J.Am. Ceran.Soe,89p3309-3324(2006 DO:10.l111551-2916.2006.01342x c The American Ceramic Society urna Developments in Oxide Fiber Composites frank zoot Materials Department, University of California, Santa Barbara, California 93106 Prospects for revolutionary design of future power generation Notwithstanding this progress, the long-term durability of systems are contingent on the development of durable high-per- SiC-Sic composites continues to be plagued by two persistent formance ceramic composites. With recent discoveries in mate problems. ()In combustion environments that contain rials and manufacturing concepts, composites with all-oxide water vapor, recession occurs by volatilization of the silica constituents have emerged as leading candidates, especially for scale. Environmental barrier coatings must then be used components requiring a long service life in oxidizing environ- in order to achieve minimum durability goals in practical de- ments. Their insertion into engineering systems is imminent. The signs. (ii) Although presently of secondary concern, long-stand intent of this article is to present a synopsis of the current ing problems with oxidation embrittlement at intermediate understanding of oxide composites as well as to identify out- temperatures remain unresolved. These deficiencies have ues that require resolution for successful implemen- purred interest and developments in all-oxide cFCCs tation. Emphasis is directed toward material systems and Indeed, oxide systems have emerged as leading contenders for microstructural concepts that lead to high toughness and long applications requiring long service lives(>10 h) in oxidizin term durability. These include: the emergence of La monazite environments and related compounds as fiber-coating materials, the introduc High toughness in CFCCs is achieved by one of three tion of the porous-matrix concept as an alternative to fiber microstructural design paths(Fig. 1). All seek to promote un- coatings, and novel strategies for enabling damage toleranc orrelated fiber failure resulting in high fiber bundle strength while retaining long-term morphological stability. Additionall and energy dissipation during subsequent pullout The most materials and mechanics models that provide insights into ma- common approach uses a fiber coating that either forms a terial design, morphology evolution, and composite properties weak interface with the fibers or has an inherently low fracture toughness (Fig. 1(b)). It has been utilized extensively in SiC/SiC, C/SiC, and C/C fiber composites, principally through C and bn coatings Similar mechanisms can be enabled through the use of fine-scale matrix porosity, obviating the need for a fiber coating (Fig. I(c). To ensure durability, the matrix must be phase compatible with the fibers, because demand for high-temperature thermostructural materials of their intimate contact in the absence of a coating. Addition- ntinues to grow, fueled principally by power generation ystems for aircraft engines, land-based turbines, rockets, and ally, the pore structure must be retained at the targeted use ly, hypersonic missiles and fight vehicles. Typi temperature. The third approach uses fugitive coatings most omponents include combustors, nozzles, and thermal insula ones that are volatilized by oxidation after composite fabrica- tion. With their high melting point, strength, and toughness, tion, leaving a narrow gap at the fiber-matrix boundary continuous-fiber ceramic composites( CFCCs)offer the greatest (Fig. I(d)). The present article highlights the most signifi- ant developments in the implementation of these design strat of these system egies for oxide CFCCs Among commercially available oxide fibers, preference has Over the past 2 decades, the vast majority of CFCC research been given to two specific types: () NextelmM610-a polycrys has focused on Sic-SiC systems. The supporting manufacturing technology has reached a high level of sophistication and talline, small-diameter(10 um) alumina fiber, with high strength maturity. Large components are routinely manufactured to 1000%.C; and (ii) Nextel"720-a polycrystalline mul and have been tested in turbine engines and burner rigs. and moderately elevated temperatures(relative to 610), but with superior creep resistance and microstructural stability at high temperatures, to about 1200oC. 9 Although most activities in D. Green-contributing editor high-performance oxide CFCCs have focused on these, some concept demonstrations have used large diameter(>100 um) sapphire and eutectic alloys. The latter are not amenable to pt No. 22223. Received August 8, 2006: approved September 7. 2006. weaving and remain too expensive to find widespread use in the Force Office of Scientific Research (award number foreseeable future. Brief references to fiber types are included in to whom correspondence should be addressed. e-mail: zok(a enginecring. this article. However, the status of oxide fibers is beyond the scope of this paper. Feature
Developments in Oxide Fiber Composites Frank W. Zokw Materials Department, University of California, Santa Barbara, California 93106 Prospects for revolutionary design of future power generation systems are contingent on the development of durable high-performance ceramic composites. With recent discoveries in materials and manufacturing concepts, composites with all-oxide constituents have emerged as leading candidates, especially for components requiring a long service life in oxidizing environments. Their insertion into engineering systems is imminent. The intent of this article is to present a synopsis of the current understanding of oxide composites as well as to identify outstanding issues that require resolution for successful implementation. Emphasis is directed toward material systems and microstructural concepts that lead to high toughness and longterm durability. These include: the emergence of La monazite and related compounds as fiber-coating materials, the introduction of the porous-matrix concept as an alternative to fiber coatings, and novel strategies for enabling damage tolerance while retaining long-term morphological stability. Additionally, materials and mechanics models that provide insights into material design, morphology evolution, and composite properties are reviewed. I. Introduction THE demand for high-temperature thermostructural materials continues to grow, fueled principally by power generation systems for aircraft engines, land-based turbines, rockets, and, most recently, hypersonic missiles and flight vehicles. Typical components include combustors, nozzles, and thermal insulation. With their high melting point, strength, and toughness, continuous-fiber ceramic composites (CFCCs) offer the greatest potential for enabling elevations in the operating temperatures of these systems. Over the past 2 decades, the vast majority of CFCC research has focused on SiC–SiC systems. The supporting manufacturing technology has reached a high level of sophistication and maturity. Large components are routinely manufactured and have been tested in turbine engines and burner rigs. Notwithstanding this progress, the long-term durability of SiC–SiC composites continues to be plagued by two persistent problems. (i) In combustion environments that contain water vapor, recession occurs by volatilization of the silica scale.1–3 Environmental barrier coatings must then be used in order to achieve minimum durability goals in practical designs. (ii) Although presently of secondary concern, long-standing problems with oxidation embrittlement at intermediate temperatures remain unresolved. These deficiencies have spurred interest and developments in all-oxide CFCCs. Indeed, oxide systems have emerged as leading contenders for applications requiring long service lives (4104 h) in oxidizing environments. High toughness in CFCCs is achieved by one of three microstructural design paths (Fig. 1). All seek to promote uncorrelated fiber failure, resulting in high fiber bundle strength and energy dissipation during subsequent pullout. The most common approach uses a fiber coating that either forms a weak interface with the fibers or has an inherently low fracture toughness (Fig. 1(b)). It has been utilized extensively in SiC/SiC, C/SiC, and C/C fiber composites, principally through C and BN coatings.4 Similar mechanisms can be enabled through the use of fine-scale matrix porosity, obviating the need for a fiber coating (Fig. 1(c)).5–15 To ensure durability, the matrix must be phase compatible with the fibers, because of their intimate contact in the absence of a coating. Additionally, the pore structure must be retained at the targeted use temperature. The third approach uses fugitive coatings: ones that are volatilized by oxidation after composite fabrication, leaving a narrow gap at the fiber–matrix boundary (Fig. 1(d)).16–18 The present article highlights the most signifi- cant developments in the implementation of these design strategies for oxide CFCCs. Among commercially available oxide fibers, preference has been given to two specific types: (i) Nextelt 610—a polycrystalline, small-diameter (10 mm) alumina fiber, with high strength to 10001–11001C; and (ii) Nextelt 720—a polycrystalline mullite/alumina fiber with a somewhat lower strength at ambient and moderately elevated temperatures (relative to 610), but with superior creep resistance and microstructural stability at high temperatures, to about 12001C.19 Although most activities in high-performance oxide CFCCs have focused on these, some concept demonstrations have used large diameter (4100 mm) sapphire and eutectic alloys. The latter are not amenable to weaving and remain too expensive to find widespread use in the foreseeable future. Brief references to fiber types are included in this article. However, the status of oxide fibers is beyond the scope of this paper. Feature D. Green—contributing editor This work was supported by the Air Force Office of Scientific Research (award number F49550-05-1-0134), monitored by Dr. B. L. Lee. w Author to whom correspondence should be addressed. e-mail: zok@engineering. ucsb.edu Manuscript No. 22223. Received August 8, 2006; approved September 7, 2006. Journal J. Am. Ceram. Soc., 89 [11] 3309–3324 (2006) DOI: 10.1111/j.1551-2916.2006.01342.x r 2006 The American Ceramic Society
3310 Journal of the American Ceramic Society--Zok Vol. 89. No. I Coating Debond Matrⅸx crack Pullout Porous Debond crack Matrix crack Debonding/sliding 圈 Interface Fig 1. Microstructural concepts for enabling crack deflection in continuousfib Il. Developments in Fiber vancement in the underpinning science and technology have been made by investigators at the U.S. Air Force Research Lab- Undoubtedly, the most significant development in fiber coatings oratory and at Rockwell Scientific(formerly Rockwell Science has been the discovery that rare-earth phosphates such as La- scientific and engineering challenges have been identified and monazite bond weakly to other oxides. - In addition to forn ing low-toughness interfaces, monazite is non-toxic; insoluble in addressed. The key developments and outstanding issues follow water, acids, and bases does not decompose up to its melting Among the numerous recipes for monazite coatings, the most promising uses rhabdophane (LaPO4. 1/ 2H2O) sols derived point(>2000%C); and is not easily reducible below 1400 C In from La(NO3)3 and H3 PO4.- To mitigate the deleterious ef. addition, it has anomalously low hardness(relative to other re- fects of nitric acid (a by-product of the reaction that forms fractory ceramics), thereby facilitating plastic deformation dur- ing fiber-matrix sliding -.Concurrent with the development monazite), the sols are repeatedly washed in de-ionized water of monazite, other mixed-oxide compounds(niobates, tung- before application. Otherwise, significant reductions in fiber states, and vanadates) have been pursued as candidate coating strength are obtained following coating( Fig. 2) materials,23-27 although none exhibits a spectrum of propertie In one successful implementation, the coating is applied by rival that of monazite cible li- The monazite discovery proved pivotal in the resurgence of quid used to minimize bridging between coated fibers.31.32The xide CFCCs. During the past decade, the most substantive ad- fibers are then passed through an in-line furnace(typically at 900-1200C)and spooled. Both Nextel610 and Nextel720 fibers endure this process with negligible strength loss. In its present form, this coating method is restricted to individual tows. Thus, to make useful shapes, the fibers must be first coated in tow form and subsequently woven into the desired architec- ture. The draw back is that the weaving can damage the coating a consequence of the weak interfacial bond The thermochemical compatibility of monazite with a wide range of oxide fibers has been definitively demonstrated. For virtually all systems of interest(including Nextel610 and 720, sapphire, single-crystal mullite, and AlO3/ZrO, and Al,O3/yt rium aluminum garnet eutectics). interfaces with monazite are sufficiently weak to allow debonding to occur when cracks ap- Nextel 720 Fiber roach from within the monazite(Fig. 3),33 even when the re- sidual radial compressive stresses are large(Fig. 4).However, 1400 the resistance to subsequent sliding appears to be considerably Temperature(°C) higher than that of C-or BN-coated fibers in SiC-based CFCCs Sliding stresses of the former systems are typically in the range Fig. 2. Effects of heat-treatment temperature on the strength of Nex of 130-250 MPa, dependent on the thermal expansion coeffi tel"720 fibers after coating with either washed or unwashed rhabdo- cients of the three constituents as well as the radial misfit strain hane sols(Adapted from Hay and boakye2) roduced by surface roughness when sliding occurs(Fig
II. Developments in Fiber Coatings (1) Monazite Undoubtedly, the most significant development in fiber coatings has been the discovery that rare-earth phosphates such as Lamonazite bond weakly to other oxides.20–22 In addition to forming low-toughness interfaces, monazite is non-toxic; insoluble in water, acids, and bases; does not decompose up to its melting point (420001C); and is not easily reducible below 14001C. In addition, it has anomalously low hardness (relative to other refractory ceramics), thereby facilitating plastic deformation during fiber–matrix sliding.23,24 Concurrent with the development of monazite, other mixed-oxide compounds (niobates, tungstates, and vanadates) have been pursued as candidate coating materials,25–27 although none exhibits a spectrum of properties to rival that of monazite. The monazite discovery proved pivotal in the resurgence of oxide CFCCs. During the past decade, the most substantive advancements in the underpinning science and technology have been made by investigators at the U.S. Air Force Research Laboratory and at Rockwell Scientific (formerly Rockwell Science Center, where the monazite discovery was made). Numerous scientific and engineering challenges have been identified and addressed. The key developments and outstanding issues follow. Among the numerous recipes for monazite coatings, the most promising uses rhabdophane (LaPO4 1/2H2O) sols derived from La(NO3)3 and H3PO4. 28–30 To mitigate the deleterious effects of nitric acid (a by-product of the reaction that forms monazite), the sols are repeatedly washed in de-ionized water before application. Otherwise, significant reductions in fiber strength are obtained following coating (Fig. 2). In one successful implementation, the coating is applied by passing continuous tows through the sol, with an immiscible liquid used to minimize bridging between coated fibers.31,32 The fibers are then passed through an in-line furnace (typically at 9001–12001C) and spooled. Both Nextelt 610 and Nextelt 720 fibers endure this process with negligible strength loss.30 In its present form, this coating method is restricted to individual tows. Thus, to make useful shapes, the fibers must be first coated in tow form and subsequently woven into the desired architecture. The drawback is that the weaving can damage the coating: a consequence of the weak interfacial bond. The thermochemical compatibility of monazite with a wide range of oxide fibers has been definitively demonstrated. For virtually all systems of interest (including Nextelt 610 and 720, sapphire, single-crystal mullite, and Al2O3/ZrO2 and Al2O3/yttrium aluminum garnet eutectics), interfaces with monazite are sufficiently weak to allow debonding to occur when cracks approach from within the monazite (Fig. 3),33 even when the residual radial compressive stresses are large (Fig. 4).24 However, the resistance to subsequent sliding appears to be considerably higher than that of C- or BN-coated fibers in SiC-based CFCCs. Sliding stresses of the former systems are typically in the range of 130–250 MPa, dependent on the thermal expansion coeffi- cients of the three constituents as well as the radial misfit strain produced by surface roughness when sliding occurs (Fig. 5). Fig. 1. Microstructural concepts for enabling crack deflection in continuous-fiber ceramic composites. Fig. 2. Effects of heat-treatment temperature on the strength of Nextelt 720 fibers after coating with either washed or unwashed rhabdophane sols. (Adapted from Hay and Boakye29). 3310 Journal of the American Ceramic Society—Zok Vol. 89, No. 11
ovember 2006 Oxide Fiber Composites 3311 (a) 1 20 um um Fig 3. Fracture surfaces of an alumina/alumina continuous-fiber ceramic composite after 5 h of exposure at 1200.C: (a)uncoated fibers, (b)monazite- coated fibers. ( Courtesy Kristin Keller, AFRL. Reprinted with permission) Although the low hardness of monazite(5GPa2) facilitates plastic option. It can be deposited readily onto tows or woven fabric ccommodation of the misfit, the coating is less effective than C or by chemical vapor deposition or through pyrolysis of organic BN in mitigating these stresses In the latter, the low radial stiffness precursors and is readily oxidized at moderately high tempera- of the coatings allows for elastic accommodation of the misfit with tures. 16, 17 Although straightforward in principle, the approach nly moderate radial pressure and hence low sliding stress has two potential drawbacks: () matrix sintering treatments Three outstanding issues remain. (i Presently, there is no es- must be performed in an inert(non-oxidizing)environment, and tablished method for coating woven fiber cloths or preforms (i once the coating is oxidized, the fibers are unprotected from (distinct from tows). Such capability would circumvent the prob- the surrounding matrix and may be susceptible to bonding lems of weaving coated tows. (ii) The sliding stress of monazite ontact points coated fibers in dense ceramic matrices is considerably higher Preliminary feasibility studies have yielded encouraging re- than that in C- and BN-coated CFCCs, by as much as an order sults. When carbon-coated Nextel"720 fibers were embedded in of magnitude If excessively high, this may compromise compos- a dense calcium aluminosilicate matrix and the carbon subse ite toughness. Relative to SiC-SiC composites, more attention must be directed to thermal expansion mismatch and surfac quently oxidized, significant enhancements in fiber pullout were roughness effects in the oxide systems (ini) Although the issue of ture exposure was also improved. A more recent investigation fiber strength retention has been addressed, an assessment of the has also shown the benefits of combining fugitive coatings with efficacy of monazite coating on Nextel" 720 fibers(the highest orous matrices. For this purpose, composites were fabricated by infiltration of a mullite-20% alumina slurry into a carbon- demonstrated. It will likely require the use of a mullite-based coated Nextel" 720 preform, repeated impregnation and pyr- matrix, to minimize residual stress and allow fiber sliding subse- olysis of an alumina precursor, followed by oxidation of the uent to debonding, while retaining chemical compatibility with carbon. Preliminary results are presented in Fig. 6. With the the fibers. The large difference in thermal expansion coefficients fugitive coating, the composite exhibits significantly greater pull- of alumina (8x 10 K )and 720 fibers(6x 10 K)pr out as well as higher notched strength and fracture energy. The ludes the use of alumina-rich compositions as matrix choices mprovements are attributable to the combined effects of matrix porosity and the interfacial gap formed following carbon re- (2) Fugitive Coatings moval. The long-term stability and role of gap thickness in such Application of fugitive coatings to oxide CFCCs has received ystems is the focus of ongoing investigation. surprisingly little attention. Carbon appears to be the be 400 Al2O3 matrix LaPO4 coating LaPo Al2O3/ZrO2 Mullite Sapphire YAG/Al2O Sapphire Fig 4. (a) Microstructure and (b) fiber pullout in a dense LaPOa matrix 1200-1000-800-600-400-2000 reinforced with large-diameter sapphire fibers. Courtesy of Janet Radial misfit stress(MPa) Davis, Rockwell Scientific. Reprinted from J. Eur. Ceram. Soc., 19 J B. Davis, D B. Marshall, and P.E. D. Morgan,""Oxide Composites Fig. 5. Effects of radial misfit stress(from both thermal expansion mis- of AlO3 and LaPO4. pp. 2421-2426, 1999, with permission from atch and microstructural roughness) on the sliding stress of several Elsevier) monazite-coated fibers( Adapted from Davis et al. -
Although the low hardness of monazite (5 GPa20) facilitates plastic accommodation of the misfit, the coating is less effective than C or BN in mitigating these stresses. In the latter, the low radial stiffness of the coatings allows for elastic accommodation of the misfit with only moderate radial pressure and hence low sliding stress. Three outstanding issues remain. (i) Presently, there is no established method for coating woven fiber cloths or preforms (distinct from tows). Such capability would circumvent the problems of weaving coated tows. (ii) The sliding stress of monazitecoated fibers in dense ceramic matrices is considerably higher than that in C- and BN-coated CFCCs, by as much as an order of magnitude. If excessively high, this may compromise composite toughness. Relative to SiC–SiC composites, more attention must be directed to thermal expansion mismatch and surface roughness effects in the oxide systems. (iii) Although the issue of fiber strength retention has been addressed, an assessment of the efficacy of monazite coating on Nextelt 720 fibers (the highest temperature commercially available oxide fiber) has yet to be demonstrated. It will likely require the use of a mullite-based matrix, to minimize residual stress and allow fiber sliding subsequent to debonding, while retaining chemical compatibility with the fibers. The large difference in thermal expansion coefficients of alumina (B8 106 K1 ) and 720 fibers (6 106 K1 ) precludes the use of alumina-rich compositions as matrix choices. (2) Fugitive Coatings Application of fugitive coatings to oxide CFCCs has received surprisingly little attention. Carbon appears to be the best option. It can be deposited readily onto tows or woven fabric by chemical vapor deposition or through pyrolysis of organic precursors and is readily oxidized at moderately high temperatures.16,17 Although straightforward in principle, the approach has two potential drawbacks: (i) matrix sintering treatments must be performed in an inert (non-oxidizing) environment, and (ii) once the coating is oxidized, the fibers are unprotected from the surrounding matrix and may be susceptible to bonding at contact points. Preliminary feasibility studies have yielded encouraging results. When carbon-coated Nextelt 720 fibers were embedded in a dense calcium aluminosilicate matrix and the carbon subsequently oxidized, significant enhancements in fiber pullout were obtained.17 The retention in properties following high-temperature exposure was also improved. A more recent investigation has also shown the benefits of combining fugitive coatings with porous matrices.34 For this purpose, composites were fabricated by infiltration of a mullite–20% alumina slurry into a carboncoated Nextelt 720 preform, repeated impregnation and pyrolysis of an alumina precursor, followed by oxidation of the carbon.34 Preliminary results are presented in Fig. 6. With the fugitive coating, the composite exhibits significantly greater pullout as well as higher notched strength and fracture energy. The improvements are attributable to the combined effects of matrix porosity and the interfacial gap formed following carbon removal. The long-term stability and role of gap thickness in such systems is the focus of ongoing investigation. Fig. 3. Fracture surfaces of an alumina/alumina continuous-fiber ceramic composite after 5 h of exposure at 12001C: (a) uncoated fibers, (b) monazitecoated fibers.33 (Courtesy Kristin Keller, AFRL. Reprinted with permission). Fig. 4. (a) Microstructure and (b) fiber pullout in a dense LaPO4 matrix reinforced with large-diameter sapphire fibers.35 (Courtesy of Janet Davis, Rockwell Scientific. Reprinted from J. Eur. Ceram. Soc., 19, J.B. Davis, D.B. Marshall, and P.E.D. Morgan, ‘‘Oxide Composites of Al2O3 and LaPO4,’’ pp. 2421–2426, 1999, with permission from Elsevier). Fig. 5. Effects of radial misfit stress (from both thermal expansion mismatch and microstructural roughness) on the sliding stress of several monazite-coated fibers. (Adapted from Davis et al. 24). November 2006 Oxide Fiber Composites 3311
3312 Journal of the American Ceramic Society--Zok Fugitive C el61 No coat Al2O3-LaPOA 3 mm 2 Fig. 7. (a) Microstructure and (b) fiber pullout in a p aPO4/ ockwell Scientific. Reprinted from J. Eur. Ceram. 02 0.6 .0 19,JB.Davis, D.B. Marshall, and P.E. D Morgan, "Oxide Compos A Displacement(mm) Elsevier) some sense)all three principal toughening schemes: porous matri- Notched tensile behavior of a porous mullite-alumina matrix d monazite coati ngs. The initial pro- ed by Nextel 720 fibers, showing the effects cessing steps are identical to those used to produce porous-matrix ( Courtesy J H. Weaver) FCCs with an interfacial gap(described above). Following oxi dation of the carbon, a monazite precursor is repeatedly impre (3) Hybrid Concepts nated and pyrolyzed, thereby filling the interface gaps for One approach that obviates the problems associated with weav occupied by carbon as well as between the matrix particles ing of coated tows and simplifies processing involves a hybrid- (Fig. 8). If successful, this approach could provide an effective ization of the coated-fiber and the porous-matrix schemes Here, route to fabricating coated fiber composites with virtually any woven uncoated preforms are infiltrated with a monazite pre- architecture and configuration. A critical assessment of perform- cursor solution containing fine alumina particles. Following pyrolysis, a layer of monazite is formed on the fibers as well as composite exhibits extremely large pullout lengths(> 100R, (1) Microstructural Concll e between the alumina particles. The resulting matrix consists of a porous two-phase mixture of LaPOa and Al2O3( Fig. 7(a). The II. Matrix-Enabled damage tolerance When introduced in the mid-1990s, the porous-matrix co Ivity in typ cept was motivated principally by two factors: (i) the lack of a performance characteristics appear to be a consequence of ()the suitable suite of coatings for oxide fibers, and ( ii) the expectation monazite coating enabling crack deflection, and (i) the low of reduced manufacturing costs resulting from the absence of matrix stiffness reducing the radial constraints on the fiber oatings. Although the concept has proven to be an effective hence reducing the sliding resistance. Contrary to other reports lternative to fiber coatings for enabling damage tolerance, it of fiber strength degradation following exposure to acidic pre- as several inherent limitations: (i)CFCCs with two-dimension cursors, the reported combination appears to be innocuous. It al (2D) fiber architectures exhibit low thermal conductivit has been suggested that the alumina buffers the solution, strength, and fracture resistance in the through-thickness direc- making the fibers less susceptible to reaction with the precur ion;(ii)regardless of fiber architecture, these composites are non-hermetic;(iii) they are expected to have l compressive strengths than the dense matrix counterparts, because of the re- In addition to the combined porous-matrix/coated fiber scheme. a second hybrid duced constraint on fiber microbuckling: and(iv) they are more Mullite Monazite lextel 720 fibe n 10 Fig 8. Scanning electron micrographs of an oxide continuous fiber ceramic composite(using backscatter electron imaging). Monazite is present within the interface gap produced by arbon as well as between matrix particles. Monazite precursor provided by Janet Davis, Rockwell
(3) Hybrid Concepts One approach that obviates the problems associated with weaving of coated tows and simplifies processing involves a hybridization of the coated-fiber and the porous-matrix schemes. Here, woven uncoated preforms are infiltrated with a monazite precursor solution containing fine alumina particles.35 Following pyrolysis, a layer of monazite is formed on the fibers as well as between the alumina particles. The resulting matrix consists of a porous two-phase mixture of LaPO4 and Al2O3 (Fig. 7(a)). The composite exhibits extremely large pullout lengths ( 100R, with R being the fiber radius; Fig. 7(b)) and virtually no detectable notch sensitivity in typical specimen configurations. These performance characteristics appear to be a consequence of (i) the monazite coating enabling crack deflection, and (ii) the low matrix stiffness reducing the radial constraints on the fiber, hence reducing the sliding resistance. (Contrary to other reports of fiber strength degradation following exposure to acidic precursors, the reported combination appears to be innocuous. It has been suggested that the alumina buffers the solution, making the fibers less susceptible to reaction with the precursors and the decomposition products formed during precursor pyrolysis).29,35 In addition to the combined porous-matrix/coated fiber scheme, a second hybrid approach has emerged, using (in some sense) all three principal toughening schemes: porous matrices, fugitive coatings, and monazite coatings.34 The initial processing steps are identical to those used to produce porous-matrix CFCCs with an interfacial gap (described above). Following oxidation of the carbon, a monazite precursor is repeatedly impregnated and pyrolyzed, thereby filling the interface gaps formerly occupied by carbon as well as between the matrix particles (Fig. 8). If successful, this approach could provide an effective route to fabricating coated fiber composites with virtually any architecture and configuration. A critical assessment of performance and durability of this class of composite has yet to be made. III. Matrix-Enabled Damage Tolerance (1) Microstructural Concept When introduced in the mid-1990s,5–8 the porous-matrix concept was motivated principally by two factors: (i) the lack of a suitable suite of coatings for oxide fibers, and (ii) the expectation of reduced manufacturing costs resulting from the absence of coatings. Although the concept has proven to be an effective alternative to fiber coatings for enabling damage tolerance, it has several inherent limitations: (i) CFCCs with two-dimensional (2D) fiber architectures exhibit low thermal conductivity, strength, and fracture resistance in the through-thickness direction; (ii) regardless of fiber architecture, these composites are non-hermetic; (iii) they are expected to have lower compressive strengths than the dense matrix counterparts, because of the reduced constraint on fiber microbuckling; and (iv) they are more susceptible to wear.36 Fig. 6. Notched tensile behavior of a porous mullite–alumina matrix reinforced by Nextelt 720 fibers, showing the effects of a fugitive carbon coating. (Courtesy J. H. Weaver). Fig. 7. (a) Microstructure and (b) fiber pullout in a porous LaPO4/ Al2O3 matrix reinforced with Nextelt 610 alumina fibers. (Courtesy of Janet Davis, Rockwell Scientific. Reprinted from J. Eur. Ceram. Soc., 19, J.B. Davis, D.B. Marshall, and P.E.D. Morgan, ‘‘Oxide Composites of Al2O3 and LaPO4,’’ pp. 2421–2426, 1999, with permission from Elsevier). Fig. 8. Scanning electron micrographs of an oxide continuous-fiber ceramic composite (using backscatter electron imaging). Monazite is present within the interface gap produced by removal of the fugitive carbon as well as between matrix particles. (Monazite precursor provided by Janet Davis, Rockwell Scientific). 3312 Journal of the American Ceramic Society—Zok Vol. 89, No. 11
ovember 2006 Oxide Fiber Composites 3313 b) 100m (c) Fig 9. Damage and fracture mechanisms in a porousmatrix oxide Fig. 11. Minimal fiber pullout on the fracture surface of a mullite. ntinuous-fiber ceramic composite(a) Crack deflection and interface ased porous matrix continuous-fiber ceramic composite strengthened as infiltrated with an excessive amount of a precursor-derived alumina.(Courtes with epoxy while under load and then sectioned and polished. (b, c) M.A. Mattoni Uncorrelated fiber failure and pullout Material consists of Nextel720 fibers in an eight-harness satin weave and a mullite-alumina matrix. matrix, coated-fiber systems. That is, matrix cracks deflect along the fiber-matrix interface and fibers subsequently fail in an un- To ensure a morphologically stable pore structure, the matri- correlated manner, leading to pullout( Fig 9). Additionally, the ces typically consist of two dissimilar phases, distinguished by degree of notch sensitivity, characterized by open hole tension their sintering kinetics. The major phase is present as a contigu- tests, is comparable in the two classes of materials(Fig. 10). In ous 3D particle network. In turn, the network is bonded by a contrast, when the matrix is sintered or densified excessively, ei- less refractory ceramic or glass binder, in the form of either ther through processing or subsequent elevated temperature(in smaller sinterable particles or the product of precursor pyrolysis. service)exposure, embrittlement ensues. This is manifested in Particles in the main network dictate the long-term stability of planar fracture surfaces with minimal fiber pullout and signi the matrix against sintering, whereas particle junctions formed antly reduced toughness(Fig. Il) y the binder control the mechanical integrity of the matrix Additionally, the junctions at the fiber surface control the inter facial toughness (2) Debonding Mechanics When properly implemented, porous-matrix CFCCs exhibit fracture characteristics similar to those of conventional dense- twofold. Firstly, the bond between the matrix and the fibers is inherently weak. That is, the interface toughness, Ti. can be no greater than that of the matrix itself; for typical porosity levels 30 Notch Insensitive: ON/oo=1 nitude lower than that of the fibers, Tf, thereby ensuring a low. o6061A toughness interface. Secondly, because energy release rates scale ith elastic moduli. the red △1018stee leads to a reduction in the driving force for matrix cracks Oxide CFCCs To achieve high toughness in CFCCs, matrix cracks must deflect into the fiber/matrix interface rather than penetrate into 后08 Mullite- alumina he fibers. (A second condition-that interface sliding occur with SIC CFCCs only moderate resistance-must also be satisfied. ) The condi tions that satisfy this requirement are plotted in Fig. 12(a) Deflection is predicted when the toughness ratio, Ti/Tf is less than the energy release rate ratio, Ga/Gp, associated with defec- Glass PMCs tion and penetration. The latter is a function of the elastic mis- atch parameter 804 S Glass/Polyester △=(Er-Em) ch Sensitive: ON/oo=/ko=0 (Er +Em) 902 where E is the plane strain modulus, and the subscripts f and m denote fiber and matrix, respectively. For porous-matrix sys- tems, A takes on high values(>0.5); hence, the allowable tough ness ratio is also higl Hole Diameter, 2a(mm) Because of similarities in the matrix and fiber constituents in oxide CFCCs of present interest, the nature of bonding at the Fig 10. Open-hole tensile strength of metals, oxide, and Sic uIs-fiber ceramic composites(CFCCs), and polymer matrix composite fiber-matrix interface is similar to that between particles in the go is the unnotched tensile strength and ko is the elastic stress concen- natrix. Consequently, their toughnesses are expected to move in tration factor. All composites have two-dimensional fiber architectures tandem: that is, Ti=ol m where o is a non-dimensional par (either laminated or woven)and loads are applied parallel to one of the meter. As the packing density of matrix particles at the fiber fiber he normalized hole diameter is a/w=0.2 for all cases except surface is lower than that in the bulk, i<Im and hence o<I the oxide CFCC, wherein afw=1/3 For conservative design, o is taken to be 1
To ensure a morphologically stable pore structure, the matrices typically consist of two dissimilar phases, distinguished by their sintering kinetics. The major phase is present as a contiguous 3D particle network. In turn, the network is bonded by a less refractory ceramic or glass binder, in the form of either smaller sinterable particles or the product of precursor pyrolysis. Particles in the main network dictate the long-term stability of the matrix against sintering, whereas particle junctions formed by the binder control the mechanical integrity of the matrix. Additionally, the junctions at the fiber surface control the interfacial toughness. When properly implemented, porous-matrix CFCCs exhibit fracture characteristics similar to those of conventional densematrix, coated-fiber systems. That is, matrix cracks deflect along the fiber–matrix interface and fibers subsequently fail in an uncorrelated manner, leading to pullout (Fig. 9). Additionally, the degree of notch sensitivity, characterized by open hole tension tests, is comparable in the two classes of materials (Fig. 10).9 In contrast, when the matrix is sintered or densified excessively, either through processing or subsequent elevated temperature (inservice) exposure, embrittlement ensues. This is manifested in planar fracture surfaces with minimal fiber pullout and signifi- cantly reduced toughness (Fig. 11). (2) Debonding Mechanics The role of matrix porosity in enabling damage tolerance is twofold. Firstly, the bond between the matrix and the fibers is inherently weak. That is, the interface toughness, Gi, can be no greater than that of the matrix itself; for typical porosity levels (B30%), the matrix toughness, Gm, is about an order of magnitude lower than that of the fibers, Gf, thereby ensuring a lowtoughness interface. Secondly, because energy release rates scale with elastic moduli, the reduction in modulus due to porosity leads to a reduction in the driving force for matrix cracks. To achieve high toughness in CFCCs, matrix cracks must deflect into the fiber/matrix interface rather than penetrate into the fibers. (A second condition—that interface sliding occur with only moderate resistance—must also be satisfied.) The conditions that satisfy this requirement are plotted in Fig. 12(a).37 Deflection is predicted when the toughness ratio, Gi/Gf, is less than the energy release rate ratio, Gd/Gp, associated with deflection and penetration. The latter is a function of the elastic mismatch parameter, D Ef E ð Þ m Ef þ E ð Þ m (1) where E is the plane strain modulus, and the subscripts f and m denote fiber and matrix, respectively. For porous-matrix systems, D takes on high values (40.5); hence, the allowable toughness ratio is also high. Because of similarities in the matrix and fiber constituents in oxide CFCCs of present interest, the nature of bonding at the fiber–matrix interface is similar to that between particles in the matrix. Consequently, their toughnesses are expected to move in tandem: that is, Gi 5 oGm where o is a non-dimensional parameter. As the packing density of matrix particles at the fiber surface is lower than that in the bulk,38 GioGm and hence oo1. For conservative design, o is taken to be 1. Fig. 9. Damage and fracture mechanisms in a porous-matrix oxide continuous-fiber ceramic composite. (a) Crack deflection and interface debonding in a notched bend specimen. The specimen was infiltrated with epoxy while under load and then sectioned and polished. (b, c) Uncorrelated fiber failure and pullout. Material consists of Nextelt 720 fibers in an eight-harness satin weave and a mullite–alumina matrix. Fig. 10. Open-hole tensile strength of metals, oxide, and SiC continuous-fiber ceramic composites (CFCCs), and polymer matrix composites. so is the unnotched tensile strength and ks is the elastic stress concentration factor. All composites have two-dimensional fiber architectures (either laminated or woven) and loads are applied parallel to one of the fiber axes. The normalized hole diameter is a/w 5 0.2 for all cases except the oxide CFCC, wherein a/w 5 1/3. Fig. 11. Minimal fiber pullout on the fracture surface of a mullitebased porous matrix continuous-fiber ceramic composite strengthened with an excessive amount of a precursor-derived alumina. (Courtesy M. A. Mattoni). November 2006 Oxide Fiber Composites 3313
3314 Journal of the American Ceramic Society--Zok between deflection and penetration. Pertinent experimental measurements and mode g 3), n a tudies matrix properties Tm and Em are presented in Section V combining with the rresponding fiber properties via Eq assessment is made Penetration of the efficacy of the porous-matrix for a specific mat rix/fiber combination(Section V(4) 0.4 1200°c IV. Evolution of porous- Matrix Materials 5 Early generations of porous-matrix CFCCs(produced by Gen- eral Electric(Cincinnati, OH)and later by CoI Ceramics(Sa 0.2 Diego, CA)) comprise alumina powder and a silica-forming polymer precursor. Commercially, the composites are manu factured using procedures adapted from the polymer composites 2h industry. Prepregs are made by immersing woven fiber cloth into 02 a dispersed ceramic slurry. They are then stacked, warm molded Elastic mismatch parameter a n an autoclave, and fired at an elevated temperature(typically 1000C)to remove organics and pyrolyze the polymer. This rocess yields a contiguous nanoporous silica phase within an alumina particle network. Although the early generation com- posites with these constituents exhibited attractive mechanical properties after fabrication, significant degradation was ob- Low porosity tained following extended heat treatments at temperatures be- yond 1000.C: a consequence of matrix sintering. A variant of this concept uses a precursor-derived alumina as the binder (in place of silica), with the intent of enhancing morphological sta bility. However, the alumina particle network remains suscep tible to densification at yet higher temperatures(1100%-1200C) High porosity typical of targeted service conditions, especially when the pa ticles are fine(0, the energy precursor route requires additional steps, beyond that of slurry release rate ratio is well described by the empirical equation infiltration, and is thus more costly (i The presence of particulate alumina can compromise the stability of the mullite network, especially if its proportion ex- (2) ceeds the percolation threshold. Conversely, if the slurry is comprised of only mullite and the alumina is introduced subse- This formula has an error <4% over the range 0<A<0.95. quently via the precursor route, the contiguity of the pon setting Gd Gp=T Tr and combining the result with eq 1), the deflection condition can be re-expressed as (iii) Because of tions on the allowable fraction of par- ticulate alumina(to inhibit densification), the slurry route results in matrices that are relatively weak. Although essential for crack ∑≡0.13 (3) deflection, this weakness compromises the off-axis properties, especially the resistance to delamination. In contrast, the pre- cursor route allows for filling of the void space between the where 2 is a non-dimensional parameter that characterizes the particles in the network(at least while the pores remain open), propensity for crack deflection resulting in increases in the mechanical integrity of the network. The requisite combinations of T m/Tr and emEr are plotted in The latter route provides access to a broader range of matrix Fig 12(b)for three assumed values of ((0.3-1). As matrix sin- propertie tering/densification proceeds, the properties follow a trajectory The morphological stability of porous mullite-alumina from the lower left corner of the diagram (when the porosity ces at the targeted upper use temperatures of oxide CFCCs has is high) to the upper right, eventually crossing the boundar been demonstrated through experiments on neat (fiber-free)ma terials(Fig. 14(a)). Specifically, compacts of I um mullite par icles exhibit no detectable shrinkage after 1000 h of exposure at 200C. Mixtures containing <% alumina particles(0. 2 um diameter) are similarly stable, with porosity changing <0.5%
An estimate of the property combination that leads to de- flection is obtained in the following way. For D 0, the energy release rate ratio is well described by the empirical equation Gd Gp ¼ 1 4 1ð Þ D 0:9 (2) This formula has an error r4% over the range 0rDr0.95. Upon setting Gd/Gp 5 Gi/Gf and combining the result with Eq. (1), the deflection condition can be re-expressed as1 S 0:13 Gf Gm 1 þ Ef Em 0:9 > o (3) where S is a non-dimensional parameter that characterizes the propensity for crack deflection. The requisite combinations of Gm/Gf and Em/Ef are plotted in Fig. 12(b) for three assumed values of o (0.3–1). As matrix sintering/densification proceeds, the properties follow a trajectory from the lower left corner of the diagram (when the porosity is high) to the upper right, eventually crossing the boundary between deflection and penetration. Pertinent experimental measurements and modeling studies on the matrix properties Gm and Em are presented in Section V. Upon combining with the corresponding fiber properties via Eq. (3), an assessment is made of the efficacy of the porous-matrix concept for a specific matrix/fiber combination (Section V(4)). IV. Evolution of Porous-Matrix Materials Early generations of porous-matrix CFCCs (produced by General Electric (Cincinnati, OH) and later by COI Ceramics (San Diego, CA)) comprise alumina powder and a silica-forming polymer precursor.6 Commercially, the composites are manufactured using procedures adapted from the polymer composites industry. Prepregs are made by immersing woven fiber cloth into a dispersed ceramic slurry. They are then stacked, warm molded in an autoclave, and fired at an elevated temperature (typically 10001C) to remove organics and pyrolyze the polymer. This process yields a contiguous nanoporous silica phase within an alumina particle network. Although the early generation composites with these constituents exhibited attractive mechanical properties after fabrication, significant degradation was obtained following extended heat treatments at temperatures beyond 10001C: a consequence of matrix sintering. A variant of this concept uses a precursor-derived alumina as the binder (in place of silica), with the intent of enhancing morphological stability. However, the alumina particle network remains susceptible to densification at yet higher temperatures (11001–12001C), typical of targeted service conditions, especially when the particles are fine (o1 mm). More significant enhancements in stability have been achieved through the use of mullite as the main matrix constituent and alumina as the binder.8,12 In a common implementation, mullite powder is dispersed in an aqueous slurry and infiltrated into a fiber preform via a vacuum-assisted technique. The alumina is introduced in one of two ways: by mixing fine alumina particles into the mullite-containing slurry, or by subsequent impregnation and pyrolysis of an alumina precursor solution.39 The two processing routes lead to distinctly different matrix topologies, shown schematically in Fig. 13(a). Compositional maps of two prototypical systems are presented in Fig. 13(b).40,41 The preceding processing routes and the resulting microstructures are characterized by three attributes: (i) The mixed mullite/alumina slurry method allows both matrix phases to be infiltrated simultaneously. By contrast, the precursor route requires additional steps, beyond that of slurry infiltration, and is thus more costly. (ii) The presence of particulate alumina can compromise the stability of the mullite network, especially if its proportion exceeds the percolation threshold.42 Conversely, if the slurry is comprised of only mullite and the alumina is introduced subsequently via the precursor route, the contiguity of the mullite network is ensured. (iii) Because of limitations on the allowable fraction of particulate alumina (to inhibit densification), the slurry route results in matrices that are relatively weak. Although essential for crack deflection, this weakness compromises the off-axis properties, especially the resistance to delamination. In contrast, the precursor route allows for filling of the void space between the particles in the network (at least while the pores remain open), resulting in increases in the mechanical integrity of the network. The latter route provides access to a broader range of matrix properties. The morphological stability of porous mullite–alumina matrices at the targeted upper use temperatures of oxide CFCCs has been demonstrated through experiments on neat (fiber-free) materials (Fig. 14(a)).42 Specifically, compacts of 1 mm mullite particles exhibit no detectable shrinkage after 1000 h of exposure at 12001C. Mixtures containing r20% alumina particles (0.2 mm diameter) are similarly stable, with porosity changing o0.5% Fig. 12. (a) Conditions for crack deflection at a fiber–matrix interface (adapted from He and Hutchinson37). Experimental data are for mullite–alumina particle mixtures, assuming a toughness ratio oGi/Gm 5 1 and fiber properties Gf 5 15 J/m2 and Ef 5 260 GPa. (b) Complementary representation of crack deflection conditions, showing the critical combination of matrix toughness and modulus as well as the effects of o. 1 Here, Poisson’s ratios of the fiber and the matrix are assumed to be the same. Consequently, the plane strain modulus ratio Ef =Em can be replaced with the Young’s modulus ratio Ef =Em. 3314 Journal of the American Ceramic Society—Zok Vol. 89, No. 11
Oxide Fiber Composites Alumina Alumina Mullite Mullite Precursor-derived particles alumina Alumina Mullite μm (a) Schematics of the matrix topologies produced by mullite/alumina particle mixtures(top) and mullite particles bonded by precursor-derived a (bottom).(b)Compositional maps produced by energy-dispersive spectroscopy of TEM foils. The top image shows a particle mixture of 80% and 20% alumina(without precursor addition), whereas the bottom one is of a mullite powder compact bonded by 15% precursor-derived during the same aging cycle. Some shrinkage occurs for higher alumina content(> 30%0), but its evolution remains extremely luggish in relation to that for pure alumina powder(inset in (a)Mullite/Alumina Mixtures Fig. 14(a)), by about four orders of magnitude. The results con- 90M/10A firm that the mullite network is effective in inhibiting matrix 100M densification, even for relatively large amounts of the sinterable 80M20A Despite the absence of shrinkage, both Young,s modulus E 0910 nd the toughness f increase appreciably with aging time, by a factor of 3-4(Fig. 14(b): a consequence of surface-diffusion- ontrolled sintering at the particle junctions. The implications 9096 0.5 for crack deflection are addressed in a subsequent section. Alumina 70M/30A Yet further enhancements in matrix properties have been C achieved through the design of all-mmullite matrices The con- 94 cept uses two particle populations with vastly dissimilar sizes (e.g, I and 0.1 um) and exploits the differences in their sinterin kinetics. When present in the appropriate proportion, the small- er particles can be readily sintered to the larger particles without (b)Mullite compromising the stability of the main network. In principle, a similar structure could be achieved using mullite precursor so- Modulus, E lutions. However, the temperatures required for mullitization are well beyond those that the present fibers can withstand An additional processing enhancement involves use of a time- delayed setting agent(e.g, AIN)in the slurry. The agent (a) Effects of aging at 1200.C on the porosity s with compositions ranging from M)to 60% mullite and 40% alumina(60M/ property changes of pure mullite. Sin diffusion nism with diffusivity =4x10-30m'Is.The 0.1 02 ferred junction toughness is Tj=3 J/'m": only slightly higher than the face energy contribution (2y=2 J/m-) ging time, t(h)
during the same aging cycle. Some shrinkage occurs for higher alumina content ( 30%), but its evolution remains extremely sluggish in relation to that for pure alumina powder (inset in Fig. 14(a)), by about four orders of magnitude. The results con- firm that the mullite network is effective in inhibiting matrix densification, even for relatively large amounts of the sinterable phase. Despite the absence of shrinkage, both Young’s modulus E and the toughness G increase appreciably with aging time, by a factor of 3–4 (Fig. 14(b)): a consequence of surface-diffusioncontrolled sintering at the particle junctions.42 The implications for crack deflection are addressed in a subsequent section. Yet further enhancements in matrix properties have been achieved through the design of all-mullite matrices.43 The concept uses two particle populations with vastly dissimilar sizes (e.g., 1 and 0.1 mm) and exploits the differences in their sintering kinetics. When present in the appropriate proportion, the smaller particles can be readily sintered to the larger particles without compromising the stability of the main network. In principle, a similar structure could be achieved using mullite precursor solutions. However, the temperatures required for mullitization are well beyond those that the present fibers can withstand without degradation. An additional processing enhancement involves use of a timedelayed setting agent (e.g., AlN) in the slurry.43 The agent Fig. 13. (a) Schematics of the matrix topologies produced by mullite/alumina particle mixtures (top) and mullite particles bonded by precursor-derived alumina (bottom). (b) Compositional maps produced by energy-dispersive spectroscopy of TEM foils. The top image shows a particle mixture of 80% mullite and 20% alumina (without precursor addition), whereas the bottom one is of a mullite powder compact bonded by 15% precursor-derived alumina. (Adapted from Fujita et al. 40). Fig. 14. (a) Effects of aging at 12001C on the porosity of mixed mullite– alumina compacts with compositions ranging from 100% mullite (denoted 100M) to 60% mullite and 40% alumina (60M/40A). (b) Corresponding property changes of pure mullite. Sintering occurs by a surface diffusion mechanism with diffusivity dSDS 5 4 1030 m3 /s. The inferred junction toughness is Gj3 J/m2 : only slightly higher than the surface energy contribution (2g2 J/m2 ). November 2006 Oxide Fiber Composites 3315
3316 Journal of the American Ceramic Society--Zok duces a gradual increase in ph and a corresponding reduc- t mechanism (n=3 for vapo tion in the s potential between par ching the iso- for lattice diffusion. and n=7 for surface dif- electric point, the slurry coagulates. The process is designed to fusion). F a/R<l, Youngs modulus of the bonded affect coagulation after completion of vacuum bagging such that aggregate scales linearly with junction radius in accordance integrity for subsequent handling Properties of Porous Matrices Ep=(2x八(R Significant progress has been made in the understanding of the where Ep is Youngs modulus of the particles; z is the particle mechanical properties of porous matrices and their dependence oordination number (approximately six for random packing) n the topology of the constituent phases as well as their evo- D is the relative packing density; and 40. 76(calculated by lution with time. The key results from analytical models, nu- the discrete element method (DEm), described in Sidebar A) merical simulations, and experimental measurements are Combining Eqs. (4)and (5)yields the time dependence of the presented below (1) Monophase Particle Networks E 0.76 work follows a power law of the form*4s nophase particle net- Junction growth due to sintering in a m by surface diffusion(n=7), the time lus increase of only This has important consequences on the long-term durability of mullite-based CFCCs under typical where a is the junction radius, R is the particle radius, I is the The relationship between toughness and junction radius for a sintering time IR is a reference time, and n is a constant; both Ir bonded particle aggregate has been obtained from numerical Sidebar A. Numerical Simulation of Bonded Particle Aggregates Youngs modulus of a bonded particle aggregate is simulated numerically using the discrete element method (DEM). " The junction response that defines the element properties is derived from finite element analysis( FEA)of a single particle in a iodic array. The interaction between particle junctions. characterized by the displacement of one junction due to the force acting on another, is also derived from FEA. An isotropic.8 generated using a computer algorithm(inset of Fig. Al(a). To nsure equilibrium, each particle is required to touch at least w 0.6 three neighbors upon placement onto the aggregate. The final 9 particle packing density is 55% and the average coordination 3 0.4 number is 6: consistent with measured values for random loose packing of spherical particles For monophase systems, junction growth is simulated by uniformly expanding the particles and re-distributing the overlapping material uniformly over the free surface of the particles. In contrast, for systems containing a precursor derived binder, the material is modeled as an aggregate of Relative density touching monophase particles, each coated with a uniform layer of the second phase. The elastic response of the junctions 5.0 is calculated by FEA of a periodic array with the two phases b) xplicitly discretized For both mono-and two-phase systems the particle network is then subjected to a prescribed macroscopically uniform strain field and the effective ela sponse is determined using DEM. Typical numerical k and comparisons with experimental measurements"" are shown in Fig. Al(a). =12(aR)2 The toughness of the aggregate is also computed by DEM (Fig. Al(b). In this case, a crack is defined by a plane eparating particles that have had the junctions between them 8 broken(inset of Fig. Al(b). The simulation proceeds by incrementally increasing the remote displacement(for tension or the remote rotation(for bending), while allowing the method simulatior junctions at the crack tip to fail at a critical junction stress, Oe given by 0.30.40.5 Junction size, a/R Inction toughness. The results of the simula- lus and (b)toughness of monophase-bonded particle aggregates. E tions(Fig. Al(b)are well described by Eq (7)in the text. perimental measurements in(a)are for alumina(from Green et al
produces a gradual increase in pH and a corresponding reduction in the z potential between particles. Upon reaching the isoelectric point, the slurry coagulates. The process is designed to affect coagulation after completion of vacuum bagging such that the green preform exhibits substantially increased mechanical integrity for subsequent handling. V. Properties of Porous Matrices Significant progress has been made in the understanding of the mechanical properties of porous matrices and their dependence on the topology of the constituent phases as well as their evolution with time. The key results from analytical models, numerical simulations, and experimental measurements are presented below. (1) Monophase Particle Networks Junction growth due to sintering in a monophase particle network follows a power law of the form44,45: a R ¼ t tR 1=n (4) where a is the junction radius, R is the particle radius, t is the sintering time, tR is a reference time, and n is a constant; both tR and n depend on the transport mechanism (n 5 3 for vapor transport, n 5 5 for lattice diffusion, and n 5 7 for surface diffusion). Provided a/R 1, Young’s modulus of the bonded aggregate scales linearly with junction radius in accordance with46,47 E Ep ¼ x zD 2p a R (5) where Ep is Young’s modulus of the particles; z is the particle coordination number (approximately six for random packing); D is the relative packing density; and x0.76 (calculated by the discrete element method (DEM), described in Sidebar A). Combining Eqs. (4) and (5) yields the time dependence of the modulus E Ep ¼ 0:76 zD 2p t tR 1=n (6) When sintering occurs by surface diffusion (n 5 7), the time dependence is weak: a 10-fold increase in time leads to a modulus increase of only 10%. This has important consequences on the long-term durability of mullite-based CFCCs under typical service conditions. The relationship between toughness and junction radius for a bonded particle aggregate has been obtained from numerical Sidebar A. Numerical Simulation of Bonded Particle Aggregates Young’s modulus of a bonded particle aggregate is simulated numerically using the discrete element method (DEM).47 The junction response that defines the element properties is derived from finite element analysis (FEA) of a single particle in a periodic array. The interaction between particle junctions, characterized by the displacement of one junction due to the force acting on another, is also derived from FEA. An isotropic random aggregate of touching spherical particles is then generated using a computer algorithm (inset of Fig. A1(a)). To ensure equilibrium, each particle is required to touch at least three neighbors upon placement onto the aggregate. The final particle packing density is 55% and the average coordination number is 6: consistent with measured values for random loose packing of spherical particles. For monophase systems, junction growth is simulated by uniformly expanding the particles and re-distributing the overlapping material uniformly over the free surface of the particles. In contrast, for systems containing a precursorderived binder, the material is modeled as an aggregate of touching monophase particles, each coated with a uniform layer of the second phase. The elastic response of the junctions is calculated by FEA of a periodic array with the two phases explicitly discretized. For both mono- and two-phase systems, the particle network is then subjected to a prescribed macroscopically uniform strain field and the effective elastic response is determined using DEM. Typical numerical results and comparisons with experimental measurements48 are shown in Fig. A1(a). The toughness of the aggregate is also computed by DEM (Fig. A1(b)).42 In this case, a crack is defined by a plane separating particles that have had the junctions between them broken (inset of Fig. A1(b)). The simulation proceeds by incrementally increasing the remote displacement (for tension) or the remote rotation (for bending), while allowing the junctions at the crack tip to fail at a critical junction stress, sc, given by: sc ¼ 2 ffiffiffiffiffiffiffiffiffiffi EpGj pa r where Gj is the junction toughness. The results of the simulations (Fig. A1(b)) are well described by Eq. (7) in the text. Fig. A1. Discrete element method simulations of (a) Young’s modulus and (b) toughness of monophase-bonded particle aggregates. Experimental measurements in (a) are for alumina (from Green et al. 48). 3316 Journal of the American Ceramic Society—Zok Vol. 89, No. 11
ovember 2006 Oxide Fiber Composites simulations of fracture using the dEm (Sidebar A). The esults of the simulations are well described by Aging time 口1000hw=10±0.3 where I is the junction toughness. (The scaling with a is a 日 consequence of the dependence of toughness on junction area. Combining Eqs. (4)and(7) yields the corresponding time de- Although this sensitivity is greater than that of the modulus, the magnitude of the effect is small when surface diffusion is the ative sintering mechanism An assessment of these models is made through comparison with measurements on pure mullite compacts(Fig. 14(b)) Consistent er-law scalings of modulus and toughness with time are obtained for aging times up to 10 h. Furthermore extrapolation, the modulus and the tough dicted to increase by only 10% and 20%, respectively, for an 口 ional 1o h components). Such extrapolations are used in estimating long 日 term durability of oxide CFCCs, demonstrated below 口日口●。8 (2) Two-Phase Particle Networks The preceding models for junction growth and property changes in monophase aggregates are extended to two-phase particle mixtures(such as those comprising pore CFCCS). For a generic mixture of A and B particles, three 10 unction types are present: A-A, A-B, and B-B. The aggregate Alumina co properties are obtained by averaging the junction properties Fig 15. Effects of composition and aging time on(a) Youngs modulus weighted by the number fraction of the associated junction type d (b)toughness. The solid lines represents model predictions(Eqs. (10) To facilitate tractable solutions, the two particle types are as- nd (II)). Adapted from Fujita et al. sumed to be the same size and arranged randomly in the mix ture. A statistical analysis yields the junction fractions, f An analogous model for the toughness of a mixed aggregate is obtained when the contribution from each junction type is fAA=XA (9a) assumed to be proportional to T (a/r)and the toughnesses of the different types of junctions are then weighted by their re- spective number fractions Eq (9). The result where XA is the number fraction of A particles. +(1-XA)2 (11) 6 Young's modulus of the two-phase mixture is modeled fol- wing an approach similar to that used for monophase sys- tems, with appropriate modifications to reflect differences in where TAA, TBB, and TAB are the junction toughnesses: Y=TAA/ junction characteristics. () The description of junction stiffness IMM; and TAB is taken to be the average of TAa and TBe gtal is modified to account for the moduli of the particles on either The models are assessed by comparison with er de of the junction. In the hertzian limit, the stiffness of dis for (Fig. 15).- The unknown parameters are and y. Fitting the a=EAEB-49(i) The area of the A-B junction is assumed to be modulus measurements yields a junction area ratio n3+I the average of the areas of the A-a and B-B junctions. This consistent with the expectation that the alumina-containing result ressed in non-dimensional form as aab/aBB junctions should sinter more rapidly than those with only mu (1 +n)/2 where n is a junction area ratio, defined by ite. Then, upon fitting the toughness measurements, the inferred m=(aAAaBB).(ini) The modulus is determined from the arith- toughness ratio is p=1.0+0.3. The implications are twofold: () metic mean of junction stiffness, weighted by the respective the toughnesses of the three junction types are similar to one number fractions, given by eq(9). The result is another, and (ii the increase in aggregate toughness with alu mina content is due largely to the increase in the average junc- tion area En=X入+2X(-x-2 (3) Precursor-Derived Two-Phase Networks +(1-XA)2 When the binder phase is produced by a precursor route, the g p Upon comparing with numerical simulations, this poses, the topology is represented by one of two limiting ideal to be accurate in the domain in which the sinterable phase com- zations. In both, the major phase is treated as a contiguou prises <40%o of the total. etwork of uniform particles, radius R, and with average
simulations of fracture using the DEM (Sidebar A).49 The results of the simulations are well described by G Gj ¼ 12 a R 2 (7) where Gj is the junction toughness. (The scaling with a2 is a consequence of the dependence of toughness on junction area.) Combining Eqs. (4) and (7) yields the corresponding time dependence G Gj ¼ 12 t tR 2=n (8) Although this sensitivity is greater than that of the modulus, the magnitude of the effect is small when surface diffusion is the operative sintering mechanism. An assessment of these models is made through comparison with measurements on pure mullite compacts (Fig. 14(b)).42 Consistent power-law scalings of modulus and toughness with time are obtained for aging times up to 103 h. Furthermore, upon extrapolation, the modulus and the toughness are predicted to increase by only 10% and 20%, respectively, for an additional 104 h of exposure at 12001C (typical of turbine engine components). Such extrapolations are used in estimating longterm durability of oxide CFCCs, demonstrated below. (2) Two-Phase Particle Networks The preceding models for junction growth and property changes in monophase aggregates are extended to two-phase particle mixtures (such as those comprising porous matrices in CFCCs).42 For a generic mixture of A and B particles, three junction types are present: A–A, A–B, and B–B. The aggregate properties are obtained by averaging the junction properties, weighted by the number fraction of the associated junction type. To facilitate tractable solutions, the two particle types are assumed to be the same size and arranged randomly in the mixture. A statistical analysis yields the junction fractions, f fAA ¼ X2 A (9a) fBB ¼ ð Þ 1 XA 2 (9b) fAB ¼ 2XAð Þ 1 XA (9c) where XA is the number fraction of A particles. Young’s modulus of the two-phase mixture is modeled following an approach similar to that used for monophase systems,46 with appropriate modifications to reflect differences in junction characteristics. (i) The description of junction stiffness is modified to account for the moduli of the particles on either side of the junction. In the Hertzian limit, the stiffness of dissimilar particle junctions is proportional to 2l/(11l) where lEA/EB. 49 (ii) The area of the A–B junction is assumed to be the average of the areas of the A–A and B–B junctions. This result is expressed in non-dimensional form as ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi aAB=aBB ¼ ð1 þ ZÞ=2 p where Z is a junction area ratio, defined by Z(aAA/aBB) 2 . (iii) The modulus is determined from the arithmetic mean of junction stiffness, weighted by the respective number fractions, given by Eq. (9). The result is: E EB ¼ X2 Al ffiffiffi Z p þ 2XAð Þ 1 XA ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð Þ 1 þ Z 2 r 2l 1 þ l þ ð Þ 1 XA 2 (10) Upon comparing with numerical simulations, this result is found to be accurate in the domain in which the sinterable phase comprises r40% of the total. An analogous model for the toughness of a mixed aggregate is obtained when the contribution from each junction type is assumed to be proportional to Gj(a/R) 2 and the toughnesses of the different types of junctions are then weighted by their respective number fractions Eq. (9). The result is G GM ¼ X2 ACZ þ 2XAð Þ 1 XA 1 þ Z 2 1 þ C 2 þ ð Þ 1 XA 2 (11) where GAA, GBB, and GAB are the junction toughnesses; CGAA/ GMM; and GAB is taken to be the average of GAA and GBB. The models are assessed by comparison with experimental measurements on the mullite/alumina system, for which l 5 2 (Fig. 15).42 The unknown parameters are and C. Fitting the modulus measurements yields a junction area ratio Z371: consistent with the expectation that the alumina-containing junctions should sinter more rapidly than those with only mullite. Then, upon fitting the toughness measurements, the inferred toughness ratio is C 5 1.070.3. The implications are twofold: (i) the toughnesses of the three junction types are similar to one another, and (ii) the increase in aggregate toughness with alumina content is due largely to the increase in the average junction area. (3) Precursor-Derived Two-Phase Networks When the binder phase is produced by a precursor route, the resulting topology is markedly different. For modeling purposes, the topology is represented by one of two limiting idealizations. In both, the major phase is treated as a contiguous network of uniform particles, radius R, and with average Fig. 15. Effects of composition and aging time on (a) Young’s modulus and (b) toughness. The solid lines represents model predictions (Eqs. (10) and (11)). Adapted from Fujita et al. 42 November 2006 Oxide Fiber Composites 3317
3318 Journal of the American Ceramic Society--Zok the same value of a. Operationally, this estimate is obtained (a) by replacing Ep in Eq (13)with the volume-weighted average of the moduli of the two solid phases, yielding the result [a=0112k E=04(1+14x2)[EPUP+EB(1-UB 4) where I-Po+vB and EB is the modulus of the binder. Upon combining Eqs.(13H (15), the estimated modulus of the two-phase system becomes E (1-Po)+(EB/Ep)VB EP (1-Po)+VB measurements predictions An analogous approach yields estimates of toughn the properties of the two phases are the same the results from DEM simulations(Sidebar A)are well described by =12a2 When adapted to the general case in which the properties of the two phases differ, the toughne ome 2VB+2 o=0.1,B=12 3(1-po) Comparisons between the model predictions and the exper- imental measurements for the mullite-alumina system are plot o ted in Fig. 16. When the parameters are selected to be b= 1.2 Alumina concentration, Va(%) and ao/R=O1, the model provides a good fit to the modulus data. In contrast, when they are taken as B= l and ao R=O, the nd model predictie modulus of the pure mullite is erroneously predicted to be 0 and Youngs modulus and (b) toughness. The material is made of a mullite the rate of increase with volume fraction is underestimated. For article network, strengthened by precursor-derived alumina(Adapted from Fujita et al. are l modeling the toughness, with Ti the only unknown parameter junction radius, ao. In one limit, the binder phase is a porous Upon fitting the data( Fig. 16), the junction toughness is inferred homogeneous continuum occupying all available space in the to be li=4 J/m*: only slightly greater than that of pure mullite interstices of the particle network. This idealization is applicable (i=3J/m) to the as-processed aluminosilicate matrices, wherein the silica glass exists as a contiguous nanoporous phase within the alu- (4) Implications for Crack Deflection mina particle network. In the other, the binder is modeled Once calibrated, the preceding models are coupled with the This is the preferred representation for the mullite- tion, as manifest in the parameter 2. The results are po lla coating, thickness h, on the particle surfaces(inset of analysis in Section Ill to assess the propensity for crack defle nalysis Nextel720 fibers In this case, crack deflection is predicted over From geometry, the normalized net junction radius a is given the entire range of compositions(0%40% alumina) and aging times(to 1000 h). As a complementary representation, a subset of these results is plotted in Fig. 12(a)(assuming @= 1). Here, 2β-VB+3 again, the property combinations lie within the crack deflection 2=R+h~V3(1-Po) ( 12) domain. Moreover, the interface sliding stress in these systems is low, typically <10 MPa (Sidebar B), re-affirming that mullite-alumina mixtures are good candidates for use in oxide composites The critical aging time le for crack penetration is obtained by Co=Mo/(R+h); and B is a non-dimensional parameter that ac- extrapolating the predictions in Fig. 17(a)to 2=0=1.It counts for preferential binder accumulation at the particle junc nges from 4000 h for mixtures of 60% mullite 40% alumina tions(= l for uniform coatings). For the case in which the roperties of the two phases are the same, computer to 60000 h for pure mullite, the latter being comparable with the targeted service lives of CFCC components. With knowledge of ions based on the DEM of Youngs modulus are accur- the activation energy of the sintering mechanism, the model can escribed by the empirical equation be readily extended to other temperatures A similar assessment is made of the mullite particle networks =0.4x(1+14x strengthened by precursor-derived alumina, again assuming the C-inforcements to be Nextel 720 fibers. The results are plotted n Fig. 17(b). Here 2 decreases with increasing alumina con where Ep is the modulus of the solid particles. More genera entration, VA, and eventually falls below the critical value. when the properties of the two phases differ, the modulus of the oRl, at VA9%. This point is expected to mark the onset of wo-phase aggregate is estimated from the weighted average of crack penetration into the fibers and a significant loss in damage the moduli of the two monophase aggregates, evaluated at
junction radius, ao. In one limit, the binder phase is a porous homogeneous continuum occupying all available space in the interstices of the particle network. This idealization is applicable to the as-processed aluminosilicate matrices, wherein the silica glass exists as a contiguous nanoporous phase within the alumina particle network. In the other, the binder is modeled as a uniform coating, thickness h, on the particle surfaces (inset of Fig. 16). This is the preferred representation for the mullite– alumina system (Fig. 13) and forms the basis for the ensuing analysis. From geometry, the normalized net junction radius a is given by40 a a R þ h ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2b2 VB 3 1ð Þ po þ a2 o s (12) where 1po is the volume fraction of the particulate phase; VB is the volumetric concentration of the precursor-derived binder; aoao/(R1h); and b is a non-dimensional parameter that accounts for preferential binder accumulation at the particle junctions (b 5 1 for uniform coatings). For the case in which the elastic properties of the two phases are the same, computer simulations based on the DEM of Young’s modulus are accurately described by the empirical equation E EP ¼ 0:4a 1 þ 14a3 (13) where EP is the modulus of the solid particles. More generally, when the properties of the two phases differ, the modulus of the two-phase aggregate is estimated from the weighted average of the moduli of the two monophase aggregates, evaluated at the same value of a. Operationally, this estimate is obtained by replacing EP in Eq. (13) with the volume-weighted average of the moduli of the two solid phases, yielding the result E ¼ 0:4a 1 þ 14a3 ½ EPuP þ EBð Þ 1 uB (14) where uP ¼ 1 po 1 po þ VB (15) and EB is the modulus of the binder. Upon combining Eqs. (13)– (15), the estimated modulus of the two-phase system becomes: E EP ¼ 0:4a 1 þ 14a3 ð Þþ 1 po ð Þ EB=EP VB ð Þþ 1 po VB (16) An analogous approach yields estimates of toughness. When the properties of the two phases are the same, the results from DEM simulations (Sidebar A) are well described by G Gj ¼ 12a2 (17) When adapted to the general case in which the properties of the two phases differ, the toughness becomes G Gj ¼ 12 b2 2VB 3 1ð Þ po þ a2 o (18) . Comparisons between the model predictions and the experimental measurements for the mullite–alumina system are plotted in Fig. 16. When the parameters are selected to be b 5 1.2 and ao/R 5 0.1, the model provides a good fit to the modulus data. In contrast, when they are taken as b 5 1 and ao/R 5 0, the modulus of the pure mullite is erroneously predicted to be 0 and the rate of increase with volume fraction is underestimated. For consistency, the same (former) values of b and ao/R are used for modeling the toughness, with Gj the only unknown parameter. Upon fitting the data (Fig. 16), the junction toughness is inferred to be Gj 5 4 J/m2 : only slightly greater than that of pure mullite (Gj 5 3 J/m2 ). (4) Implications for Crack Deflection Once calibrated, the preceding models are coupled with the analysis in Section III to assess the propensity for crack deflection, as manifest in the parameter S. The results are plotted in Fig. 17(a) for mullite–alumina particle mixtures combined with Nextelt 720 fibers. In this case, crack deflection is predicted over the entire range of compositions (0%–40% alumina) and aging times (to 1000 h). As a complementary representation, a subset of these results is plotted in Fig. 12(a) (assuming o 5 1). Here, again, the property combinations lie within the crack deflection domain. Moreover, the interface sliding stress in these systems is low, typically o10MPa (Sidebar B), re-affirming that mullite–alumina mixtures are good candidates for use in oxide composites. The critical aging time tc for crack penetration is obtained by extrapolating the predictions in Fig. 17(a) to S 5 o 5 1. It ranges from 4000 h for mixtures of 60% mullite–40% alumina to 60 000 h for pure mullite, the latter being comparable with the targeted service lives of CFCC components. With knowledge of the activation energy of the sintering mechanism, the model can be readily extended to other temperatures. A similar assessment is made of the mullite particle networks strengthened by precursor-derived alumina, again assuming the re-inforcements to be Nextelt 720 fibers. The results are plotted in Fig. 17(b). Here, S decreases with increasing alumina concentration, VA, and eventually falls below the critical value, o1, at VA9%. This point is expected to mark the onset of crack penetration into the fibers and a significant loss in damage tolerance. Fig. 16. Summary of measurements and model predictions of (a) Young’s modulus and (b) toughness. The material is made of a mullite particle network, strengthened by precursor-derived alumina. (Adapted from Fujita et al. 40). 3318 Journal of the American Ceramic Society—Zok Vol. 89, No. 11