Materials Science and Engineering A244 (1998) 58 – 66 Model-based optimization of consolidation processing Ravi Vancheeswaran, Haydn N.G. Wadley * Intelligent Processing of Materials Laboratory, Department of Materials Science and Engineering, Uni6ersity of Virginia, Charlottes6ille, VA 22901 -2442, USA Abstract The performance of fiber reinforced titanium matrix composites (TMC) made by consolidation of spray deposited monotapes is strongly influenced by the processing conditions used. This high temperature consolidation step must simultaneously increase the relative density while minimizing fiber microbending/fracture and the interfacial reaction product layers at the fiber–matrix interface. These three microstructural variables have conflicting dependencies upon the consolidation process variables (temperature, pressure and time), and it has been difficult to experimentally identify process pathways that lead to composites of acceptable quality (where the fiber damage and the reaction layer thickness are kept below some bounds, while matrix porosity is eliminated). Here, models for predicting the microstructure’s dependence upon process conditions (i.e. the time varying temperature and pressure) are combined with consolidation equipment dynamics to simulate the microstructure evolution. We then introduce the idea of process failure surfaces and show how a model predictive control algorithm, is able to design ‘locally’ optimal process cycles that minimize fiber damage, reaction product layer thickness and porosity. As an example, we explore the planning of process schedules that process failure surfaces for several TMC systems. © 1998 Elsevier Science S.A. All rights reserved. Keywords: Model-based optimization; Consolidation processing; Titanium matrix composites 1. Introduction Metal matrix composites are attractive alternative materials to monolithic titanium and nickel base alloys in gas turbine engines because of their higher specific stiffness, strength, crack growth resistance, creep rupture life and fatigue endurance [1 – 6]. Those based upon conventional or intermetallic titanium alloys have attracted considerable attention because they allow a radical redesign of high performance aircraft engines [7,8]. Many laboratory scale manufacturing processes exist for the manufacture of these materials; however, they all involve a two step sequence that seeks to reduce the aggressive chemical reaction between all liquid titanium alloys and most candidate fibers. First a monotape is manufactured by plasma spray deposition, slurry casting or physical vapor deposition coating of colinear fiber arrays [9 – 14]. The tapes are then cut to shape, stacked to create an appropriate fiber architecture and consolidated by either vacuum hot pressing (VHP) or hot isostatic pressing (HIP) [15]. Ideally, this * Corresponding author. Tel.: +1 804 9825670; fax: + 1 804 9825677; e-mail: [email protected] 0921-5093/98/$19.00 © 1998 Elsevier Science S.A. All rights reserved. PII S 0 9 2 1 - 5 0 9 3 ( 9 7 ) 0 0 8 2 6 - 5 results in a near-net-shape composite component with no residual porosity (to avoid premature matrix failure during static or fatigue loading), minimal fiber microbending/fracture (to avoid degrading composite strength and creep/fatigue endurance) and a limited thickness of reaction product at the fiber–matrix interface (to avoid increases in interfacial sliding stress or loss of fiber strength). Unless a ‘goal state’ combination of these microstructural attributes (i.e. pore concentration or equivalently, relative density, fiber microbending stress/fracture and reaction product layer thickness), the competitive advantage of these composites over other materials will not be realized. A dynamic model has been developed that simulates the evolution of these three microstructural parameters for any hypothetical process cycle (i.e. temperature and pressure schedules) for a lay-up of plasma sprayed monotapes [15]. It reveals that the goals for each microstructural parameter have conflicting dependencies upon the variables of the process (temperature and pressure). For example, densification (the elimination of matrix porosity) of plasma sprayed monotape preforms is most quickly accomplished by consolidating at the highest temperature and pressure. But fiber microbend- R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 59 Fig. 1. Simulation of the evolution of the microstructural states using the process schedule shown in (b). ing/fracture can only be minimized by consolidating at high temperatures while applying the pressure very slowly (i.e. using long duration cycles), and reaction product layer thickness can only be reduced by shortening the high temperature exposure time (i.e. by using short, cold consolidation cycles). Dynamic models of this type enable a mapping from the process state space (with temperature and pressure) to a microstructure state space (in this case with axes of relative density, fiber fracture density and reaction layer thickness). An example is shown in Fig. 1 for the Ti–6Al–4V/SCS-6 system (refer to [15] for the monotape geometry and matrix/fiber thermophysical properties used for the calculation). The successful consolidation processing pathway is seen to be a nonlinear mechanism for transferring the microstructural states from an initial configuration to a final ‘goal’ state. From this viewpoint, it no longer matters if pressure or temperature are applied first, or if the full heating rate or temperature/pressure capacities of the equipment are utilized. All that is important is to find a trajectory in the process variable space (Fig. 2) that takes the lay-up from its initial state to the goal state. The dynamic simulations could be used as a trial and error approach to find process pathways that accomplish this. For the more ‘processible’ composite systems like Ti–6Al–4V/SCS-6, several potentially successful cycles have been identified in this way [15]. However, the processability of this system originates from the low resistance to creep of the matrix at typical processing temperatures. Efforts to extend the upper service temperature by using more creep resistant alloys rapidly degrades the processability of the composite, and it becomes increasingly difficult to find acceptable process 60 R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 schedules, or to be sure at the outset of a materials development campaign, that one even exists. Here the models for microstructural evolution are coupled with a model predictive control (MPC) algorithm to develop a methodology for calculating locally optimal process pathways. Control design methods based on the MPC concept have been studied and have found wide acceptance in industrial applications [17,18]. MPC techniques use a multistep predictor and a receding horizon philosophy to predict the optimal set of control variables using the present knowledge of the process and candidate planner actions. This is especially useful because the microstructural variables like relative density, fiber fracture and reaction layer thickness are irreversible processes. Therefore the planner needs to be ‘conservative’, since the consequences of a ‘wrong’ decision by the planner can make the microstructural states overshoot the desired goal state thereby rendering the composite unusable. However, if extensive use is made of predictions well into the plant rise time, disastrous overshoots can be avoided. The plant can be modeled as two distinct parts, the machine model (i.e. HIP/VHP) and the material (microstructural evolution) model [15]. The interconnection and interaction between these two models and the automatic planner is shown in Fig. 3. Mathematical models capturing the dynamic behavior of the machine and the material take the form of coupled systems of nonlinear ordinary differential equations [15,19–23]. Fig. 2. Schematic of the reachable set from a given point in state space Fig. 3. Block diagram of the control system. 2. The path planner A simulated response of a matrix alloy Ti–6Al–4V/ SCS-6 silicon carbide fiber titanium matrix composite (TMC) to a commonly used ‘ramp and soak’ temperature and pressure schedule is shown in Fig. 1. This particular schedule fails to achieve full density, causes about nine fiber fractures per meter of fiber and results in the growth of :0.3 mm thick reaction layer at the fiber–matrix interface. It would result in a composite of poor mechanical performance [2] due to the large number of fractures. The optimization problem that we pose is schematically illustrated in Fig. 2. The objective is to reach a goal state at the completion of the process. At any moment in the consolidation process, the material’s state can be represented as a point in a relative density, fiber deflection, reaction layer thickness space. It will have reached this point along a path originating in the lower left corner of the space. The consolidation process is ‘irreversible’1 hence all future feasible paths are enclosed in a conically shaped region, called the reachable set, originating from the current point. If the goal state is not inside this region, then it is impossible to process the composite to achieve that goal. One seeks to find the optimum process path in state space that will take the current material state to the desired goal state. Ideally the indicated planning problem would be tackled by computing the exact reachable set of microstructural states for a given set of process states. Then a complete path can be found in a single shot beginning at the starting conditions and ending at the 1 The relative density, fiber fracture density and reaction zone thickness are non-decreasing functions of time. R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 goal state. However, solving the problem this way is very difficult because the problem is computationally intensive (NP Hard). Instead, as described below, a predictive planning method is used where a series of small steps are taken toward the goal state and at each step a viable approximation to the exact reachable set is computed. A block diagram showing the planner is presented in Fig. 3. The path planning system accepts as input a vector signal, xg, for the goal state that contains desired final values for the relative density, the fiber deflections of a set of representative unit fiber cells and the reaction zone thickness. The planning system also accepts as input the state vector, xh, giving measured or estimated values of the microstructural variables (density, fiber deflection and reaction layer thickness), and the process environment vector, h, giving the current machine temperature and pressure. The planner calculates appropriate temperature and pressure slews and commands the HIP or VHP, thereby constructing a temperature and pressure schedule that will steer the current material state to the desired goal state, or keep the current material state as close to the goal state as possible. Mathematically, therefore, the planner’s task is to calculate u(t) such that xh (t) xg (see [16] for details). 61 as a Taylor series and retaining only first order terms. The discontinuity (plasticity) is not included in this step. Letting x̃= xh − x ch and h̃= h− h c denote deviations from the current material state and machine environment, respectively, the linearization of the vector field about x ch, h c can be written dx̃h = Ahx̃h + Bhh̃+ fh. dt (2) 3.3. Cascading Once the linearization of the material model is calculated for a given operating point, x ch, h c, the overall material-machine model can be written dx̃ = Ax̃+ Bh̃ + f. dt (3) Numerically integrating the system of differential Eq. (3) by the forward Euler algorithm leads to the recursive relationship x̃((k +1) t)=(I+ tA)x̃(kt)+ tBu(kt)+ tf (4) where t is the integration time step. Solving the difference Eq. (4) starting from x̃(0)= 0 (recall that x̃ represents deviations from the current operating point) yields n−1 x(nt)= % t(I+ tA)n − 1 − k(Bu(k)+ f ). 3. Path planner design 3.1. Normalization To begin, a change of variables is made in the material model so that each component of the states xh and inputs h lie in the interval [0, 1]. This normalization is done to provide good numerical conditioning in the later stages of the path planning design where a convex program is solved. Transformations of the density, the unit cell deflections, the reaction zone growth, temperature and pressure are defined as follows Dn = = D −D0 i 6i r T −Tmin 6 n = i rn = T = P 1− D0 6 crit rcrit n Tmax −Tmin n P − Pmin Pmax − Pmin (5) k=0 (1) where the n subscript denotes the normalized variable. The parameters 6 icrit and rcrit are defined as the maximum (worst case) values that these states are allowed to have. The critical value for the reaction zone is chosen to be 0.6 mm and for the deflection states it is the value that causes a failure probability of 0.1. 3.2. Linearization A linear approximate model about the current operating point is computed by expanding the vector-field Hence it is seen that at an operating point, [x ch, h c], an approximation to the reachable set is given by the range of the right hand side of Eq. (5) when subjected to the machine constraints. 3.4. Con6ex sol6er The relevant planning task is steering the material state, xh, to the goal state xg, and so a logical choice of convex objective is the Euclidean distance between x(nt) and xg. This leads to a quadratic program which can be efficiently solved. The planner minimizes a weighted sum of the distance between the goal state and the projected future material states (based on the linearization) given by Nl k (x(kt)− xg )TW(x(kt)− xg )22 N k=1 l % Nl k = % k = 1 Nl ÆW D × (Dg − Dl )k))2Ç ÃW 6 × (6g − 6 il (k))2 Ã . ÈW r × (rg − rl (k))2 É (6) The planner attempts to drive the relative density to the desired final value while not fracturing fibers or accruing a reaction product, and the weightings, WD, 62 R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 Table 1 Goal states for the different material systems Composite 8f rcrit (mm) D 6i r Ti– 6Al – 4V/SCS-6 Ti– 24Al – 11Nb/SCS-6 0.1 0.1 1.0 1.0 0.999 0.999 0.47 6crit 0.5 6crit 0.6 rcrit 2.0 rcrit W6 and Wr in the objective function allow the trade-off between these competing goals to be made quantitatively and precisely. Note that the method in fact allows cross-weightings between the states because it only requires that W= W T \0 and not W to be a diagonal matrix. Once optimal values of temperature and pressure slew rates are found, they are used as inputs to the full machine and material (nonlinear) models (which now includes the effects of plasticity) which are then integrated forward in time by a higher-order method (e.g. Runge–Kutta) to give a new operating point which includes the discontinuities of the vector field (plasticity). A linearization is then found about the new operating point and the entire planning design process is repeated until either the goal state is reached or a final time limit condition is reached. 4. Results To explore the optimization approach, we choose a goal state density of 0.999, fiber cell deflections that result in a cumulative probability of fracture B 0.1 (thereby holding the fiber fracture density below a reasonable bound of 1 fracture per meter), and a reaction layer thickness lie below 0.6 mm for a Ti–6Al–4V matrix system and 1 mm for the Ti – 24Al – 11Nb system. For weights, the selections, WD =10, W6 = 10, and Wr = 10 were chosen. Hence, roughly, the algorithm assumes an equal concern for densification, fiber/matrix reaction and fiber microbending/fracture. A sampling interval of 1 min was used (t= 60 in Eq. (5)), and a look ahead horizon for the planner of 10 min (N= 10 in Eq. (5)). Therefore for each 1 min interval, the planner solves a quadratic program of 20 variables and about 40 constraints. This takes a very short time ( : 1 s) on a multi-user Sun SPARC-20™ station. A higher weighting is given to states that are further away in time from the operating point, thereby making the planner aggressive in reducing process time to consolidate the matrix, since the process is completed either when the density reaches 0.999 or the time elapsed is 10 h. An extra constraint was placed on the planner to constrain the pressure rate at zero till the first fiber deflection cell was created. The extra constraint caused the planner to serially apply the temperature before applying pressure. Since the critical fiber unit-cells of interest are not created during the early stages of densification, and therefore if this constraint is not added, the planner makes the mistake of increasing the pressure since it does not know of the existence of the fiber cells. When the fiber cells are created, the planner is not able to correct the pressure rate soon enough, to reduce the deflection of the fiber cells beyond their critical limit and this causes fracture (see Table 1). 4.1. The Ti– 6Al– 4V/SCS-6 system A path for the microstructural variables to the given goal state for the Ti–6Al–4V/SCS-6 composite monotape is computed. Fig. 4 shows a result where we simulate this system with a nominal machine (machine 1) see Table 2. In this case, the goal state density was reached and the goal states for fiber deflections and reaction layer thickness were held below their critical values. The planner computed a very interesting and intuitive input schedule. First the temperature is ramped at the maximum rate till the maximum machine temperature was reached. At this time, the planner ramped the pressure at the maximum rate till the bounds on the deflections of the fiber cells were reached. Then the planner reduced the pressure rate so as to make sure that the deflection bound is not violated while the densification rate continued to increase. It was during this period that the pressure profile underwent cycling or hunting. It is worth mentioning that the density continued to increase while the deflection of the fiber cells was kept constant or decreased. In the latter stages of the process, the fiber deflection mechanism is not important anymore (see [15]), so it then becomes essentially a race to the finish between the densification and the reaction layer kinetics. If the reaction layer reached the critical state first, then the temperature was ramped down and this stopped the densification of the composite. This schedule is nearly time optimal since the planner kept a constraint active at all time. 4.2. The Ti– 24Al – 11Nb/SCS-6 system The Ti–6Al–4V matrix was then replaced with a more creep resistant matrix, Ti–24Al–11Nb. A typical industrial input schedule was simulated and compared with a path planned simulation using HIP M/C 1. The final states for these simulations have been tabulated in Table 3. R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 63 Fig. 4. Simulation showing evolution of the microstructural states for the Ti – 6Al – 4V/SCS-6 system using a path-planned process schedule. On comparison of the results for the direct simulation and the path-planned simulation, one finds that in both cases the desired reaction layer thickness of 1 mm is not met. The path-planned result provides better performance for the fiber fracture density while trading off the reaction layer thickness performance. In the path-planned simulation, if the reaction layer thickness requirement is made any smaller, the planner drops the temperature to constrain the growth of the reaction layer. An interesting effect of this problem is that when the planner drops the temperature, applying high pressures does little to increase density and thus the planner drops the pressure. 4.2.1. Machine limitations The optimal planning approach can be used to determine what states are reachable. Fig. 5 shows the boundary that separates those states for which optimal process trajectories exist and those for which no trajectory exists. It shows that for a fixed relative density at process completion, fiber fracture and reaction layer thickness are inversely related. The best composite would lie at the origin of Fig. 5 and we can explore how the limitations affect this window. Four sets of numerical experiments have been performed using enhanced temperature and pressure HIP machines (see Table 2) to find what effect this has on the bounds of the reachable set. It is interesting to note that regardless of machine, as the goal state density is reduced, the performance of the reaction layer thickness and fiber fracture density are improved. As more stringent reaction layer thickness goal states are chosen, the curves become asymptotic with the y-axis while the limitation of how many fibers are broken is completely dependent on the pressure ramp rate. One can think of reaction layer thickness as a ‘fuel’ constraint and only a limited thermal exposure (as monitored by the reaction layer thickness) is allowed during consolidation. If this constraint is violated, the computed temperature trajectory drops off, and severely limits further densification. The nominal HIP was found to have the worst performance. The performance of the planner was bet- R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 64 Table 2 Physical limits for the different HIP machines HIP M/C T: min (°C min−1) T: max (°C min−1) P: min (MPa min−1) P: max (MPa min−1) Tmax a (°C) Pmax (MPa) 1 2 3 4 −20.0 −30.0 −20.0 −30.0 20.0 30.0 20.0 30.0 −1.0 −1.0 −2.0 −2.0 1.0 1.0 2.0 2.0 825.0 825.0 825.0 825.0 100.0 100.0 100.0 100.0 a The maximum temperature for the Ti–24Al–11Nb/SCS-6 composite is 1000°C. ter when the temperature rate was increased (M/C 2). This was because if the temperature is increased at a faster rate, the total (integrated) thermal exposure that the composite was exposed to during consolidation was reduced. The effect of increasing the pressure rate (M/C 3) had an even more significant effect on the consolidated final states of the composite. Pressures in the HIP are ramped at a much slower rate than temperatures. Following the same trend, M/C 4 performed better than all the other HIPs. Even with these enhanced HIP machines an acceptable goal state fracture and reaction layer thickness were never reached. For this system, fiber deflection and fracture becomes less of a problem since the material creeps readily, but reaction zone growth becomes critical for processability. The density and reaction layer thickness requirements were met using the enhanced pressure HIP and the enhanced temperature and pressure HIP but at the expense of breaking 10 fibers m − 1. To make this system processible, either a better interfacial coating (with slower reaction kinetics) is needed, or a stronger fiber must be developed. For example, if the fiber reference strength were increased from 4.5 to 6.2 GPa it becomes possible to reach the composite’s goal state with M/C 4. fibers are easily deflected and damaged by the application of pressure. Therefore the planner ramps the temperature as fast as possible while holding the pressure at 1 MPa until the matrix starts to creep rapidly. Once the temperature reaches the maximum temperature allowable, the pressure is aggressively ramped till the critical fiber deflections are in sight. At this time the pressure rate is reduced to trade off the densification achieved while not increasing the fiber deflections beyond the critical values. After the density reaches 0.92 (the critical transition density between stage 1 and stage 2 densification), it becomes a ‘race’ for the planner to achieve the highest possible relative density without overshooting the upper bound on the reaction zone. This strategy is nearly time-optimal as the densification rate is maximized while the fracture rate is held acceptably low. A short cut to a near optimal process can be found by viewing consolidation as a process in which fiber fracture and fiber–matrix reaction are failure mechanisms for the process. For processes of the type shown in Fig. 6, it is possible to plot surfaces of constant reaction layer thickness and fiber fracture density in a three-dimensional (Prate − T− t) process space. we have chosen failure surfaces corresponding to 1 fracture m − 1 5. Discussion The fundamental trade-off between the final composite density, the number of fibers fractured and the reaction layer thickness is evident in all path-planned runs. Evidently, consolidation can be naturally divided into two distinct time intervals. The first interval is roughly the time taken to densify the matrix to a relative density of 0.92. Here the fiber deflections and thus number of fractures of the composite are the critical parameters. It has been shown previously [15] that in the beginning while the temperature is low, Table 3 Final states for the direct and path planned simulation Simulation D Nf (m−1) r (mm−1) Direct Path planned 0.999 0.999 10.1 0.15 1.17 1.9 Fig. 5. The bounds on the process window for the Ti –24Al–11Nb/ SCS-6 system for a densification goal state of D =0.995 using different HIP machines R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 65 Fig. 6. Microstructure failure surface in process space for a Ti – 6Al – 4V/SCS-6 composite. and r =0.47 mm. When the optimal trajectory (shown by the black line) is superimposed, the optimal process is seen to have found a way to avoid the fracture and reaction layer failure surfaces while still densifying the composite completely. We see that the trajectory is initially pressure rate limited by the fracture surface. However, after : 200 min of consolidation, the Prate bound disappears and is replaced by the need to avoid a reaction layer failure. The relatively low consolidation temperature and reactivity of the Ti–6Al–4V matrix enables this to be accomplished. Thus, the optimal process is seen to hug the first failure surface encountered and to then jump to follow the second failure surface until complete densification is accomplished. By mapping the failure surfaces in a process trajectory space, a convenient graphical method can be used to estimate a near optimal process. 6. Conclusions A model predictive planner has been developed for guiding the microstructural evolution in TMCs to a predefined goal state during consolidation. The path planning method uses constantly updated linearizations and constrained convex optimization to compute planner actions. For long look-ahead times and short integration time steps, sequentially local reachable goals are very nearly global reachable goals. It identifies a viable process path that reaches the desired goal state for a Ti–6Al–4V/SCS-6 composite system while showing that no such path exists for a similar composite with a Ti–24Al–11Nb matrix. Reachable process windows have been constructed by this method for each composite system. The path planning method can also be used to redesign the processing equipment, the fiber’s reaction inhibiting coating, the fiber’s strength or the monotape surface roughness so that difficult material systems, such as the Ti–24Al–11Nb matrix composite, can be successfully processed. A simple graphical method is also used for the estimation of near optimal process trajectories for processes where the microstructural states have conflicting dependencies upon the process variables. Acknowledgements We are grateful to R. Kosut for helpful discussions about this research and to D. Elzey for suggestions in implementing the models. This work has been funded by DARPA through a contract with Integrated Sys- R. Vancheeswaran, H.N.G. Wadley / Materials Science and Engineering A244 (1998) 58–66 66 tems, Santa Clara, CA (Dr Anna Tsao, Program Manager). References [1] D.M. Elzey, J.M. Kunze, J.M. Duva, H.N.G. Wadley, in: L. Gibson, D.J. Green, K. Sieradzki (Eds.), Mechanical Properties of Porous and Cellular Materials, vol. 207, MRS, Pittsburgh, PA, 1991, p. 109. [2] J.M. Duva, W.A. Curtin, H.N.G. Wadley, Acta Metall. Mater. 43 (3) (1995) 1119–1126. [3] M.Y. He, A.G. Evans, W.A. Curtin, Acta Metall. Mater. 41 (3) (1993) 871 – 878. [4] W.A. Curtin, J. Am. Ceram. Soc. 74 (11) (1991) 2837 – 2845. [5] Z.Z. Du, R.M. McMeeking, Creep models for ceramic matrix composites with ling brittle fibers, unpublished paper, 1994. [6] D.M. Elzey, J.M. Duva, H.N.G. Wadley, in: F.H. Froes, J. Storer (Eds.), Recent Advances in Titanium Metal Matrix Composites, TMS, Rosemont, CA, 1995. [7] J.R. Stephens, in: R.B. Bhagat, A.H. Clauer, P. Kumar, A.M. Ritter (Eds.), Metal and Ceramic Composites: Processing, Modelling and Mechanical Behavior, TMS, Warrendale, PA, 1990, p. 3. [8] D.L. Anton, D.H. Shah, in: C.T. Liu, A.I. Taub, N.S. Stoloff, C.C. Koch (Eds.), MRS Symp. Proc. 133, MRS, Pittsburgh, PA, 1989, p. 361. [9] E.S. Russell, Thermomechanical effects in plasma-spray manufacture of MMC monotapes, in: E.A. Thornton (Ed.), Thermal Structures for High Speed Flight, Progress in Aeronautics and Astronautics 140, AIAA, Washington DC, 1992. . . [10] N.A. James, D.J. Lovett, C.M. Warwick, in: S.W. Tsai, G.S. Springer, Composite: Design, Manufacture and Application, ICCM VIII, SAMPE, 2 (19), 1992, p. 11. [11] P.K. Brindley, in: N.S. Stoloff, C.C. Koch, C.T. Liu, O. Izumi (Eds.), High Temperature Ordered Intermetallic Alloys II, vol. 81, MRS, Pittsburgh, PA, 1987, p. 419. [12] H.C. Hartwick, R.C. Cordi, in: D.L. Anton, P.L. Martin, D.B. Miracle, R. McMeeking, Intermetallic Matrix Composites, vol. 194, MRS, Pittsburgh, PA, 1990, p. 65. [13] P.G. Partridge, C.M. Ward-Close, Int. Mater. Rev. 38 (1993) 1. [14] H.E. Deve, D.M. Elzey, J.M. Warren, H.N.G. Wadley, in: P. Vincenzi (Ed.), Advances in Science and Technology, 7, Advanced Structural Fiber Composites, Techna, Florence, 1995, pp. 313 – 327. [15] R. Vancheeswaran, D.M. Elzey, H.N.G. Wadley, Acta. Metall. Mater. 44 (6) (1996) 2175 – 2199. [16] R. Vancheeswaran, D.G. Meyer, H.N.G. Wadley, Optimizing the Consolidation of Titanium Matrix Composites, unpublished research. [17] C.E. Garcia, D.M. Prett, M. Morari, Automatica 25 (3) (1989) 335 – 348. [18] D.W. Clarke, C. Mohtadi, P.S. Tuffs, Automatica 23 (1987) 137 – 160. [19] R. Gampala, D.M. Elzey, H.N.G. Wadley, Acta Metall. Mater. 42 (9) (1994) 3209 – 3221. [20] D.M. Elzey, H.N.G. Wadley, Acta Metall. Mater. 41 (8) (1994) 2297 – 2316. [21] Z. Qian, J.M. Duva, H.N.G. Wadley, Acta Metall. Mater. 44(12) (1996) 4815 – 4824. [22] R. Gampala, D.M. Elzey, H.N.G. Wadley, Acta Metall. Mater. 44 (4) (1996) 1479 – 1495. [23] D.M. Elzey, H.N.G. Wadley, Acta Metall. Mater. 42 (12) (1994) 3997 – 4013.
© Copyright 2026 Paperzz