Modelling the Precipitation of Al 3 X Dispersoids in Aluminium Alloys and their Effect on Recrystallization

Transcription

1 Materials Science Forum Vol. 55 (27) pp online at (27) Trans Tech Publications, Switzerland Modelling the Precipitation of Al 3 X Dispersoids in Aluminium Alloys and their Effect on Recrystallization J. D. Robson 1, a, P. B. Prangnell 1,b, B. J. McKay 2,c and C. P. Heason 3,d 1 School of Materials, University of Manchester, Grosvenor Street, Manchester, UK 2 Department Metallurgie, Institut für Gießereikunde, Montanuniversität Loeben, Austria 3 Corus Swinden Technology Centre, Rotherham, South Yorkshire, S6 3AR, UK a joseph.robson@manchester.ac.uk, b philip.prangnell@manchester.ac.uk, c brian.mckay@mu-leoben.at, d chris.heason@corusgroup.com Keywords: Dispersoids, 7xxx aluminium alloys, Scandium, Modelling, Recrystallization Abstract. A combined model is presented that predicts the non-uniform distribution of Al 3 X dispersoid particles in commercial aluminium alloys containing zirconium and scandium and uses these predictions as inputs to a simple recrystallization model. The recrystallization model relies on knowledge of the stored energy in the sub-structure after deformation and this has been measured using electron backscattered diffraction (EBSD) techniques. The recrystallization model is based on the concept that partial recrystallization results from the non-uniform distribution of dispersoid particles due to their precipitation from a segregated cast structure. The model has been used to devise an improved homogenization treatment for AA75, which uses an isothermal hold during heat up to maximize dispersoid nucleation. It has also been applied to predict the effect of scandium additions on recrystallization, investigate the factors that control the through thickness variation in recrystallized fraction, and interpret the results of experiments where the effect of strain rate have been studied. Introduction In many wrought aluminium products it is desirable to maintain an unrecrystallized grain structure after thermomechanical processing to optimize properties [1]. To achieve this, dispersoid forming elements are used to precipitate a stable distribution of fine particles that suppress recrystallization by grain boundary pinning. Zirconium is a commonly added dispersoid forming element in modern high strength wrought aluminium alloys, precipitating as fine L1 2 Al 3 Zr particles during homogenization heat treatment [2]. Although this phase is metastable, these dispersoid particles are resistant to transformation, dissolution, and coarsening over a typical thermomechanical processing cycle, and are therefore effective recrystallization inhibitors. Despite the addition of zirconium, partial recrystallization often occurs during solution treatment of rolled or extruded products [2]. This has been attributed to the highly non-uniform distribution of dispersoid particles, which itself is due to the initial segregation of zirconium in the as-cast structure from which the dispersoids are precipitated [3]. Zirconium is segregated towards the dendrite centres during casting, and towards the edges its concentration is insufficient for precipitation of Al 3 Zr during homogenization, resulting in dispersoid free regions. These regions are also adjacent to the grain boundaries, which in commercial 7xxx alloys contain coarse constituent particles that have been shown to promote recrystallization through the particle stimulated nucleation (PSN) mechanism [2]. To control and suppress this partial recrystallization requires promoting dispersoid precipitation towards the dendrite edges. This can be partly achieved through optimization of the homogenization heat treatment [4], but is most effectively accomplished by adding other dispersoid forming All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of the publisher: Trans Tech Publications Ltd, Switzerland, (ID: /5/7,12:45:2)

2 46 Fundamentals of Deformation and Annealing elements. The most effective addition is scandium. This element substitutes for zirconium to form an L1 2 Al 3 (Zr,Sc) phase [5]. Scandium is segregated towards the dendrite edges after casting, and thus dispersoid precipitation in these regions is greatly enhanced. Experiments have shown that only a very small scandium level (e.g..1wt%), when added in combination with zirconium, is sufficient to completely suppress recrystallization in conventionally hot rolled 7xxx plate [5]. In recent work, other alloying additions (such as hafnium) which substitute into the Al 3 X phase have also been explored to assess their potential for enhancing recrystallization resistance [6]. To fully understand the interactions between alloy chemistry, heat treatment, and recrystallization resistance requires a large experimental programme. An alternative approach is to develop a model, which can then be used to guide alloy and heat treatment development. This paper describes the modeling approach developed at Manchester over recent years, and shows how it has been used to help understand how to optimize alloy composition and processing. The Model Precipitation Model. The dispersoid precipitation model is described in detail elsewhere [7], and only the key points will be summarized here. The model is based on the Kampmann and Wagner Numerical (KWN) method [8]. In this method, the full particle size distribution is divided into a large number of discreet size classes. The evolution of the particle size distribution is tracked during a timestep by allowing nucleation to add new particles into the size class slightly larger than the critical radius, and swapping existing particles between size classes according to whether the particles in that class are growing or shrinking. In the implementation of the model used here, particles are allowed to leave one size class and be distributed into a number of other size classes, depending on the variation in growth rate across the class. Numerical integration is performed using a second order Runge-Kutta scheme [7]. Classical theory is used to determine the nucleation rate and growth (or shrinking) rate for each size class. Interfacial compositions are corrected for the Gibbs-Thomson effect, which has the effect of naturally shrinking small particles at the expense of large ones as the matrix solute level falls (i.e. coarsening is automatically predicted in the model). In modelling the precipitation of nonstoichiometic phases with containing two (or more) diffusing species, it is necessary to solve the coupled requirement for flux balance and local equilibrium to find the interfacial composition. This will evolve with time, as the mean matrix composition changes [7]. The model can therefore predict the variation of dispersoid composition with time. Finally, in situations where precipitation is occurring from a segregated initial state (e.g. the ascast condition), multiple KWN models are run simultaneously, each corresponding to a particular initial position and composition (ranging from dendrite centre to edge). Solute is also allowed to exchange between positions, driven by diffusion. The initial composition variation in the as-cast structure is calculated using a 1-dimensional Scheil model [7]. Recystallization Model. To quantitatively estimate the effect that changing the dispersoid distribution will have on the recrystallized fraction, a simple recrystallization model has been developed. It is based on the principle that recrystallization will be suppressed in regions where the Zener pinning pressure due to the dispersoid particles exceeds the driving pressure for recrystallization due to the energy stored in the as-deformed structure. The as-deformed structures of interest (obtained after hot deformation) consist of well defined subgrains and the stored energy can therefore be related to the subgrain size and boundary energy, which can itself be determined from the boundary misorientation [9]. Recrystallization is assumed to be possible in regions satisfying equation (1):

3 Materials Science Forum Vol CV f r γ m γ s <. (1) D Where the left hand term represents the Zener pinning pressure of the coherent dispersoid particles, with V f the particle volume fraction, γ m the energy of the boundary the particles are pinning, and r the particle radius. C is a constant, with a theoretical value which depends on the approach used to model pinning, but in this work is calibrated by fitting predicted and measured recrystallized fractions for one condition. The right hand term represents the driving pressure for recrystallization, with γ s the subgrain boundary energy (which depends on the subgrain misorientation) and D the mean subgrain diameter. As a first approximation, it is assumed that the stored energy is uniformly distributed within the deformed grains. Therefore, the only factor that varies within a deformed grain is the V f / r ratio of the dispersoid particles. The other factor controlling the occurrence of recrystallization is the availability of suitable sites for nucleation of new grains. In commercial aluminium alloy hot rolled plate, it has been demonstrated that particle stimulated nucleation (PSN) of recrystallization at coarse constituent particles located on grain boundaries is the primary initiation mechanism for new recrystallized grains [2]. The critical particle size for PSN under typical hot rolling conditions for 7xxx plate has been shown to be 5µm [2]. Therefore, in the model, it was assumed that recrystallization will be possible in regions adjacent to at least one particle greater than this critical size, but will not occur in regions without a particle exceeding this size. The distribution of PSN particles was determined experimentally from FEGSEM images. Experimental Method To test the model, a number of experiments have been performed using commercial AA75 plate. To validate the predictions of dispersoid precipitation kinetics, heat treatments were performed on as-cast material, heat treated in the range 35-5 o C, followed by characterization of the dispersoid size, number and distribution using FEGSEM (Philips XL-3) and TEM (Philips CM2). To study the effect of model optimized thermal treatments, as cast AA75 was homogenized using standard and modified routes, and then laboratory hot rolled. This rolling was carried out at 45 o C in a single pass to a true strain of.9, followed by solution treatment for 3 minutes at 48 o C. To investigate the effect of adding scandium and study through thickness variations in recrystallization (scandium free alloy only), three AA75 plates (.13wt% Zr) were produced with scandium levels of (scandium free),.1, and.3wt%. The plate was commercially manufactured by direct chill casting, homogenization (24h at around 48 o C), and hot rolling in the temperature range o C from an initial thickness of approximately 45mm down to a thickness of 13mm. Particle analysis was performed using FEGSEM and EBSD. An operating voltage of 1kV and backscattered imaging mode was used to image the large constituent particles and dispersoids, which were quantified using image analysis. Prior to EBSD analysis, specimens were electropolished at 3 o C and 12V in a 7% nitric acid 3% methanol solution. For acquisition of EBSD maps, an operating voltage of 2kV was used, specimen tilt of 7 o, and a step size of.5µm were used. EBSD data were analysed using the VMAP software package [1] to identify grain and subgrain sizes and boundary misorientations.

4 48 Fundamentals of Deformation and Annealing To determine the effect of deformation conditions on the recrystallized fraction, channel die deformation experiments were conducted on homogenized commercial AA75 using the Alcoa Thermomechanical Simulator to a total true strain of 1 at a temperature of 45 o C and strain rates between.1 and 1s -1. Following deformation, the specimens were rapidly quenched, before being prepared for EBSD examination as described above. To determine the fraction of recrystallization obtained after solution treatment, specimens were first overaged and then etched using Graff-Sargents reagent (.5% HF, 15.5% HNO 3, 84% H 2 3g CrO 3 ) to reveal the grain structure and identify recrystallized regions. Results Dispersoid Precipitation Model. Validation of the dispersoid precipitation model is described in detail elsewhere [7]. The model contains one unknown parameter, which is the interfacial energy of the dispersoid phase. This was found by fitting model predictions and data for both zirconium and scandium containing aluminium alloys. An example prediction, showing comparison of model predictions and measurement of Al 3 Zr precipitation in AA75 at 5 o C is given in Fig. 1(a,b). Fig. 1(c,d) shows the predicted effect of scandium additions on the dispersoid precipitation kinetics in AA75. It can be seen that scandium is predicted to accelerate precipitation, which is to be expected given the greater diffusivity of scandium than zirconium in aluminium [7]. The model also predicts that scandium additions have a minor effect on the final particle size reached, and changes in kinetics with alloy content are therefore largely due to changes in the particle number density. Another consequence of the faster diffusivity of scandium than zirconium is that the dispersoids are predicted to form with a composition that is initially scandium rich (close to Al 3 Sc). The matrix therefore first becomes depleted in scandium (Fig. 1(e)) before finally the dispersoids enrich in zirconium at the surface (Fig 1(f)). Diffusion within the dispersoids is very limited, so that this variation in composition is maintained, leading to particles with a scandium rich core and zirconium rich outer shell, as confirmed by atom probe measurements [11]. This has the important effect of reducing the coarsening rate of the dispersoids when zirconium is present. As discussed, for the purpose of recrystallization control, it is the dispersoid distribution that is critical. Fig 2(a) shows the initial zirconium variation across a typical dendrite in AA75 after casting, predicted using a Scheil model [7]. Fig 2(b) shows the predicted variation in dispersoid number and radius across the dendrite after homogenization. Towards the edge of the dendrites, a dispersoid free zone is predicted. It is this zone that is particularly susceptible to recrystallization. The model has been used to investigate the effect of homogenization treatment, alloy composition, and through thickness position on this dispersoid free zone and the subsequent recrystallized fraction, as discussed below. Optimization of Homogenization. One possible method to enhance the dispersoid distribution is to modify the homogenization treatment. Slow heating to the homogenization temperature occurs naturally in industrial processing as a result of the large size of the ingots being heated. This has a beneficial effect, since the dispersoid nucleation rate in zirconium depleted regions is enhanced at temperatures below the required final homogenization temperature. Fig 3(a) shows the predicted variation in dispersoid nucleation and growth rate with temperature in AA75 (for which the homogenization temperature is typically in the range o C). Dispersoid nucleation in low zirconium regions is maximized at around 43 o C; below this temperature, the kinetics are too slow for precipitation within industrially useful timescales. By holding at 43 o C during the heat up to homogenization, the model predicts that dispersoid nucleation towards the dendrite edge is promoted. This is shown in Fig. 3(b), where the predicted variation in the dispersoid pinning

5 Materials Science Forum Vol effectiveness (characterized by the Vf/r) ratio is plotted in the region near the dendrite edge. The reduction in recrystallized fraction associated with this improvement in dispersoid distribution has been demonstrated experimentally on laboratory hot rolled specimens (Fig. 3(c)). However, it is not possible to completely suppress recrystallization using this method, since close to the dendrite edges the zirconium concentration is too low for dispersoid precipitation within industrially acceptable timescales regardless of the temperature used (as Fig. 3(b) shows). Number Density / µm -3 (c) (e) (a) Volume Fraction Time / h 6 x 1-3.7Zr-.1Sc.13Zr-.2Sc 5.13Zr-.1Sc.13Zr-.18Sc (d) Predicted Error bar Measured Time / h (f) Mean Radius / nm Mean radius / nm (b) Time / h radius.7zr-.1sc.13zr-.2sc.13zr-.1sc.13zr-.18sc Time / h Sc concentration / atomic fraction 8 x Initial matrix composition α + Al 3 (Sc,Zr) Final matrix composition α X Zr concentration / atomic fractionx 1-4 X Concentration.25 Al-Zr-Sc Zr Sc Time / h Fig 1: (a,b) Predicted and measured evolution of Al 3 Zr in AA75 number density and radius at 5 o C. (c,d) Predicted effect of combined scandium and zirconium on Al 3 (Sc,Zr) precipitation kinetics (e) Predicted pathway showing the variation in matrix content of Zr and Sc during precipitation (AA75 with.1sc-.13wt% Zr) (f) Predicted change in interfacial composition of dispersoids during precipitation at 3 o C (Al-.6Sc-.6Zr at%).

6 2.5.6 Zr (b) Zn 1 Mg.2.5 Cu.5 Centre Distance 2 Radius / nm 3 Conc. Zr / at.% (a) Radius 1 2 Number density 1 Edge 4 Centre.5 Distance 1 Edge Number Density /µm -3 Fundamentals of Deformation and Annealing Conc. Cu,Mg,Zn / at.% 5 Fig 2: (a) Distribution of alloying elements across a typical dendrite (centre to edge) in AA75 (.13wt% Zr) predicted using a Scheil model. (b) Predicted number density and radius of dispersoids across dendrite after homogenization (48oC, 24h). (a) (b) f.1 V /r /µm Growth Rate / nms-1 Nucleation Rate / µm-3s Nucleation Rate Growth Rate o Temperature / C (c) isothermal ramp ramp + hold Normalized Distance.4 (d) Fig 3: (a) Predicted variation in Al3Zr nucleation and growth rate in AA75 with temperature (b) Predicted variation of pinning effect (Vf/r) near dendrite edge for isothermal, 2h heat up, and 2h heat up + 5h hold at 43oC homogenization treatments (48oC, 24h). (c) Microstructure after rolling and solution treatment with standard ramp heating and homogenization (d) Microstructure after rolling and solution treatment with modified ramp heating, hold, and homogenization. Effect of Scandium Additions. Scandium is known to be highly effective in suppressing recrystallization, particularly when added in combination with zirconium in aluminium alloys. Scandium is an expensive alloying element however, and it is therefore desirable to use it sparingly. The model was used to predict the effect of small scandium additions on the dispersoid distribution and fraction of recrystallization. Fig 4(a) shows the measured distribution of scandium and zirconium across a typical as-cast dendrite in three AA75 alloy variants, along with the predicted distribution obtained from the Scheil model. As expected, scandium segregates in the opposite direction to zirconium.

7 Materials Science Forum Vol The effects of scandium additions on the variation of dispersoid size and number across the dendrite are shown in Fig. 4(b). Fig. 4(c) shows the predicted effect of small scandium additions on the variation in the pinning effectiveness (V f /r ratio) of the dispersoids. It can be seen that even very small scandium additions are predicted to have a potent effect on enhancing V f /r, particularly towards dendrite edges. The effect on recrystallization can be estimated if it is assumed that a constant value of V f /r is required to prevent recrystallization, regardless of composition (this assumption ignores any effects of scandium additions on the stored energy obtained after deformation). Assuming a typical fraction of recrystallization in scandium free material of 25%, the critical V f /r ratio above which no recrystallization occurs can be estimated. It can be seen that a scandium addition of.3wt% is predicted to result in a V f /r ratio that exceeds this critical value at all positions across the dendrite. Therefore, in such an alloy, recrystallization will be completely suppressed (assuming no variation in stored energy and ignoring local variations). Whilst it is yet to be proved that such low scandium levels can be completely effective, it has been shown that as little as.1wt% Sc can completely prevent recrystallization in commercially rolled AA75 plate [5]. (a) (b) (c) Mean Radius / nm Sc.3Sc Sc.1Sc.3Sc Number density 1 Mean radius Sc Distance Across Dendrite (normalized) Number Density / µm -3 V / r / m -1 f 4 x Sc.5Sc.3Sc.1Sc Sc V f /r crit Distance Across Dendrite (normalized) Fig 4: (a) Variation of as-cast composition across a dendrite (centre to edge) in 75 (.13wt% Zr) with and without scandium addition as predicted and measured (b) Predicted variation in dispersoid size and number after homogenization (c) Predicted variation in pinning effectiveness (V f /r) for AA75 (.13wt% Zr) with small scandium additions. Through Thickness Variation of Recrystallization. In thick plate hot rolled aluminium alloys, there is often a substantial through-thickness variation in the recrystallized fraction observed after processing. Thick AA7xxx plate is often characterized by a high fraction of recrystallization at the plate surface and centre compared to that at quarter thickness. There are a number of potential factors contributing to this variation. These include a variation in composition through thickness due to macrosegregation, a variation in size and distribution of coarse constituent particles, and a variation in stored energy. To investigate the significance of each of these variables, commercially rolled 75 plate was studied using EBSD and FEGSEM to determine the as-rolled grain and subgrain structure and number of particles above the critical size required for PSN. The dispersoid model was used to predict the variation in dispersoid distribution and dispersoid free zone width through the thickness, taking into account the macro- and micro-scale composition variation. Fig 5(a) shows the measured variation in mean subgrain size and misorientation after rolling. Near the surface, the subgrains are smaller with a greater misorientation, leading to a greater stored energy (Fig 5(b)). This decreases until approximately a third way through thickness (.33T), after which the stored energy does not change greatly. Measurement of the number of particles capable of PSN showed that this remains reasonable constant from the surface to the.4t position, after which

8 52 Fundamentals of Deformation and Annealing it increases continuously, almost doubling at the T/2 position [12]. Using this information, the variation in recrystallized fraction was calculated through the thickness and compared with measurements (Fig. 5(c)). The measurement at T/2 was used to calibrate the model, and thus perfect agreement is assured at this position. Given this point, the model reproduces the form of the measured variation at the other positions reasonably well. Using the model, it is possible to isolate the variables responsible for the observed variation by studying the effect of removing that variable from the analysis. Fig 5(d) shows the effect of ignoring the through thickness change of each of the variables in turn. This demonstrates that the increase in recrystallized fraction in the near surface region can be explained by the increased stored energy after deformation, whilst the additional PSN particles at mid-thickness (T/2) are responsible for the increase in recrystallized fraction at this position. The variation in dispersoid distribution due to through thickness composition variations has a relatively minor effect. (a) (b) (c) (d) Fig 5: (a) Measured variation in mean subgrain size and misorientation through the thickness of AA75 as rolled plate. (b) Calculated variation in stored energy. (b) Measured and predicted recrystallized fraction after solution treatment. (c) Effect of eliminating variables on the predicted through thickness variation in recrystallized fraction. Effect of Deformation Conditions. Strain rate is critical in determining the as-deformed structure and hence the driving force for recrystallization. Channel die deformation at different strain rates (.1-1s -1 ) was used to study this effect in AA75. Fig. 6 shows EBSD maps for specimens after channel die compression, deformed at the two extremes of strain rate used in the experiments. From such maps, the average subgrain size and misorientation were calculated, and used to estimate the average driving force for recrystallization (Table 1). The scatter of subgrain sizes and misorientations around this average was typically large, as evidenced in Fig. 6(b), where subgrains ranging from approximately 2 to 2µm were observed.

9 Materials Science Forum Vol The dispersoid precipitation model was used to calculate the variation in Vf/r across a typical dendrite for the alloy and homogenization conditions used in this study (AA75, homogenized at 48oC for 24h). The width of the region where the calculated driving pressure for recrystallization due to the stored energy exceeded the predicted pinning pressure from the dispersoids was then calculated, and is shown in Table 1. For strain rates less than 1, it can be seen that the width of the region where the pinning pressure (Pz) exceeds the driving pressure (Pd) for recrystallization is less than the average subgrain size. Ignoring local variations in subgrain size, this suggests that for these conditions static recrystallization will be suppressed because further outward growth of subgrains to create high angle grain boundary will not be possible. (a) (b) 25µm 25µm Fig 6 EBSD maps showing substructure after deformation at 45oC with a strain rate of (a).1s-1, (b) 1s-1. Fig. 7(a) shows micrographs after solution treatment of specimens deformed at a strain rate of.1s-1 and 1s-1. In the specimen deformed at strain rate 1, clear recrystallized regions can be seen, distinct from the unrecrystallized regions. In the specimen deformed at a strain rate of.1s-1 there is no such clear distinction. The structure consists of unrecrystallized grains containing subgrains of variable, but generally large, size (as confirmed by EBSD analysis [12]). Table 1 Mean subgrain size (D), misorientation (θ) after deformation, calculated stored energy and predicted width of region where boundary migration is possible (pinning pressure less than driving pressure). Strain Rate D / µm θ/o Stored Energy/kJm Width of region where Pz < Pd / µm Fig. 7: Solution treated specimens, deformed at 45oC, strain rate (a).1s-1, (b) 1s-1.

10 54 Fundamentals of Deformation and Annealing Summary A model has been developed for predicting the precipitation and distribution of Al 3 (Zr,Sc) dispersoid particles in commercial aluminium alloys. This model has been coupled to a simple model to predict the recrystallized fraction in these alloys after thermomechanical processing. The model is based on the principle that recrystallization will only be possible in regions of the microstructure where the Zener pinning pressure due to the dispersoids is less than the average driving pressure for boundary. Furthermore, recrystallization will only occur in regions that contain a particle above the critical size for PSN to initiate recrystallization. A number of applications of the coupled model have been presented: Predicting optimized homogenization treatments which contain a hold during heat up to promote dispersoid nucleation in solute depleted regions, with the effect of reducing recrystallization Estimating the minimum level of scandium required to prevent recrystallization when added in conjunction with zirconium in AA7xxx alloys. Understanding the observed through thickness variation in recrystallization in AA7xxx plate, which is shown to result from the reduction in stored energy towards the plate centre, countered by the availability of additional PSN sites close to the centreline. Interpreting the observed recrystallized fractions obtained in AA75 after deformation at different strain rates. References [1] J. T. Staley, Properties Related to Fracture Toughness, ASTM STP 65, American Society for Testing and Materials, 1976 [2] O. Engler, E. Sachot, J. C. Ehstrom, A. Reeves, R. Shahani, Mater. Sci. Tech. Vol. 12 (1996), p [3] J. D. Robson, P. B. Prangnell, Acta Mater. Vol. 49 (21), p [4] J. D. Robson, Mater. Sci. Eng. A Vol. A338 (22), p [5] F. A. Costello, J. D. Robson, P. B. Prangnell, Mater. Sci. Forum Vol (22), p [6] H. Hallem. W. Lefebvre, B. Forbord, F. Danoix, K. Marthinsen, Mater. Sci. Eng. A Vol. A421 (26), p [7] J. D. Robson, Acta Mater. Vol. 52 (24), p [8] R. Kampmann, R. Wagner, Materials Science and Technology Vol. 5, (VCH Weinheim, Germany 1991). [9] F. J. Humphreys, M. Hatherly, Recrystallization and Related Annealing Phenomena (Pergamon Press, UK 1996). [1] F. J. Humphreys, Orientation Mapping and Quantitative Metallography by EBSD (Manchester Materials Science Centre, 2). [11] B. Forbord, W. Lefebvre, F. Danoix, H. Hallem, K. Marthinsen, Scripta Mater Vol. 51 (24), p [12] J.D. Robson, B.J. McKay, C. P. Heason:, Proc. ICAA-9 (Institute of Materials Engineering Australasia 24), p. 913