus is 10x 106 psi, the shear modulus is 6 x l06 psi, and the thermal coefficient of expansion is 18.7 x 10-6 cm per cm per "C from 20" to 250°C. Usually, properties of intermetallic compounds as functions of temperature have been determined by hardness, compression, or impact behavior, and almost all of these experiments have been performed on polycrystalline materials. Tammann and Dahl,2 for instance, determined the onset of ductility in a series of binary intermetallics by ball-hardness measurements and found the onset to range between 0.71 and 0.96 T/T,. Lowrie's3 investigations included some tensile properties, and he found ductility beginning around 0.65 T/Tm for most materials examined. His work did include some data on polycrystalline, nonstoichiometric MgCu2, however. The present work is an attempt to extend the examination of the particular compound MgCu2 in single-crystal form in order to aid the formation of general concepts about mechanical be-

Jan 1, 1965

By Lawrence A. Shepard, Richard H. Krock

A series of ductile, two phase W-Ni-Fe composites, sintered in the presence of a liquid phase, were tested in tension. Identical room temperature stress-strain curves were obtained for specimens containing from 80 to 92 wt pct W (58 to 75 vol pct W particles). The composites exhibited a maximum elongation of 29pct at room temperature, and 10.7 pct at 77 °K. The tungsten particles in the composite elongated by the same amount at these temperatures. Single-phase alloy specimens matching the composition of the composite matrix showed about one half the flow stress of the composites. The test results demonstrated that the mechanical properties of W-Ni-Fe composites are determined by the tungsten particles alone and are independent of matrix volume fraction or mean free path over the composition range studied. A general prediction of the plastic behavior of two-phase alloys or composites from a knowledge of the properties, size, shape, and dispersion of the individual components is not, at present, possible. The excellent theoretical and experimental exposition of the dispersion-strengthening problem applies only to cases where the stronger phase is present as fine particles and in volume fractions of a few percent. Materials in which the stronger phase constitutes a significant fraction of the volume, although of broad engineering interest, have received comparatively little analytical attention. This important group of materials is the concern of the present study. The degree of deformation of each phase during the plastic working of a ductile two-phase alloy has been established by Boas and Honeycombe,1 Clare-brough,2 and Clarebrough and berger.3 These workers empirically correlated the recrystallization temperature of the individual phases with the degree of cold work sustained by each. It was shown that the deformations of both phases are equal and therefore the same as that sustained by the composite whole when the volume fraction of the stronger phase exceeds 35 pct. At yielding, however, deformation begins in the weaker phase. A direct correlation of composite strength, as a function of deformation, with the strength of the individual components cannot be made. In the presence of stronger phase particles, the flow stress of the weaker matrix phase is enhanced by an additional hydrostatic component. The development and potential magnitude of matrix strength promotion upon straining has not even been experimentally established. Quantitative work in this field has been confined to ceramic-metal composites. These materials fracture at strains considerably below 1 pct by cracking of the ceramic particles.4-6 An indication of the potential constrained matrix strength capacity is given by the observation by Nishimatsu and Gur-land4 that the maximum tensile strength of a WC-Co composite is almost double the bulk strength of the cobalt matrix. The relationship between composite strength and geometry is ordinarily expressed in terms of the mean-free matrix path (referred to subsequently as MFP) between the stronger particles.* This factor takes into account both the volume fraction of matrix phase and the average size of the particles. unke17 concluded that composite strength varied directly with the decreasing log MFP, following the Gen-samer8 relationship. However, Gurland and ~radzil~ reported a strength maximum at an MFP of the order of 1 p in WC-Co composites. A ductile two phase metallic composite was chosen for the present investigation of composite mechanical properties with the view of avoiding the difficulties and ambiguities noted above. The W-Ni-Fe heavy alloy of Green, Jones, and pitkin9 contains 80 to 94 wt pct W, and nickel and iron in a weight ratio

Jan 1, 1963