Analysis and Significance of Non-metallic Inclusions in Non Grain-oriented Electrical Steel

Abstract

NMI in electrical steels are known to have a very complex chemical makeup and history of formation, and exert a large influence on the handling of this steel in secondary metallurgy and casting. We have developed routines to analyse the non-metallic inclusion contents of non grain-oriented electrical steels by automated FEG-SEM analysis. We automatically analyze NMI of sizes down to 0.08 μm² in a steel that has significant amount of alloyed Si (2.3 wt%) by employing a matrix spectrum subtraction routine that leaves only the signal of the NMI itself to be quantified. We describe data reduction procedures for the NMI populations that consist out of duplexing oxides, sulfides, and nitrides. The chemical complexity can be represented and understood in terms of using a multicomponent projection technique, and based on the analysis of the particles it is possible to calculate quantitatively the amount of elements contained in the NMI assemblage of a steel sample. We find that this mass balance gives results in good to excellent agreement with bulk steel analyses for elements that are dominantly inclusion bound, such as Ca, if the complete inclusion size range above 0.08 μm² is taken into account. Based on these data analysis methods, we compare the NMI development in two heats through the secondary metallurgy process. Although the same steel composition was alloyed, the NMI developed vastly different depending on the details of treatment on the ladle furnace.

1. Introduction

In recent years, the advance of automated Field Emission Gun Scanning Electron Microscope (FEG-SEM) based inclusion analysis has greatly changed the quantity and quality of non-metallic inclusion (NMI) data available to secondary metallurgy process, as well as steel product metallurgy improvement. This is especially useful for the development of relatively high alloyed steel grades, such as Si alloyed electrical steel grades where NMI assemblages can be chemically very complex yet have significant influence on castability and product performance.^1,2) For non-grain oriented (NGO) electrical steels, it is well understood that NMIs are developed along both liquid and solid processing routes and they are the assemblages of oxides, sulfides, and nitrides. This makes the automated-SEM inclusion analysis (AIA) of such steels very challenging, compared to simpler micro-alloyed steels. The understanding of the evolution of NMIs can be very helpful to understand the complex chemical reactions occurring in the production of electrical steels.

Here we report on the development and metallurgical usage of fully quantitative AIA data for a NGO electrical steel grade with mid-range Si content, as currently produced at Tata Steel Europe (Table 1). This steel is broadly comparable (lower in Al, higher in Si) to the NGO electrical steels for which Steiner Petrovic et al.³⁾ reported highly unusual Mg-rich NMI. The production of this type of steel involves alloying of large amounts of FeSi and Al-wire additives and it is well known that the obtained steel melts can be difficult to cast due to their aggressivity with respect to conventional refractories and the potential occurrence of CaS in the NMI population leading to CaS-induced clogging. We have performed AIA of NGO electrical steel during steelmaking process, as part of a program to find a robust secondary metallurgy procedure. The aim of this paper is to describe the instrumental procedures and data treatment necessary to obtain AIA data of sufficient quality for the chemically complex NMI populations of this type of steel, and to give an example of the NMI datasets that can be produced for a thorough secondary metallurgy process analysis.

Table 1. Typical composition of the mid-Si non-grain-oriented electrical steel (in ppm wt).

Element	C	N	Al	Si	P	S	Mn	Ti	V	Nb
Average	30	15	4000	23000	90	30	1900	40	30	10

2. Secondary Refining Process and Sampling for NGO Electrical Steel

The samples investigated here are routine lollipop steel production samples, used for online evaluation of the steelmaking process. Two different heats, A and B, with slightly different secondary steelmaking procedures were sampled. Out of a large through-process set of samples, we report here on a set of four such samples for each heat, covering specifically the evolution of the steels and their NMI populations during the ladle furnace process. The general production sequence for this type of steel is: After tapping from a 335 ton basic oxygen furnace (BOF) converter into a ladle, the heats are degassed on a RH-OB degassing station. Thereafter, they are alloyed, refined and desulfurized on a ladle furnace (LF) using an artificial slag. During this LF treatment, samples are taken at different times. The finished heats are then sent to a continuous caster. The difference between heat A and heat B studied here lies in the timing of the main Al deoxidation: heat A is fully deoxidized during RH-OB treatment, heat B is deoxidized only on the LF, and treatment on the LF starts with FeSi addition. FeSi and Al wire are added in the further ladle treatment of both ladles until the alloy target is reached. For both heats A and B, four lollipop samples (A1-4, B1-4) are routinely taken: before start of the ladle treatment, at an early stage as well as at the end of ladle treatment, and at the tundish during casting (about half discharged heat). The lollipop samples were used for routine spark-OES composition analysis to evaluate the metallurgical progress online. Chemical compositions of the samples A1 to 4 and B1 to 4 are given in Table 2. The industrial test procedures did achieve the desired contrast of Al content: heat A arrived on the LF is already with most of its target Al, whereas heat B arrived there contains limited Al content (similar to B1 in Table 2). Corresponding slag composition for heats A and B in LF are given in Table 3. This LF slag is artificially synthesized from raw material additions (lime and bauxite amongst others) during LF treatment, which also involves continuous stirring. Total treatment time in LF is varying but normally in the range of 60 to 90 minutes. Heat target temperature at LF is in the range of 1580 to 1590°C.

Table 2. Compositions of production lollipop samples for heats A and B (spark-OES, all in ppm).

	Time*	Mn	Si	Al**	N	Cr	Ca	S	C
A1	beforeLF	1400	50	2600	30	160	1	52	4
A2	Early LF	1950	20240	2900	25	230	41	22	32
A3	End LF	1950	22550	3780	25	240	18	13	30
A4	Tundish	1950	22580	3740	26		18	10
B1	beforeLF	1346	10	50	21	150	0	52	6
B2	Early LF	1840	21330	4190	16	220	21	30	20
B3	End LF	1960	22480	3460	19	220	23	19	26
B4	Tundish	1950	22580	3740	26		18	10

*:  Time in relation to ladle furnace process.

**:  “soluble” Al from spark OES (disregarding inclusions).

Table 3. Ladle Furnace slag analyses heats A and B (in wt. % oxides).

LF slag	MgO	Al2O3	SiO2	CaO	P2O5	TiO2	MnO	FeO	S
Heat A	5.0	31.5	3.7	58.6	0.01	0.09	0.02	0.6	0.18
Heat B	5.0	25.7	8.7	59.3	0.01	0.16	0.05	0.64	0.20

3. Measurement of Non-metallic Inclusions and Data Processing

3.1. Automated Inclusion Analysis (AIA) System

For the analyses of NMI, the lollipop samples received after spark-OES analysis were cut and polished so that a small section (~10×10 mm) of steel surface directly under the circular OES spark impingement spots could be used for AIA. This was done to ensure that the automated SEM (AIA) inclusion data were as much as possible representative for the same steel analyzed for overall composition and avoiding potential sample bias. The ~10×10 mm steel blocks were mounted in conductive resin for SEM, and polished to standard 1 μm diamond paste.

The AIA work was performed at the Ceramics Research Center, Tata Steel Europe IJmuiden Works, using two FEG-SEMs with Energy Dispersive Spectroscopy (EDS) systems attached. The JEOL 7001F FEG-SEM is equipped with two Silicon Drift detectors (hardware: Thermo Scientific NORAN system 7; software version 3.0) for high speed micro-analysis, and the ZEISS Ultra55 FEG-SEM is equipped with a liquid nitrogen (LN) cooled SiLi detector (hardware: Thermo Scientific NORAN system 6; software version 3.0). A customized inclusion analysis setup, which is integrated into these micro-analysis systems, was used. Since it became clear that number densities of NMI vary with the size of NMI, a full quantitative counting of the NMI was performed separately for small (0.08–0.75 μm²) and large (> 0.75 μm² ) inclusions. The standard setting for AIA of both small and large inclusions are: 15 kV, 500× magnification and 0.112 μm pixel size. Objects (inclusions or other features) were detected on a thresholded Backscattered Electron (BSE) image, and their EDS spectrum acquired using beam rastering mode instead of single point measurement, to prevent loss of duplexing object information. Objects were accumulated up to a preset number of raw objects, or to completing a scan of the programmed observational area, whichever came first. In general, datasets consist out of ~2000 raw objects over the whole observed area of steel blocks (see basic dataset numbers in Table 4). Further, as this steel is relatively highly alloyed, some non-Fe elements of metallurgical interest appear in a generic EDS spectrum of the steel matrix (here: Si and Al) and would thus be imprinted on a straightforward EDS quantification of an inclusion smaller than the excitation volume (which is the case for most of the small inclusions). This problem limits the reliable compositional analysis of NMI in significantly alloyed steels, such as our electrical steels but also in e.g. stainless steels⁴⁾ or TWIP steels. To deal with this, a pre-acquired matrix (steel) EDS spectrum was subtracted from every object spectrum and only the residual spectrum (raw inclusion minus steel matrix) was quantified, allowing reliable compositional analysis of inclusion objects much smaller than the electron beam excitation volume. The matrix subtraction before quantification for the analysis of very small inclusions/precipitates is an indispensable step in steel compositions such as this that have alloy element EDS peaks interfering with target compositions. An example of a relatively small object is given in Fig. 1. The inclusion shown there is AlN and its area measured via AIA is 0.08 μm². The residual spectrum clearly shows the constituent elemental peaks and allows a good quantification. This matrix subtraction step is the main advance that allowed us to reach a smaller size limit for compositional quantification compared to conventional AIA wherein particle analysis limits range from 0.4 μm⁴⁾ to 0.9 μm.¹⁴⁾

Table 4. Basic statistics of the cleaned NMI datasets for the 8 analyzed samples.

Sample	Observed area [μm2] (sizes0.08–0.75 μm2)	Number of NMI (0.08–0.75 μm2)	Volume fraction [ppm] (0.08–0.75 μm2)	Observed area [μm2] (> 0.75 μm2)	Number of NMI (> 0.75 μm2)	Volume fraction [ppm] (> 0.75 μm2)
A1	*	*	*	4.3E+07	892	150
A2	3.7E+06	1928	220	1.8E+07	918	200
A3	4.9E+06	1629	140	2.6E+07	1505	100
A4	9.0E+06**	2330	94	9.0E+06**	701	180
B1	*	*	*	3.6E+07	1293	160
B2	4.8E+06	1737	120	2.0E+07	739	300
B3	5.3E+06	2036	130	3.0E+07	915	150
B4	1.7E+07	759	14	1.2E+07	341	170

*:  Small inclusions were not collected.

**:  Same area, small and big inclusions collected together.

Fig. 1.

Example of an inclusion near the lower size limit of the employed FEG SEM detection. Left – original appearance as seen during the analysis, with automatically calculated thresholded object area. Right – residual object spectrum after subtraction of the sample’s matrix spectrum. Below the images: extract from the obtained quantification using Gaussian method.

For an NMI assemblage in which the precipitating metals can bind to all of the volatiles O, N, and S, a good quantification of the light-element part of the spectrum is critical. The low energy section of the spectrum (<1 keV) contains N–K, O–K, Mn–L and Fe–L peaks. Imperfections in the lower energy region after matrix spectrum subtraction can induce serious errors in the light-element quantification if the improper quantification method is selected. We have evaluated two spectral processing methods whereby Gaussian filtering using Kramers’s Law⁵⁾ was preferred over a digital top hat filter⁶⁾ due to the “backgroundless” spectrum after matrix subtraction. The PRZ correction method as developed by Bastin and Heijligers⁷⁾ was used to convert net counts into elemental wt%. While the light-element quantification is good enough to unambiguously identify the phases present, it should be noted that O and N carry a substantial analytical imprecision.

All object residual spectra and quantification analysis results were recorded together with a predefined set of morphological parameters including cross sectional area, perimeter, and location. Basic observational data (numbers of inclusions, observed areas, volume fractions) are given in Table 4.

3.2. Data Reduction

Datasets were obtained for all eight samples A1–A4 and B1–B4 by the described SEM EDS quantification method. Such a raw dataset can contain artifacts (such as scratches, holes, dust on the surface, preparation residues) that are analyzed due to being thresholded from the SEM BSE image. These false positives need to be cleaned from a raw dataset to obtain the final analytical NMI datasets per sample, We do this using an in-house developed routine written in the commercial software IGOR™(WaveMetrics), which handles data reduction and offers options for NMI data plotting on selected chemical plots as well as calculations of size distributions and bulk-steel NMI mass balances.

3.3. Inclusion Mass Balance

Based on the obtained inclusion composition analyses, and their sizes resp. cross sectional areas, it is possible to calculate quantitatively how much of the given elements are tied up to the particulate phase, which means how much of a given steel bulk analysis (such as routinely obtained by spark-OES such as in Table 2) is bound to the observable particles. This quantification, loosely termed “mass balance” of inclusions, can yield critical metallurgical information, especially for secondary metallurgy where the true state of the liquid steel at a given time during processing is of prime interest.^1,2) However, the calculation of an inclusion mass balance is not straightforward. Two approaches can be used: (i) A simple average of the bulk inclusion composition can be calculated. by averaging the area-weighted individual inclusions, and by using an estimated overall inclusion density and the measured absolute inclusion volume fraction, an estimated bulk mass balance contribution is obtained. The volume fraction of inclusions is a directly measured feature of the datasets, since for a sufficiently large analyzed 2D sectional area, the cross sectional area fraction of a particulate phase and its true 3D volume fraction are identical.⁸⁾ Mass balance from approach (i) gives only an approximate result, as the density of the individual inclusions can vary significantly according to their individual phase makeup.

Approach (ii) for a mass balance calculation takes into account varying phase densities; the actual contribution of phases to each inclusion must be found and weighted by its density to arrive at a bulk inclusion population composition. This is a classical problem of inversion, or, transformation of the compositional space from element to phase coordinates.⁹⁾ If x is the composition vector of each particle analysis in elemental coordinates, and A is the coefficient matrix set up by the composition vectors of each of the constituent phases in terms of the same element coordinates, then   

$$\mathrm{A}\text{ * }\mathrm{y}\text{ = }\mathrm{x}$$(1)

where y is the desired composition vector of each particle in phase coordinates. The solution for y is obtained by straightforward inversion of the phase composition matrix A and multiplication with the measured particle compositions. However, this method has limiting practical requirements. The set of phases to convert into must be known (from direct observation), and must be a linearly independent set of compositions. With a number of phases equal to the number of relevant elements, a unique solution results. For the electrical steel NMI datasets, the method is problematic due to the error of analysis in light elements mentioned above. As the particle population is dominated by binary compounds (AlN, MgS, MgO, CaS), the accuracy of the metal to light element ratio in the analysis becomes critical to a realistic transformation of compositions into a phase makeup. Large deviations in e.g. the apparent measured Al/N ratio in an AlN dominated small particle will force a full matrix inversion based phase transformation to express the excess apparent Al or N through unphysical phase compounds attributed to that inclusion. This can create significant errors in the overall summation, as the contribution of these unphysical contributions gets fully weighted with the respective phase density. Due to the metal/light element errors, the results were estimated by using phase matrix inversion method by excluding the elements measured with the highest relative error, which are the light elements (O, N).

We have calculated an inclusion contribution to a bulk-steel analysis according to both methods of mass balance calculation (average assuming homogeneous inclusions, and phase matrix inversion). The results are given in Table 5. For calculation via approach (i), we used an average inclusion density of 3.5 for the homogeneous approximation (a value halfway between AlN and MgO densities). For (ii) approach, i.e. the simplified matrix inversion method, we used a phase model of spinel (MgAl₂O₄), CaS, MnS, CaO and Al with densities of 4.0, 2.58, 3.99, 3.345, and 3.3 g/cm³ respectively (Al as a proxy for AlN). From the data in Table 5, which are expressed in inclusion-bound elements as ppm of a bulk-steel composition, it can be seen that the agreement between the methods is generally acceptable. The main difference appears in the value for the amount of Al, which is dominated by the AlN content. Since the Al as measured in our data is actually affected by the measurement error for N through the closure effect (normalization), we believe that here the imperfect homogeneous-inclusion approximation (method (i)) is closer to the true value, whereas we believe the matrix inversion numbers to be better for the other elements. The good agreement between the methods for elements which are measured with low error such as Ca, S, and Mg, lets us believe that the approximation values for the inclusion-bound O and N – for which the poor light element quantification precludes matrix inversion method - is of a similar precision. For future attempts to refine the mass balancing method, it should be pointed out that the main error source is the light element quantification: as this can be improved through using e.g. sample-internal phase reference standardization, it may conceivably become sufficiently robust to include in a phase model inversion of the full composition space.

Table 5. NMI composition of the steel samples, expressed as inclusion-bound elements as ppm-wt contribution to a bulk steel analysis. For each sample, two numbers are given:(1) according to homogeneous-inclusion averaging method; (2) according to simplified phase model inversion method.

Heat A	A1		A2		A3		A4
element	(1)	(2)	(1)	(2)	(1)	(2)	(1)	(2)
Ca	0.1	0.12	36	36.9	11	12.4	9	17.6
Al	24	33.3	50	61.3	36	41.5	33	31.5
Mg	0.1	0.02	3	3.66	2	3.04	2	2.18
S	0.1	0.02	20	17.3	6	5.95	6	11.1
Mn	1	0.69	1	0.69	0.5	0.66	0.3	0.32
O	34		17		6		4
N	1		34		25		31
Heat B	B1		B2		B3		B4
element	(1)	(2)	(1)	(2)	(1)	(2)	(1)	(2)
Ca	0	0.51	13	19.0	13	19.3	12	17.8
Al	31	40.1	61	78.4	38	49.7	22	26.8
Mg	0	0.13	3	7.0	3	4.9	1	1.66
S	0	0.26	8	4.38	5	9.62	2	2.19
Mn	0	0.25	2	1.89	0	0.56	0	0.17
O	41		52		30		28
N	1		21		16		3

3.4. Compositional Plots

The inclusions in the electrical steel samples are compositionally complex and often duplexing between multiple phases. In general terms, the bulk of the inclusion compositions is constituted by sulfides, nitrides and oxides of Ca, Al, and Mg. There is no single graphical representation for compositional variations in this six-component system. Thus, projections in graphically representable subsystems have to be employed for plotting and analyzing the inclusion composition evolution along the process route. For inclusions in which metals are partitioned between nitridic, oxidic or sulfidic bonding, simple ternaries (such as Al–Mg–Ca) can yield apparent plotting configurations that are difficult to interpret. This is illustrated in the tetrahedron for the subsystem Al–Mg–Ca–O–S (Fig. 2), which encompasses the compositional ternaries that are frequently employed for analyzing Ca containing steel NMI, such as in Ca-treated Al-killed steels. If a series of sulfide-oxide duplexing inclusions (composed of CaS and spinel in variable proportions), a simple Al₂O₃–MgO–CaO ternary is plotted (equivalent to projecting from the SO_–1 corner of the compositional tetrahedron), then the plotting positions of the inclusions in the ternary will vary as indicated, and it is not possible to recognize whether the spinel is duplexing with CaS or CaO. Similarly, not correcting for the spinel-bound Al makes inclusion plotting positions on the popular Ca(O)–S–Al(₂O₃) ternary ambiguous as to whether the non-CaS fraction of the inclusions is due to calcium aluminates or spinel. One solution to this representation problem is to project complex inclusion compositions into the simplified ternaries not from chemical component apices, but from the compositions of the most abundant phases themselves. This suppresses spurious apparent variations in the simplified ternaries as variations in off-ternary phase contents are by definition parallel to the projection. This principle is illustrated in Fig. 2: projecting from spinel (instead of MgO) into CaO–S–Al₂O₃ removes spinel-caused variations in that triangle; projecting from CaS into Al₂O₃–MgO–CaO removes CaS-caused variations in that triangle (while leaving variations in CaO content visible).

Fig. 2.

Sketch of the projection methods used for the Al–Mg–Ca–O–S subsystem of inclusion compositions. Top tetrahedron illustrates that inclusions lined up between CaS and spinel (MA) (i.e. spinels with varying amounts of CaS attached) project with variable apparent “CaO” contents when projected onto the base triangle by simple renormalization neglecting S. Lower tetrahedron illustrates the projection method eliminating off-ternary phase effects: a given multicomponent composition is projected from the non-ternary phases present onto the side face of interest (in our case, from CaS onto the oxide face, and from spinel onto the sulfide-calcium aluminate face of the tetrahedron).

We therefore plot our NMI data using a chemographic projection as explained in Fig. 2 to show the oxide/sulfide variation in our data. Inclusions are projected from CaS resp. spinel on the two sides of the tetrahedron, and the sides are folded open into a set of two triangles joined at the CaO–Al₂O₃ (calcium aluminate) edge. This double-ternary is illustrated in Fig. 3 with the plotting positions of pure phases marked. As the plot consists of projections into two adjoined ternaries, every inclusion composition is plotted in both triangles simultaneously, and their actual compositional variation in the multicomponent inclusions can be evaluated from the two triangles. As the projection removes non-ternary variations, both triangles can also be directly compared with the relevant phase diagrams and the thermodynamic state of the inclusions at process conditions deduced. The ternaries are drawn in mole units, for easy placement of phase compositions, and the mathematical expressions for the respective projections are indicated in the corners: (Ca–S) instead of Ca in the oxidic triangle, (Al2–Mg) instead of Al2 in the oxide-sulfide duplexing triangle. It should be noted that the presence of other sulfides than CaS will cause projected inclusion compositions to fall outside the triangles. This simply follows from the mathematical expressions for the projection. An MgS inclusion, plotted into the “oxidic” triangle, would have a mole value for (Ca–S) of –0.5 and therefore plot towards the upper right outside of the triangle – correctly indicating that there is no oxidic part to that inclusion. Similarly, MgS would have a mole value for (Al2–Mg) of –0.5 and plot to the lower left outside of the left triangle, correctly indicating that there is no CaS-oxide duplexing in this inclusion. In contrast, since this diagram is only a projection from the Al–Mg–Ca–O–S subsystem, it does not chemographically correct for the the presence of nitrides (i.e. the projection is simply from N into this subsystem). Consequently, AlN appears simply as “Al” on this diagram and plots identical to Al₂O₃. It should be noted that the diagrams, drawn in molar units, require the exchange vector SO_–1 instead of the component S at the sulfur-rich apex, to keep stoichiometry consistent (CaO + SO_–1 = CaS).

Fig. 3.

Double-ternary plot sketch (on the left) together with plotting positions of relevant pure phase compositions (on the right) chosen for representing the inclusion composition variation of the inclusions in this NGO electrical steel. MgS and MnS plot outside of both triangles towards the upper right and lower left. Expressions for plotting coordinates are given in the corners: (Ca–S) instead of Ca in the oxide triangle, Al2–Mg instead of Al2 in the sulfide-oxide triangle. All ternaries are in molar units. Black lines in left triangle are tie lines between CaS and the Calcium aluminates on the Ca–Al axis. Every individual inclusion is plotted simultaneously on the left and the right triangle. AlN plots identical to Al₂O₃.

To graphically represent the nitride part of the inclusion compositions, we choose a quadrilateral Al₂O₃–MgO–AlN–Mg₃N₂ diagram, which is a section through the Al–Mg–O–N subsystem. Duplexed oxides and nitrides are coplanar in this section. The sulfide part of inclusions can not be plotted herein, and it serves mainly as a supplementary diagram to the double-triangle illustrated in Fig. 3. It should be noted that this oxide-nitride diagram does contain the considerable potential nitrospinel solid solution series (spinel MgAl₂O₄ to Al₅O₆N, complete miscibility at 1750°C).¹⁰⁾

To help with analyzing that plot, examples of individual inclusions are shown together with their respective plotting positions on the coupled triangles in Fig. 4 for Heat A, and Fig. 5 for Heat B.

Fig. 4.

Examples of individual inclusions in Heat A, illustrating the actual phase nature of various significant parts of the compositional variation of inclusions. Phases are labeled, and individuals a) to d) are placed on the Al–Mg–Ca–O–S double projection. Phase name CaA means “calcium aluminate” (quenched slag).

Fig. 5.

Examples of individual inclusions in Heat B, illustrating the actual phase nature of various significant parts of the compositional variation of inclusions. Phases are labeled, and individuals a) to d) are placed on the Al–Mg–Ca–O–S double projection. Phase name CaA means “calcium aluminate” (quenched slag).

4. Results and Discussion

4.1. Al–Mg–Ca–O–S Subsystem

The compositional variations of the NMI in Heat A and B during the LF process are shown in Figs. 6 and 7, respectively, using the double-triangle projection discussed above. As can be seen, the inclusion compositions shift rapidly during the processing steps. For both heats, the state before ladle treatment is comparable. In both the NMI assemblage is dominated by alumina, as is expected for Al-killed steel. Heat A has a marked content of aluminous spinel in the inclusion assemblage (as can be seen from the spread of the NMI compositions between the alumina corner and the theoretical MA spinel composition in Fig. 6). In both samples, remains of alumina clusters are observed. No AlN formed during the solidification of steel is observed. A more detailed NMI analysis is below.

Fig. 6.

Compositional variation of the inclusions in Heat A, successive samples along-process, as presented on the Al–Mg–Ca–O–S double-ternary projection. Colours of inclusion points are logarithmically scaled to their size (cross sectional area, see colour legend). Red star is ladle furnace slag composition in Heat A.

Fig. 7.

Compositional variation of the inclusions in Heat B, successive samples along process, as presented on the Al–Mg–Ca–O–S double-ternary projection. Colours of inclusion points are logarithmically scaled to their size (cross sectional area, see colour legend). Red star is ladle furnace slag composition in Heat B.

In the beginning of ladle furnace treatment the inclusion assemblage changes dramatically in response to FeSi and Al wire alloying, buildup of artificial slag, and stirring. The slag composition (Table 3) is given on Figs. 6 and 7 for comparison (red stars). Although the heats are not deliberately Ca-treated, the change in inclusion compositions is comparable in nature to that seen during Ca treatment of steels.^11,12,13) The source of Ca may be suspected to be Ca impurities in the alloyed FeSi raw materials (the actual Ca/Si in FeSi is 0.0005–0.001), which can produce 12–24 ppm of Ca in the fully alloyed heats. However, we observe both in the lollipop bulk-sample spark-OES data (Table 2), as well in the quantification of the inclusion (Table 5) that actual Ca value is higher than 24 ppm. This means that Ca impurity in FeSi alloy is not entire Ca source for the NMI assemblage during ladle treatment. The second evident source is the artificial slag, which is used to refine the steel. As in the case of the NMI in Ca treated steel,¹³⁾ the initial alumina NMI assemblage is destroyed immediately with the beginning of ladle furnace treatment, and replaced by a CaS–Ca aluminate–spinel assemblage.

The evolution of NMI in heat A and heat B shows dramatic differences. In Heat A, the calcium aluminate apex of the reformed inclusion assemblage in the vicinity of C3A (i.e. on the CaO-rich side of the liquid window of CaO–Al₂O₃–MgO system) appeared already shortly after the beginning of the LF treatment (Fig. 6 sample A2). In fact, this calcium aluminate composition corresponds very closely to the composition of the artificial LF slag (red star in Fig. 6). Spinel is only seen as a relict phase with compositions significantly higher than 0.5 (atomic) in Mg:Al, indicating the destabilization of spinel already at this early phase. Consistent with this, inspection of the actual large, emulsified slag droplets at the calcium aluminate apex (Fig. 4(a)) shows them to be MgO and CaS saturated. At the end of the ladle treatment of this heat (sample A3), no calcium aluminate (liquid) remains to the inclusion assemblage, which consists entirely out of CaS, MgO, and relics of the earlier destabilized oxides (MgAl₂O₄, CaO) (Fig. 6). The AlN-dominated quench assemblage appears in both heats as soon as Al is highly alloyed and is clearly different from the process inclusions in terms of particle size.

In the case of heat B, the initial inclusion is composed of small amount of Al₂O₃. The appearance of spinel is especially marked in heat B which did not have significant spinel before LF treatment, and itself confirms the involvement of mass transfer from slag to the NMI in molten steel. The reformed inclusion assemblage then develops with progressing ladle treatment towards a CaS–Ca aluminate assemblage in heat B. In heat B, the NMI at the end-of-ladle treatment shows a compositional apex corresponding to liquid window calcium aluminates. The development of a calcium aluminate liquid in the inclusions is also directly observable as these tend to be the largest and droplet shaped inclusions, see for an example Fig. 5(a). However, the NMI overall are not fully comparable to Ca treated NMI, as there is a quench population of AlN (as well as AlN overgrowths on process inclusions) and in some measure MnS and MgS, as soon as Al is highly alloyed, causing the appearance of large off-triangle data clouds that are not seen when Ca treated NMI are similarly plotted.

Contrasting the inclusion development in heat A and heat B shows that the development follows a comparable reaction pathway at the beginning of Ca treatment (A2, B2 in Figs. 6, 7, respectively). The NMI in heat A appears to be far more progressed towards destruction of slag and destabilization of aluminate than heat B where it only progresses to liquid window calcium aluminate compositions comparable to Ca treated steels. Consequently, although the overall alloy composition of the steel itself is very similar, the character of the NMI carried to the caster is different: liquid calcium aluminates with CaS and spinel in the case of heat B, only CaS and MgO without liquid phase in the case of heat A. This can give rise to very different final inclusion in casting, as can be seen in samples A4 and B4 of Figs. 6 and 7. In both cases, the state of the NMI in the last stage does not change dramatically, but shows a tendency to decrease in numbers and increase in individual grain size, indicating ripening. In heat A, where the steel has ended slag-undersaturated, some reappearance of larger spinel inclusions is seen; however, given the low absolute number densities it is not sure whether this is a reversal of the original reactions. Again in heat A it is remarkable that quench sulfides (MnS, MgS) which did appear in A2, but not in A3, reappear in A4; this may indicate incipient resulfurization.

4.2. Nitride Occurrence

The occurrence of nitrides in relation to the process is shown in Fig. 8, representing the coplanar Al₂O₃–MgO–AlN–Mg₃N₂ section through the Al–Mg–O–N system. This plot is shown only for samples A2-4 and B2-4, as there is no nitride occurrence in the lollipop samples before LF treatment. As seen in Figs. 4 and 5, the AlN, which is formed on quench, is duplexing with the preexisting process inclusions. The significant difference between both heats lies not in the occurrence of AlN, but in the amount of shift of the inclusion population towards the magnesium rich side – pronounced in heat A, much less pronounced in heat B. Even though the assemblage in heat A (sample A3) is MgO saturated, the absolute amount of MgO is not large enough to give rise to a MgO–AlN duplexing line in Fig. 8. Instead, as can be seen in Fig. 4, most AlN grows on the CaS rich inclusions remaining in this heat after the demise of the calciumaluminate slag in the later stages of ladle treatment.

Fig. 8.

Compositional variation of the inclusions in the diagram Al₂O₃–MgO–Mg₃N₂–AlN, which is a coplanar section through the Al–Mg–O–N ternary. Left column heat A, right column heat B. Only samples A2–A4 and B2–B4 are plotted as nitrides do not occur in samples A1 and B1. Grey line is the theoretical mixing line connecting oxide spinel (MA) and nitrospinel Al₆O₅N.

4.3. Development of NMI Mass Balance

The mass balance (amount of elements bound to the inclusions in each sample, in ppm contribution to the bulk composition), calculated according to both methods described above, is given in Table 5 and the data are plotted in Fig. 9. Both sets of numbers have errors arising from their associated assumptions (average density for method 1, phase model for method 2) and these assumptions cause the discrepancies seen in Fig. 9. In general, it can be expected that the numbers from the rigorous phase model inversion method are more precise. Their actual precision can be seen in comparing the inclusion-bound element data to the bulk-steel compositional data measured by spark-OES (Table 2). The agreement for Ca is excellent – within a few ppm for all samples. This agreement for Ca is only achieved through quantification of the full inclusion size range and it would not be nearly as good if only inclusions > 1 or 2 μm were summed. This indicates that the disagreement between NMI based Ca data and bulk-steel analysis Ca data that is noted in the literature – especially with regard to Ca treatment inclusions – is largely due to incomplete sampling of the NMI population by disregard of small inclusions.

Fig. 9.

Evolution of the NMI mass balance over time during processing of Heats A, B. Plotted is the amount of each element that is contained in the inclusions in each steel sample. Process steps: 1, 2, 3, 4 correspond to before LF, early and end of LF treatment, and tundish. Filled squares: Heat B, white diamonds: Heat A. Both numbers estimated from homogeneous-inclusion approximation as well as calculated from phase model inversion are plotted for each sample (refer to Table 5). For O and N, only numbers from approximation method are given.

The data shown in Fig. 9 illustrate the strong contrast between the two heats along the processing route, in spite of the similarity of the overall bulk-steel composition. Heat A shows a much higher load of CaS (Ca and S) at the beginning of ladle treatment, and a consistent and ongoing decrease of O up to the tundish. In contrast to this, heat B shows inclusion bound O that remains high at all times in the ladle furnace. Even though this heat shows comparable uptick in S and N during the ladle treatment stage, both then decrease again towards the tundish. Consequently, even though the overall absolute inclusion bound Al profile of both heats is quite similar, the character of this Al in the final as-cast steel is rather different – nitridic in Heat A, oxidic in heat B. One remarkable feature of the NMI quantification described here is its ability to track the absolute amount of Mg taken up in the steel during ladle treatment (Fig. 9), which reaches levels of 3–5 ppm before decreasing again towards the tundish.

From the general steel composition, it is clear that AlN is not stable at liquid processing temperatures in the ladle furnace, and forms in the samples only upon quenching. Data shows that there is a clear size limit to the occurrence of AlN; this break in sizes occurs roughly in the vicinity of the 0.75 μm² limit that we have employed here as the dividing size between the analytical runs for small and for big inclusions in samples A2-4 and B2-4. Thus, approximately, it may be possible to quantify separately the amount of elements (ppm per bulk composition) that is bound to inclusions actually present during the processing steps (ladle furnace), and the amount of elements that is bound to quench inclusions specifically. This is a useful distinction as it potentially allows the direct measurement in steel samples of elements dissolved in the actual steel liquid during processing, through quantification of their solidus and subsolidus precipitates. This is especially interesting for elements like Ca, Mg, for which the actual solubility in steel during processing is controversially debated, and for volatile elements such as S, O and N, whose activities control the secondary metallurgical reactions themselves. However, such a breakdown involves a detailed analysis of the size distributions of the NMI in addition to their compositions. As size distributions by themselves require detailed methodological discussion,^14,15) we leave the development of such quench-precipitate based process liquid analysis to a subsequent paper.

5. Summary

In summary, we have described methods to characterize the particle content (both process NMI and quench inclusions/precipitates) by automated FEG-SEM with full EDS characterization of the particles down to ~0.08 μm². The automated inclusion analysis (AIA) described here can characterize complex particle populations consisting of mixed oxides, sulfides and nitrides both qualitatively and quantitatively. We have developed a chemographic projection method that allows the actual phase makeup of the inclusions to be assessed from diagrams that can be directly compared with the respective phase diagrams. EDS quantification of inclusions down to 0.08 μm² is possible, but requires careful optimization of the quantification method. While progress on this issue can be expected, we find that matrix subtraction methods are able to deal with high-alloyed background steel, and are indeed indispensable for the quantification of small inclusions. Quantitatively, the inclusion populations can be converted into the absolute amount of the various elements (metals and volatiles) that is inclusion bound. Together with a process-metallurgical evaluation of the inclusion generations, these data can serve as fundamental input for models to assess reactions between the steel liquid, the inclusion populations, and other reservoirs, in order to develop optimised steelmaking procedures for this steel.

Acknowledgements

We want to express our gratitude for stimulating discussion to a large number of colleagues at Tata Steel Europe, especially to S. Van der Laan, F. Mensonides, and for technical technical support, especially F. van der Does. We express our gratitude to Tata Steel Europe, and the OX2 plant at IJmuiden Works for support of the study and permission to publish these results. We also thank I.-H. Jung, McGill University, Canada and W. Crama with Tata Steel Europe for discussing aspects of the ideas expressed here. Any errors remain our own.

References

• 1)   A  Chaudhury,  R  Khatirkar,  N.  Viswanathan,  V  Singal,  A.  Ingle,  S.  Joshi and  I.  Samajdar: J. Magn. Magn. Mater., 313 (2007), 21.
• 2)   D.  Steiner Petrovic: Mater. Technol., 44 (2010), 317.
• 3)   D.  Steiner Petrovic,  B.  Arh,  F.  Tehovnik and  M.  Pirnat: ISIJ Int., 51 (2011), 2069.
• 4)   T.  Taniguchi,  N.  Satoh,  Y.  Saito,  K.  Kubota,  A.  Kumagai,  Y.  Tamura and  T.  Miki: ISIJ Int., 51 (2011), 1957.
• 5)   H. A.  Kramers: Philos. Mag., 46 (1923), 836.
• 6)   F. H.  Schamber: Proc. 8th Nat. Conf. on Electron Probe Analysis, Microbeam Analysis Society, New Orleans, (1973), 85A.
• 7)   G. F.  Bastin and  H. J. M.  Heijligers: Scanning, 12 (1990), 225.
• 8)   A.  Okayama and  T.  Nishi: ISIJ Int., 51 (2011), 2064.
• 9)   F. S.  Spear: Metamorphic Phase Equilibria and Pressure-Temperature-Time Paths., Min. Soc. Am. Monograph, Mineralogical Society of America, Washington, (1993), 118.
• 10)   A.  Granon,  P.  Goeuriot,  F.  Thevenot,  J.  Guyader,  P.  L’Haridon and  Y.  Laurent: J. Eur. Ceram. Soc., 13 (1994), 365.
• 11)   N.  Verma,  P.C.  Pistorius,  R. J.  Fruehan,  M. S.  Potter,  H. G.  Oltmann and  E. B.  Pretorius: Metall. Mater. Trans. B, 43 (2012), 830.
• 12)   L.  Holappa,  M.  Hamalainen,  M.  Luikkonen and  M.  Lind: Ironmaking Steelmaking, 30 (2003), 111.
• 13)   E.  Zinngrebe,  P.  Seda,  S.  van der Laan and  B.  van Arendonk: Proc. 5th Int. Cong. on the Science and Tech. of Steelmaking, (2012), CD-ROM.
• 14)   E.  Zinngrebe,  C.  van Hoek,  H.  Visser and  A.  Westendorp: ISIJ Int., 52 (2012), 52.
• 15)   M. A.  van Ende,  M.  Guo,  E.  Zinngrebe,  B.  Blanpain and  I.-H.  Jung: ISIJ Int., submitted.