Corrosion of Potlining Refractories: A Unified Approach. R. Pelletier, C. Allaire, O.-J. Siljan* and A. Tabereaux**

Transcription

1 Corrosion of Potlining Refractories: A Unified Approach R. Pelletier, C. Allaire, O.-J. Siljan* and A. Tabereaux** CIREP-CRN Department of Engineering Physics & Materials Engineering, Ecole Polytechnique of Montreal (CRIQ campus), 8475 Christophe Colomb Street, Montreal Quebec H2M 2N9 Canada *Norsk Hydro ASA, Research Centre Porsgrunn P.O. ox 2560 Research Centre HPI, Porsgrunn 3907, Norway **Alcoa Inc Second Street, Muscle Shoals, Alabama 35661, USA ASTRACT The corrosion chemistry of aluminosilicate refractories in potlinings has been always studied with respect to one corrosive agent either molten Na or Na vapour. or this reason, both approaches can explain only a fraction of in-field observations. This paper presents a unified approach taking into account the effect of the above-mentioned corrosion agents on aluminosilicates. Along with the model, mathematical tools have been developed to help interpret experimental results as well as make predictions of onservice behaviours. Such predictions are based on the use of corrosion maps which vary according to the sodium ratio [R Na = N Na(V) /(N Na +N Na(V) )] as well as to other parameters. Predictions made with the novel model match almost perfectly with the results obtained in a wide range of corrosion conditions. It appears that the behaviour of cell potlinings can be divided into two distinct groups based on a chemical criterion. Depending on the quantities of fluorides that a lining lets percolate and of metallic sodium diffusing, cells can be described as "wet" or "dry". Industrial aspects regarding this "wet"/"dry" distinction are also discussed. INTRODUCTION In the past, two corrosive agents have been identified as the main responsible for aluminosilicate corrosion in potlining: liquid sodium fluoride and sodium vapour. Two distinct theoretical approaches have been developed to interpret laboratories and postmortem observations. However, each approach can only explain a fraction of the wide range of observations made in used potlinings. If these two approaches could be

2 combined, a better understanding of real life conditions would be possible. In the following, a theoretical strategy to unify the two approaches is presented. The following development is inspired from the work of Schøning et al. [1]. In the approach that proposes molten Na as the corrosive agent, the reaction pattern is: Na + [aluminosilicate refractory] Na 3 Al 6 + [Al-Si-Na-O compounds] (1) asically, Na attacks alumina to produce cryolite and Na 2 O, which combines with the remaining refractory. In the other approach, the reaction pattern is: Na (g) + [atm] + [aluminosilicate refractory] [Al-Si-Na-O compounds] (2) The sodium reacts with the atmosphere (gaseous oxygen or carbon monoxide) and/or with the silica contained in the refractory to produce Na 2 O, which reacts with the remaining refractory. The reaction with the atmosphere should be favoured if the gaseous oxygen renewal in the lining is fast. The availability of gaseous oxygen depends on the lining permeability and on the gas tightness of the cell. In the present development, it is assumed that all the sodium is oxidised by the atmosphere. Under this assumption, the reaction pattern becomes: Na 2 O + [aluminosilicate refractory] [Al-Si-Na-O compounds] (3) The reaction patterns presented in equations (1) and (3) are linked by the following basic reaction: 3Na 2 O + 2Al 3 6Na + Al 2 O 3 (4) THEORETICAL DEVELOPMENT Combining reaction pattern (1) and (3) with the help of the reaction (4) results in the following unified corrosion pattern: 2

3 Na 2 O + Na + Al 3 + [aluminosilicate refractory] [Na-Al- compounds] + [Al-Si-Na-O compounds] (5) A good mean of visualising the predictions resulting from this corrosion pattern is by drawing what could be called "corrosion maps". Corrosion maps Corrosion maps are designed to visualise the evolution of the mineral composition of refractories during the corrosion process. undamentally, they are projected views of coexistence diagrams along certain paths. igure 1 shows the trigonal composition prism proposed by Rutlin and Grande [2], which summarises the coexistence lines of the system Na 6 6 -Si 3/2 6 -Al 2 6 -Na 6 O 3 -Si 3/2 O 3 -Al 2 O 3. All the aluminosilicate refractories lie on the Si 3/2 O 3 -Al 2 O 3 line, which is the silica-alumina line. During the corrosion process, at least those described by the corrosion pattern (5), the mineral composition of a refractory shifts from a point on this line (initial composition) to another located in the Na 6 O 3 -Na Al 2 6 plane (ultimate composition). The exact location of the latter point depends on the proportion of Na 2 O, Na and Al 3 in the mixture corroding the refractory. These two points delimit a line corresponding to a corrosion path. Each path is different because it passes through a certain number of domains at specific proportions of refractory and corrosive agents. Note that in coexistence diagrams such as the one shown in figure 1, the domains are bounded by four triangular planes that are each delimited by three coexistence lines. ecause all the paths that converge to the same ultimate composition form a plane, they can be represented on a single corrosion map. Therefore, a map shows the behaviour of all the aluminosilicate refractories in a specific corrosion condition. With only a few maps, it is possible to predict the corrosion path of any aluminosilicate 3

4 refractory in all the conditions taken into account by the corrosion pattern (ex. equation (5)). To draw these maps, all the coexistence domains have to be expressed as mathematical functions. Here is an example of how each boundary on the corrosion map is calculated. irst, the reactive side of the corrosion pattern is made into an equation: Z M Na 2O Y Y X (100 X) Na2O+ Na+ Al 3 + SiO 2 + Al 2O3 M M R M M Na Al3 SiO2 Al2O3 (8) Where X,Y and Z are the number of grams of each chemical, the M i 's are the molecular weights and R is the Na/Al 3 mass ratio of the initial fluorides melt. This equation remains the same for all the domains. or the products side, we have to go systematically through all the existing triangular planes inside the trigonal prism of the coexistence diagram. The three compounds delimiting each plane are used as reaction products to complete the reaction, for example: A Na 3Al 6 + Al 2O3 + C NaAlSiO 4 (9) Where A, and C are the number of mole of each products. The reaction expressing the location of the plane Na 3 Al 6 -Al 2 O 3 -NaAlSiO 4 contains six unknowns since R is imposed by the corrosion conditions. Writing the mass balance of each element provides only five equations. To solve the system, lets introduce a new parameter, R Na, defined as the molar fraction of the sodium introduced as oxide with respect to all sources of sodium, knowing that the only other source, under the assumed conditions, is the sodium fluoride from the percolating bath. R Na 2 Z M Na 2O = 2 Z Y + MNa O M 2 Na (10) 4

5 This parameter is not only mathematically useful, it also characterises the lining behaviour in a very original and meaningful way. This aspect will be treated later on. With this new equation, each system of equations can be solved to get Y or Z, as a function of X, the mass fraction of silica in the uncorroded refractory. The diagram shown in figure 1 is valid at subsolidus temperature, which is probably close to 650 C [2]. At higher temperatures mullite must be considerate and Na 2 Si 6 becomes unstable. or this reason, the coexistence diagram has to be slightly modified. Thermodynamic calculations (.A.C.T., École Polytechnique, Canada) suggested that Mullite should be coexistent with albite, cryolite and alumina or silica. The compounds Na 2 Si 2 O 5 and Na 2 SiO 3 are assumed to be coexistent with Na instead of Na 2 Si 6. Note that there is possibility that one or both of these compounds could be coexistent with cryolite instead of Na. It is beyond the scope of this paper to clarify this particular aspect. Also, theoretically, the solid solution luorotopaz (2Al 2 O 3 xsio 2 Si 4 ) should also be considerate. However, because only Na, Na 3 Al 6 and Na 2 O are added, in the proposed unified corrosion pattern, some domains of coexistence cannot be reached, including any domain containing luorotopaz. The easiest to visualise among these unattainable domains is the one containing SiO 2, Al 3, Na 5 Al 3 14 and Si 4. It can be calculated that to enter any domain containing Na 5 Al 3 14, the following condition must be met: R < 3 Na 2+ ( 1 4 R R Na) (11) This condition is certainly rarely met because even in the most favourable conditions, i.e. R Na = 0, the corroding melt would need to have a mass ratio lower than

6 Many combinations of parameters can be used as axes when drawing corrosion maps. A good choice is the weight percent of cryolitic agents, i.e. Na 3 Al 6 + Na + Na 2 O, introduced (Y-axis) as a function of the silica content in the refractory (X-axis). This choice for the Y-axis is particularly useful in an engineering point of view. Since the quantity of cryolitic agent under carbon blocks increases as the potlining ages, the Y- axis can be seen as time. or a similar reason, it can also be seen as the vertical position into the refractory lining. Depending on the assumed corrosion conditions, the number of domains visible and their shape vary significantly. Two important parameters control the number and the shape of domains: R Na and R. The bath ratio (R ) has lesser effects on the maps. Its most important effect is to inflate upward the domains in which cryolite is the stable fluoride. The parameter R Na has a much more dramatic effect on the corrosion maps. igures 2 shows the evolution of the corrosion map as the sodium ratio passes from R Na = 0 to R Na = 1. On the maps, the thin lines represent equilibrium involving only cryolite as fluoride, while the thick lines correspond to equilibrium involving only villiaumite as fluoride. Therefore, when a domain is surrounded by only one type of line, the assemblage contains only one fluoride, either cryolite or villiaumite. Note that figure 2 (a) is very similar to the diagram proposed by Schøning et al. [1]. They both constitute cases in which no Na 2 O is present. The only difference between the two is that figure 2 (a) is drawn for R = 3.2 instead of infinity. Consequently, all the domains containing cryolite are slightly inflated upward in figure 2 (a) compared to the diagram proposed by Schøning et al. The corrosion map presented in igure 2 (e) has a particular interest since it has been drawn for R Na = 1. Although it shows domains in equilibrium with Na and 6

7 others with Na 3 Al 6, in fact, it contains no trace of fluorides since all the sodium comes from Na 2 O. As such, this map is the one that would have been obtained by Allaire [3], if the author had plotted the ternary SiO 2 -Al 2 O 3 -Na 2 O coexistence diagram with the present choice of X-Y axes. Consequently, by varying R Na from 0 to 1, a very wide range of conditions can be covered with the proposed unified corrosion pattern. This shows that the two previously antagonist approaches constitute in fact the limiting cases of a more general corrosion pattern. Dry cells versus wet cells It can be noted that for lower values of R Na, the approach predicts that aluminosilicate linings are ultimately converted into one of three assemblages in which there is always presence of both cryolite and villiaumite. or higher values of R Na, cryolite disappears from the intermediary and ultimate assemblages. This is due to the consumption of Al 3 by Na 2 O, as presented in reaction (4). The critical value of R Na at which cryolite free domains begin to appear at the top of a map, is given by the following relation: R Na > 3/(2R + 3) (12) Of course, for those domains to be an important portion of the map, significantly higher values of R Na are necessary. This criterion, or similar ones, can be used to differentiate two totally distinct types of lining behaviours: 1) - wet linings, in which the main corrosive agent is the percolating melt, 2) - dry linings, for which sodium vapours is the controlling agent. Wet linings, or "wet cells", should be ultimately converted into one of the three following mixtures, according to their A/S ratio: 1) - nepheline, β-alumina and fluorides, 2) - nepheline, glass and fluorides or, 3) - glass and fluorides. If R Na is a little 7

8 higher than zero, nepheline, β-alumina and fluorides should be the ultimate mixture for any aluminosilicate refractory. or dry cells, the picture is more complex. It depends on how "dry" is the cell. or a relatively dry cell, a realistic prediction would be: nepheline, Na, possibly sodium aluminate and either β-alumina or a glass. One important difference between both types of lining behaviours is that the bath found in dry cells should have a much higher Na/Al 3 mass ratio than the one found in wet cells. Use of corrosion maps Corrosion maps can be used to interpret experimental results or to make predictions regarding the on-service behaviour of a lining. If an estimation of R and R Na is known, the map can be drawn and a simple vertical line above the desired silica content will tell the evolution of that refractory. Note that the silica content has to be calculated as if the refractory was composed of only silica and alumina. If R Na is not known but chemical analyses are available, an estimate of R Na can be calculated. The theoretical development leading to the R Na estimation is the following. irst, the mixture of oxides in the corrosion products has to be considered as a single nonstoechiometric compound. The products side of the reaction then becomes: A M Na Na+ M A R Al3 Al 3 + NaAl Si a b O ( 3a+ 4b+ 1) 2 (13) Where a and b are respectively the Al/Na and Si/Na molar ratios of the oxides found in the analysed sample. The parameter R is the Na/Al 3 mass ratio of the fluorides also found in the corroded sample. Defined as such, the system can be solved to calculate the 8

9 theoretical refractory composition that would eventually lead to the found mixture. The analytical solution for this composition is: X* = M M Al2O3 SiO ( 2b H) { a H + ( R R )} + 2b H (14) where: H R ( ) ( ) ( ) R Na 2R R R 1 R = Na The composition X* is a function of R, R Na, R, a and b. Reorganising this function allows to obtain the sodium ratio, R * Na, necessary so that X* = X, where X is the uncorroded refractory composition. The expression for R * Na is the following: R * Na = X * M M Al2O 3 SiO 2 * Al ( ) 2O3 * R R X ( 3a + 1) ( 100 X ) M M 6b * [( R R ) a R ( 2R + 3) ] + ( 100 X ) 2b R ( 2R + 3) SiO2 (15) Once R * Na is known, the map can be plotted and the evolution of the refractory can be predicted by drawing a vertical line at the refractory initial composition. To locate the analysed sample on the map, its vertical coordinate has to be calculated with the following equation: where: Y * = Y = [ Z + Y ( 1+ 1 R )] [ Z + Y ( 1+ 1 R ) + 100] MAl R X 3 MSiO b 2 * 2 R RNa * ( 1 R ) Na ( 2 R + 3) ( 2 R + 3) + 6 ( R R ) * (16) 9

10 M Z = 2 M Na 2 O Na ( 1 R ) However, a special attention has to be taken when determining R. As can be seen in the presented corrosion maps, the value of R may be restricted in some domains. In most domains, R is fixed to either 1,5 or infinity depending on if the oxides are in equilibrium with cryolite or villiaumite, respectively. If the mineralogical analysis is not clear enough on whether or not both these phases are present, it is preferable to associate all the luoride to only one of the two above-mentioned compounds, when interpreting the chemical analysis. Moreover, sample inhomogeneity can sometimes lead to apparently incoherent analysis. EXPERIMENTAL VALIDATION To validate the proposed unified approach, experimental results obtained in different degrees of "dryness" are compared with predictions from the model. The experimental results obtained from a new corrosion test are used for that purpose. This novel test allows to expose a bulk refractory sample to the two major corrosive agents which are liquid fluorides and sodium vapour. It also permits the variation in sodium vapour/fluorides ratio. igure 3 shows schematics of this new corrosion test. In configuration (J-1) the sample is directly exposed to melt but in configuration (J-2), melt and aluminium are placed into the upper crucible so that the sample is exposed only to the percolating fluorides (the amount of the latter is controlled by varying the porosity of the crucible). Moreover, aluminium reacts with the melt to generate sodium that diffuses towards the refractory sample. or more detail on the experimental aspects, see Allaire et al. [4]. Descriptions of the materials used and of the testing conditions are given in Table R Na Na Y 10

11 1 and Table 2. The chemical and mineralogical analyses of the samples have been used to calculate their Al/Na, Si/Na and R ratios. or each sample, a corrosion map has been plotted using an estimated value for R Na. The summary of the predictions and the experimental results is presented in Table 3. Knowing that albite has a strong tendency to form an amorphous phase [2,5], the prediction is perfect for samples 19, 20, 25 and 26, and very good for samples 27, 24-N and 24-D. In sample 27 the model predicts that corundum should be present as a major compound while mullite should not be detected. The experimental point is however very close to the domain in which mullite is present. This suggests that the sample is not completely at equilibrium. As a matter of fact, the mineralogical analysis of that sample shows the simultaneous presence of cristobalite, mullite and corundum, which is impossible to get under equilibrium condition. A similar remark can be made regarding the sample 24-N. Therefore, the present experimental results support very well the unified corrosion model. Another interesting aspect in these results, is the trend that follow the calculated R * Na values. The calculated R * Na decreases as the fluorides exposure becomes more direct. This supports the adequacy of the approach and of the concept underlying the parameter R Na. This also confirms that the newly developed corrosion test is capable of producing a wide range of corrosion conditions. inally, in test 26, performed with the J-1 configuration, it was not possible to find a value of R Na to reach the condition X * = X. The origin of this problem is most probably the partial separation of the corrosion products. As a matter of fact, there is a tendency to form a fluoride rich liquid and an oxides rich liquid in the direct exposure 11

12 configuration when a refractory having a low alumina to silica ratio is tested. The analyses for test 26 have been performed only on the oxides rich zone. In this condition, the mass balance equations underlying the formulas used to calculate R * Na and X * do not strictly apply. INDUSTRIAL ASPECTS As mentioned earlier, post-mortem analysis made in the past showed that there is a whole range of potlining behaviours. ased on a chemical criterion, it is proposed to divide these behaviours into two groups: 1) - dry cells and, 2) - wet cells. The cells that let the molten fluorides percolate the most through the cathode blocks will be in the second group. The parameters that determine the relative proportion of fluorides and sodium that reach the refractory lining are: 1) - wettability of the carbon blocks with respect to the melt, 2) carbon block permeability, 3) apparent sodium diffusion coefficient in the carbon block, and 4) - the presence of gaps and cracks. Definitively, it should be the properties of the carbon blocks and the quality of their installation that control the type of corrosion that the refractories will be subjected to in service. The type of corrosion could very well influence the best choice of refractory. Dry cell linings are exposed to much less fluorides than linings in wet cells. Laboratory experiments should be done to determine which type of refractory behaves best in each type of cells. Classifying the existing cell technologies would also be useful. inally, additional care should be taken when defining the optimum properties of carbon blocks. CONCLUSIONS The development of a unified approach has led to the following conclusions: 12

13 1. The two previous approaches developed by runk [6] and Siljan [7], and Allaire [3] have been unified in a model that allows to make predictions in intermediary conditions. 2. The predictions made with this unified approach match almost perfectly the results obtained in a wide range of conditions. 3. Along with the model, mathematical tools have been developed to help interpret experimental results as well as making predictions of on-service behaviours. 4. ased on a chemical criterion, it is possible to divide cell potlining behaviours into two groups: wet cells and dry cells. 5. Wet cells are those that let molten fluorides percolate the most into the refractory lining. It is the properties of the carbon blocks and the quality of their installation that determine the amount of percolating bath. REERENCES 1. C. Schøning, T. Grande and O.-J. Siljan, "Cahode Refractory Materials for Aluminium Refractory Reduction Cells", Light Metals 1999, J. Rutlin and T. Grande, "Phase Equilibria in Subsystems of the Quaternary Reciprocal System Na 2 O-SiO 2 -Al 2 O 3 -Na-Si 4 -Al 3 ", J. Am. Ceram. Soc., Vol. 82 N 9, , C. Allaire, "Refractory lining for alumina electrolytic cells", J. Am. Ceram. Soc., 75, (8), , C. Allaire, R. Pelletier, O.-J. Siljan and A. Tabereaux, "An Improved Corrosion Test for Potlining Refractories", to be published in Light Metals O.-J. Siljan, "Sodium Aluminium luoride Attack on Alumino-silicate Refractories - Chemical Reactions and Mineral ormation", Thesis IUK-1990:61, Trondheim, Norway, runk, "Corrosion and ehaviour of ireclay ricks of Varying Chemical Composition Used in the ottom Lining of Reduction Cells", Light Metals 1994, O.-J. Siljan, "Reaction of ireclay Refractories in Aluminium Reduction Cells", UNITECR '95 Congress, Kyoto, Japan, 1995,

14 Table 1: Details of the experimental conditions. Test Material Configuration Crucible Porosity [%] 19 J J A J A J-1-26 J-1-27 A J

15 Table 2: Chemical analysis (XR) given in weight percents and mineralogical analysis (XRD) of material A and. Oxide A SiO Al 2 O e 2 O MgO CaO Na 2 O < K 2 O TiO Minerals A Major Medium Mullite Corundum Cristobalite Cristobalite Mullite Cordierite 15

16 Table 3: Comparison between the unified approach and the observations. Sample a b R R Na * Prediction Analysis, XRD Albite Nepheline Villiaumite Cryolite Albite Nepheline Villiaumite Cryolite b Villiaumite Cryolite Albite Nepheline Albite Corundum Nepheline Cryolite 24-N Mullite Albite Corundum Cryolite 24-D c Nepheline Albite Corundum Cryolite d 0.09 e Nepheline Albite Cryolite Corundum Glass (57,2) a Nepheline (34,8) Villiaumite (5,0) Cryolite (0,8) Glass (51,8) Nepheline (35,3) Villiaumite (8,8) Cryolite (1,3) Glass (63,8) Villiaumite (16,2) Cryolite (7,5) Nepheline (2,0) Glass (39,3) Mullite (32,3) Nepheline (18,3) Corundum (10,2) Cristobalite (0,8) Cryolite (0,1) Glass (39,9) Mullite (30,7) Corundum (16,3) Cristobalite (8,6) Nepheline (3,6) Glass (41,2) Nepheline (35,8) Corundum (11,9) Cryolite (6,2) Mullite (3,9) Cristobalite (1,2) Nepheline (43,0) Glass (37,2) Cryolite (14,4) Corundum (11,5) a: Values between parenthesis are percentages obtained by a semi quantitative method. b: This value doesn't allow to have X * = X. c: Since no value allows X * = X, this intermediary value is chosen. d: Value imposed to maintain consistency with the predicted phases. e: Imposing a value of zero would make no significant difference to the prediction. 16

17 Si Na Si O Na SiO 2 3 Na SiO 4 4 O 3/2 3 NaAlSi O 3 8 NaAlSiO 4 Al Si O NaAl O Al O 2 3 NaAlO 2 Na O 6 3 Si 3/2 6 Na Si 2 6 Al 2 6 Na 6 6 Na5Al314 Na3Al6 igure 1: Coexistence lines in the quaternary reciprocal system Na 2 O-SiO 2 -Al 2 O 3 -Na- Si 4 -Al 3 at subsolidus temperature proposed by Rutlin and Grande [2]. 17

18 (See following pages) igure 2: Evolution of the corrosion map as the sodium ratio passes from R Na = 0 to R Na = 1. 18

19 (a) (b) igure 3: Schematics of the test configurations: (a) where the refractory sample is indirectly exposed to fluorides (J-2) and (b) where the refractory sample is directly exposed to fluorides (J-1). Note that the dimensions have been kept in inches for simplicity. 19