19. Registration of Filtration Efficiency of Active or Reactive Filters in Contact with Steel Melt in a Steel Casting Simulator

A metal-casting simulator (Systec, Karlstadt, Germany) located since 2010 at the Institute of Ceramics, Refractories and Composite Materials in Freiberg was used. The system was designed to evaluate refractory materials in direct contact with steel melts. The system consists of a water-cooled, gas tight chamber, that is filled with argon to protect all components within (see Fig. 19.1). It is composed of a separated inductive-heated melting unit (150 kVA) with a crucible that can contain up to 100 kg of steel. A revolver system allows to introduce probes to measure the temperature of the melt as well as the dissolved oxygen. In addition, another device is used to immerse prismatic samples (with maximum dimensions of 25 × 25 × 125 mm³) into the melt and rotate them at 30 rpm. It is also possible to extract melt samples for chemical analysis. The revolver system is equipped with a small chamber, in which an almost oxygen-free atmosphere is obtained through vacuum and argon flowing. After contact with the melt, the sample is extracted and can cool down inside this chamber, which is essential to prevent oxidation of e.g. carbon-bonded specimens. Finally, a charging device allows to introduce alloying elements and synthetic slags into the melt.

A schematic diagram of the experimental setup of the metal casting simulator. It is equipped with a melting crucible, a tundish with an oxide crucible inserted into the carbon crucible, a gas tight chamber, a nozzle testing area, and water cooled copper molds.

After conditioning of the molten metal, the crucible is tilted and the melt is cast into an inductive-heated (100 kVA) tundish, which presents three nozzles in the basement (two for tests and one for safety reasons) and a stopper rod system to control the melt flow. After some time, the stoppers are simultaneously lifted and the melt flows through parallel channels. Each consists of a protective carbon rod with a CFC heating unit to achieve a reproducible temperature level and prevent the metal from freezing. From the testing channels, the melt flows into separate water-cooled copper molds for solidification. Both the tundish and the copper molds are constantly weighed, which allows to monitor the melt flow for each of the two testing channels.

All experiments performed in the steel casting simulator can be classified as dynamic experiments, i.e. where either the refractory specimen or the melt are in motion. This results in more intensive interactions compared to the static experiments, in order to approach industrial conditions. In the so-called “finger-test”, the samples are immersed and rotated for a defined time into the metal (or slag) melt. This is the easiest and fastest type of experiment. More complex experiments can be also performed, by casting the melt from the main crucible into a tundish unit and then through the parallel channels into the copper molds. Before each experiment, the steel casting simulator is evacuated and then filled with argon gas, in order to eliminate oxygen and protect the inductively coupled graphite crucibles required for the heating. In addition, to create defined alumina impurities in situ in the steel melt, 0.5 wt% of an iron oxide mixture was added. The commercially available product (Mineralmühle Leun, Germany) consisted of 75 wt% hematite and 25 wt% magnetite. After setting a temperature of 1650 ℃, the oxide mixture was added directly to the melt and an increase of the dissolved oxygen from approximately 10 ppm up to approximately 60 ppm was measured. To create the required endogenous alumina inclusions, 0.05 wt% of pure aluminum metal was added to the melt. Due to this deoxidizing step, the dissolved oxygen content of the melt decreased to the starting values. Through the reaction between oxygen and aluminum, finely dispersed aluminum oxide was formed. After the experiments, the solidified steel as well as the tested samples were thoroughly analyzed.

In spite the fact that both nozzles had approximately the same starting inner diameter, different results of casted steel mass versus time were obtained. After an initial phase of temperature adjustment and filling the test zones, differences in the mass flow could be registered from about 60 s (Fig. 19.5) [1]. The smaller amount of casted steel versus time in the case of the alumina-coated nozzle was caused by a more intense clogging of nonmetallic particles at the inner surface of the tested nozzle, which was macroscopically observed after the test. The clogging layer inside the alumina-coated nozzle consisted of two areas, a dense zone with thickness of about 35 µm and a coral-like zone with thickness in the range 60–80 µm. Investigations by SEM/EDS and EBSD of the clogged particles identified compositions such as spinel and alumina. In case of the mullite-coated nozzle, no remarkable difference of the steel flow with time was detected. Several samples were taken from different regions of the test nozzle. Depending on the position, different morphologies were observed. The coating of a sample taken from the upper part of the nozzle showed a highly porous morphology without any fine particles as present in the coating before the steel contact (Fig. 19.6a). The composition of the coating approached Al₂O₃ with some traces of SiO₂ and some elements from the steel melt such as manganese, iron, and chromium. In contrast, a sample taken from the lowest part of the nozzle (Fig. 19.6b) presented a partial delamination of the former coating from the alumina-carbon substrate and pores with size in the range of 20–50 µm. The former mullite coating consisted of dense parts composed of two phases, darker grains (under BSE) embedded in a light gray matrix, metal particles as well as closed and open pores. An adhesion of exogenous and endogenous particles could not be detected. EDS analysis showed that the gray particles had nearly Al₂O₃ composition with some traces of silicon. The lighter continuous phase consisted of silicon, manganese and aluminum with traces of sodium, titanium, iron, and magnesium. Element mapping demonstrated that the alumina grains were surrounded by silicon and manganese, while iron was mainly present as separate particles. The investigations of samples between 0 and 1 showed a mixture of the previous results, with less pronounced delamination and lower amounts of the Mn-Si-Al-containing phase. The nozzles used in the experiments were scanned by CT before and after the test in the steel casting simulator. Only the presence of small steel particles in the clogging layer allowed to distinguish between the alumina-carbon matrix and the decomposed coating, due to the enrichment of Mn and Fe in the decomposed mullite layer. Delamination and partial destruction of the former coating was clearly visible through CT. In conclusion, a mullite coating on a carbon-bonded alumina component is not suitable for applications at >1520 ℃ due to the decomposition of mullite. On the other hand, a pronounced clogging for the nozzles with the active alumina coating was observed. Correlating these clogging values to a type of “filtration efficiency” in means of more particles to be deposited on the ceramic collector surface (nozzle or filter), there was a first indication that at least a higher filtration efficiency is expected in case that carbon-bonded filters are coated with an alumina active coating.

A line graph plots flow rate in kilograms per second versus casting time in seconds. 2 lines for carbon bonded alumina nozzle and carbon bonded alumina nozzle with alumina coating follow an increasing trend with some fluctuations between 0.05 and 0.30 with respect to the casting time from 0 to 140.

2 microscopic images of the coating sample at 20 and 100 micrometers. a. The morphology of the alumina carbon nozzle surface has a high porosity. b. The morphology of the sample has a decomposed coating in the steel contact region.

The finger-tests used carbon-bonded alumina filters with active or reactive coatings. Despite the strong thermal shock during the immersion itself as well as during the rapid cooling process, the filter samples did not show macroscopic damage after the experiment. By comparing the color of the sample surfaces before and after the immersion test, it was observed that no oxidation occurred on the areas without direct melt contact. The change from black to grayish color (Fig. 19.7a and b) in the areas with direct steel contact was caused by several reactions, resulting in decarburization in areas near the surface due to the high solubility of carbon in the steel melt, the interaction between steel, the carbon, and the refractory oxide, and by the deposition of a newly formed phases.

3 macroscopic images of the coated and uncoated A l 2 O 3, M g O C, and alumina samples at 2 centimeters. Each sample has a porous structure with different colored regions.

It is worth to mention that the trends for temperature and dissolved oxygen during all performed immersion tests were nearly the same. Table 19.3 presents the values for temperature and dissolved oxygen measured before and after oxidation (the addition of an iron oxide mixture), after deoxidation with aluminum, and after an immersion test for 3 different coated filters [2]. It could be seen that the process of increasing and decreasing the oxygen content by the addition of iron oxide and deoxidation with the aluminum metal was sufficiently reproducible. However, carbon loss in the steel was measured by spark emission spectroscopy after the experiments, in comparison to the delivered steel quality.

Investigations of the filter samples by digital light microscopy delivered interesting results. In general, all filter surfaces were microcrack-free before the immersion procedure. After the test, newly formed and loose crystalline phases were found on the surface of uncoated carbon-bonded alumina filters. SEM investigations of carbon-bonded alumina filters after immersion in a steel melt showed agglomerated particles in various shapes, such as platelets and dendrites, which are typical for endogenous inclusions as reported by Dekkers [13, 14]. When testing carbon-bonded alumina filters, the material buildup observed after an immersion test consisted of different structures (from the center of a strut to the surface), reported in several publications [2, 8, 15‐19]:

A similar layer buildup was observed throughout all finger-test samples, provided the coating (if present) was porous enough to allow for diffusion of elements from and into the filter material. A thorough discussion of the interactions taking place during melt contact is given in the next section. Zone 1 represents the unreacted carbon-bonded alumina filter material, composed of approximately 70 wt% alumina with typical grain size of 0.5–0.8 µm (d₅₀) and 30 wt% carbon as bonding matrix. Zone 2 consists of alumina grains with typical size in the range 500 nm to 1 µm, partially sintered together into a porous layer with a typical thickness of 5–10 µm. Iron particles were observed at the border between the first 2 zones. The secondary layer (4) has a thickness usually in the 100–400 nm range, sometimes up to 1 µm. It seems to be totally dense even at high magnifications and it reflects the topography of the zone below. This secondary layer is extremely efficient for collecting inclusions with similar chemistry from the melt, and its roughness increases the wetting angle with the steel melt, possibly contributing to the deposition of inclusions too. On top of this layer, the clogging zone can usually be divided into two parts. The first one is dense and consists of relatively small plate-like particles. This is followed by a zone with larger, loose particles in complex shapes.

The first reactive coating consisted of MgO-C, which should generate gaseous Mg when in contact with a steel melt and in turn remove the oxygen by creating secondary MgO [2]. By means of digital light microscopy a dense, crack-rich layer was detected on the filter surface after the immersion test. In addition, the carbon-bonded coating turned to a grayish-white color due to reactions with the steel melt. Under SEM a dense, in situ-formed fine-grained and carbon-free MgO layer with some MgO particle agglomerates on the surface was followed by a porous (3) zone containing some carbon, primary magnesia, and iron droplets. EDX investigations of the elemental composition of the newly formed dense layer as well as the agglomerated particles showed that both consisted primarily of magnesium and oxygen. In addition, newly formed whiskers consisting of MgAl₂O₄ were found inside the hollow struts generated from the burnout of the polyurethane foam used as template during the filter production (see Fig. 19.9).

2 microscopic images of the M g O C coated sample at 50 and 5 micrometers. a. The morphology of the sample has a cracked surface. b. The morphology of the sample has a needle like structure.

Carbon-bonded calcium aluminates were also tested as highly reactive coatings [20]. After the immersion test, a lot of pores infiltrated with steel (white spots) were found over the CA2-C surface. The rest of the surface consisted of polyhedral particles partially sintered together (see Fig. 19.10). The EDS detector revealed Al, Ca, and O as main constituents of this zone. The CA6–C surface was smoother and did not show any major porosity with entrapped steel (see Fig. 19.11). The difference should be related to the different wettability and the greater potential for the melt to penetrate CA2 than CA6. In addition, gaseous calcium and/or suboxides may be produced from the reaction between the calcium aluminate and the steel melt, with the formation of pores, which would be readily infiltrated by liquid metal. Given the different amount of calcium oxide in the compositions (three times higher in one mole of CA2), this effect should be much stronger in the case of the CA2–C sample than CA6–C. As usual, the applied coating was still detected with a thickness of approximately 40 µm. In some spots, a thin secondary layer was also observed on which endogenous inclusions were collected. Thermodynamic calculations showed that the reduction of CaO from the coatings to Ca vapor should proceed at a slow rate.

2 microscopic images of the C A 2 C filter sample at 50 and 20 micrometers. a. The morphology of the sample as a fractured surface with voids. b. The cross section of the sample has a formation of 2 layers.

2 microscopic images of the C A 2 C filter sample at 20 and 5 micrometers. a. The morphology of the sectional view of the sample has a layer of clustered particles. b. The morphology of the sample has a thin layer with inclusions.

Nano-sized materials were also selected for their high specific surface area (thus reactivity) and used in combination with a binder to produce reactive coatings [8, 15‐17, 19, 21, 22]. The used water-based slurries had a low solids content, hence very thin coatings were obtained. Due to this, the high reactivity of the nano-sized materials and of the residual carbon from the binders, no residuals from the nano-based coatings were detected on any filter after a finger-test, even for very short (10 s) immersion time. For this reason, after contact with a steel melt, the surface of nano-coated filters showed exactly the same features as uncoated carbon-bonded alumina filters (see Fig. 19.12). However, differences in the development of the dense and loose collection zones (5 and 6) were observed, with the nano-coated filters generally showing an increased clogging for the same testing time. It can be assumed that the presence of a nano-based coating before the test resulted in stronger interactions with the steel melt and a faster formation of the secondary (4) layer, which in turn promoted the collection of endogenous inclusions from the melt.

2 microscopic images of the nanocoated sample at 50 and 2 micrometers. a. The morphology of the sectional view of the sample has 5 layered structures. b. The morphology of the sample has a flake like layer at a size of 209 nanometers.

Regarding active coatings, pure alumina was applied by spray coating at first [2, 18]. Such filters were treated at 1400 ℃ to allow for sintering of the alumina coating. The coating thickness varied between 10 and 100 µm. The determined cold crushing strength of coated filters was (0.58 ± 0.08 N)/mm², which is about double that of uncoated ones (0.27 ± 0.04 N)/mm². After the experiment, this coating was still present on the filter surface, although cracks due to thermal shock were detected. They were the more pronounced the longer the immersion time. Frozen steel in the form of droplets was also found as usual. From SEM investigations, the secondary layer (4) with a thickness of a few hundred nm was already detected after a 10 s immersion. Filters immersed for 60 s showed plate-like particles and also fine inclusions in clusters on top of these in some areas (Fig. 19.8). After 120 s, the clusters completely covered the investigated filter surface. EBSD investigations of the thin secondary layer suggested that it consisted of tiny crystals embedded in an amorphous phase. Plate-like particles consisted of trigonal alpha-alumina. Grain orientation within these particles was determined, confirming that they did not grow directly from the coating material but more likely consisted of endogenous inclusions that sintered together due to the high process temperatures. Furthermore, carbon-bonded alumina filters were coated using the flame spray technique, in which a ceramic feedstock is melted at approximately 3160 ℃ and a dense coating (90 µm) is produced. Very high cooling rates (10⁶ K/s) resulted in metastable cubic γ-alumina that was frozen at room temperature. Another difference with cold spraying, thermal spray leads to inhomogeneous coatings due to the difficulty in reaching the filter core. It was observed that the oxygen content of the steel remarkably decreased after immersion of the flame-coated filter, indicating a strong interaction with the melt. Moreover, the alumina coating was extensively covered by a continuous crystalline layer with an average thickness of 45 µm. Compared to a cold-sprayed filter, the microstructure was overall much denser, with plate-like particles covering nearly the whole surface. In this study, no evidence of an in situ formed thin layer was found. It is not surprising that this layer was absent as its formation is promoted by raw material impurities like SiO₂, Na₂O, and K₂O. During the flame spraying process, the raw materials were heated above 3000 ℃. At this temperature, these impurities evaporate and hence, are not transported to the filter surface, leading to an alumina coating with high purity.

Besides the investigation of ceramic components after contact with molten metal, the analysis of the solidified steel in terms of remaining inclusions was of critical importance for the subproject C01. Standard chemical analysis by means of spark emission spectroscopy was performed first, followed by automated scanning electron microscopy (ASPEX).

Regarding the steel chemistry after contact with carbon-bonded components, the first experiments involving both uncoated and alumina-coated nozzles already showed some interesting trends [1]. In particular, the Al content decreased in all ingots, suggesting the possible generation of endogenous alumina inclusions. In addition, other elements such as C, Mn, Ni, S and P were also present in lower amounts after the experiments. Further studies involving the use of carbon-bonded alumina filters (Table 19.4) confirmed the trend of aluminum and carbon loss, regardless of the presence and type of coating [2, 15, 18, 19]. MnS inclusions were found in some cases on the filters’ surface, as well as regularly in the solidified steel [16, 19, 23, 24]. Sulfides, and particularly MnS, are common impurities in steel and can explain the observed depletion of both elements.

The first ASPEX investigations were used to compare the performance of uncoated carbon-bonded alumina versus nano-coated filters, due to the promising clogging results. One steel melt was used as a reference, in which no filter was immersed. Both uncoated and nano-coated filters showed positive effects: the density of particles with area < 20 µm² and especially < 3 µm² decreased significantly [19]. On the other hand, the amount of inclusions larger than 50 µm² slightly increased. The results were also supported by light microscopy investigations. The nano-coated filter performed significantly better than the uncoated one, especially for a 5 min immersion time. Regarding the chemistry, the majority of inclusions consisted of alumina, followed by MnS (due to the high S content of the steel) and Mn-spinel. Very good filtration efficiency (up to 90% were registered). Despite the good performance of nano-coated filters, these also generated new inclusions (classified as mixed silicates) which were otherwise not detected for the other experiments. In addition, especially after 10 s immersion, the coated filters produced a high amount of coarse particles, which are most detrimental for the fracture toughness of steel. This particular behavior of nano-coated filters was observed in other studies as well [17, 25] (Tables 19.5 and 19.6).

Filters with cold-sprayed alumina coatings were tested for different immersion times, from 10 up to 120 s. The proportions of different chemical as well as size classes remained similar for all experiments. Filtration efficiencies in the 50–70% range were registered [18]. Interestingly, with increasing immersion time, there was a trend to more inclusions remaining in the steel, together with a growing amount of SiO₂ particles. On the other hand, neither Ca-aluminate, Mg-spinel, Al– Mn–Mg–Fe–Ca–silicate nor CaO–CaS type inclusions were identified in the steel samples. In addition, the density of inclusions > 50 µm slightly increased when immersing alumina-coated filters.

A further study involving both nano-coated and alumina-coated filters showed that the majority of inclusions detected by ASPEX was located in the 0.1–20 µm² area region, with a peak at 1–3 µm². The use of a filter had always a positive impact on the steel purity, but the effect decreased with increasing immersion time [17]. In addition, particle growth was observed again for the nano-coated filters.

A study involving carbon-bonded calcium aluminate coatings showed the same trends for inclusion size distributions [20]. Most of the detected particles were classified as alumina as expected. Filtration efficiencies close to 100% were registered. However, the steel samples in this case were extracted directly from the steel melt (before and after filter immersion), so the results cannot be directly compared.

SEM investigations on nano-coated and CA2-C coated filters showed very similar features to what reported previously, since the conditions during the immersion were similar to those of simple finger-tests. In case of the active filters instead, at the end of the process, some of the steel did not manage to escape the macropores and froze. As a consequence, any remaining melt also solidified directly on top of the filter, which could hardly be analyzed. Generally, the bottom part showed the flame-sprayed alumina coating with collected inclusions and steel droplets. Light microscopy pictures are presented in Fig. 19.13. The alumina coating survived the impact of the molten metal and the thermal shock for the most part, but in some areas it cracked and was lost into the melt or during handling after the experiment. A glassy phase apparently covered the alumina coating after multiple experiments. The bottom surface of the sandwich filter seemed quite rough, with a lot of inclusions over the alumina coating. Since the top portion of the filter consisted of uncoated carbon-bonded alumina, this material reacted on contact with the steel melt and the resulting endogenous inclusions were immediately forced through the bottom (coated) part, in a way that a high number of them was collected on the filter surface.

2 light microscopic images of C A 2 C and nano nanocoated filter sample at 2 millimeters. The morphology of the sample has a molten metal with a cracked surface.

Under SEM, the surface of the other three samples appeared for the most part very smooth (as suggested by light microscopy investigations), with some interesting formations on top. These usually consisted of steel droplets resting over inclusions with a different composition (see Fig. 19.14b). The EDS signal revealed the presence of glass-forming elements such as Si and P. In particular, about 70% of the smooth layer consisted of SiO₂, with the rest made up of alumina, Na₂O and minor amounts of other oxides. On the other hand, the inclusions most likely consisted of iron phosphate or a similar phase. These observations confirm the presence of a glassy layer on the surface which was speculated before. It was not possible to determine the thickness of such layer in cross section, since it merged with the alumina coating underneath. From multiple images from different filters it was estimated to be at least a few micrometers thick. Since only a thin slag layer is produced during the immersion tests, this layer most likely originated from the impurities contained in the refractory adhesive used to keep the active filter in place. Small grains can be observed within the glassy layer at high magnifications, but the fast cooling of the filter after casting of the steel melt probably did not leave enough time for a complete crystallization. This particular microstructure was observed for the first time on coated carbon-bonded filters in this work. Previous studies only focused on immersion experiments, in which no casting process was involved.

2 SEM images of the with and without nanocoated filter at 50 and 20 micrometers. a. The morphology of the sample has a rough surface with a distribution of spherical particles. b. The morphology of the sample has a steel droplet with inclusions.

As in simple finger-tests, with combined filtration the amount of carbon decreased considerably for every test (Table 19.7). Interestingly, when performing only the immersion (which is a much faster experiment), the final carbon content in the steel was similar. On the other hand, loss of Cr was much larger in this case. Some Mn was also lost in all experiments, and especially when no filter was immersed. The amounts of other elements such as M, P and S were not affected significantly. Si was lost in all experiments, contributing to the formation of the glassy layer on the active filters (see above) and also to the creation of complex inclusions. Aluminum was strongly depleted whenever a filter was immersed, which is in agreement with previous studies [20]. Most likely this element was involved in the creation of new alumina inclusions during contact between the reactive filter and the steel melt. When no reactive filter was immersed, the Al content was not remarkably affected. The total oxygen content increased for all experiments, and particularly for the combined filtrations.

From the ASPEX size classification, it is clear that all curves tend to follow a log-normal distribution (Fig. 19.15). The combined filtration experiments resulted in larger inclusions on average compared to the reference test. As expected, the majority of inclusions consisted of alumina due to the steel treatment before immersion of the reactive filter, followed by MnS. Some improvement in comparison to simple finger-tests performed under similar conditions were noticed. However, less inclusions were detected in the solidified steel in the case no reactive filter was immersed, i.e. when only casting through the active filter. During the combined filtration experiments, after extraction of the reactive filter, the crucible was immediately tilted and the steel melt cast through the second filter. In this way, any new inclusions which were not already collected at that point were cast together with the steel melt and did not have time to escape to the slag layer via flotation. For this reason, further experiments were performed, with pure alumina filters floating on the steel melt to promote inclusion removal.

A multi-line graph plots the particles per 100 millimeter square versus area root. 5 steel samples begin at (0.4, 0), follow an increasing trend, reach the peaks between 60 and 260, again follow a decreasing trend, and end at (100, 0). Values are estimated.

As mentioned above, one characteristic layer buildup was observed on almost all tested filters, regardless of the presence and type of coating employed. It appears that the in situ formed thin layer can be amorphous, crystalline (α-Al₂O₃) or a mixture of amorphous and crystalline, which depends on the level of impurities of ceramic raw materials, impurities in steel, and alloy elements [18]. Focused ion beam investigations detected only the structure of alpha alumina, despite the variable composition [9]. Other authors detected a vitreous phase on the surface of a decarburized Al₂O₃-C submerged nozzle after contact with steel [26]. They reported that this phase consisted of alumina, silica, and alkali oxides. According to the authors, the aluminum activity in steel increased with the amount of carbon dissolved. The resulting CO gas was then dissolved and provided oxygen for the reoxidation of Al and deposition of alumina. Regarding the clogging layers, plate-like particles clearly consisted of trigonal alpha-alumina. Grain orientation within these particles was determined, confirming that they did not grow directly from the coating material but more likely consisted of endogenous inclusions that sintered together due to the high process temperatures [18]. Small crystals as well as collected inclusions might act as nuclei for the polycrystalline structures observed on the filters’ surface.

In presence of a reactive or porous active coating, reaction products and gases can easily move to the filter/steel interface. Consequently, two mass transports probably contribute to the layer buildup observed in many studies: on the one hand, products (e.g., CO, alumina sub-oxides, gaseous Mg and Ca) originating from the carbon-bonded substrate, and, on the other hand, endogenous inclusions as well as dissolved elements from the steel. Regarding the actual reactions taking place, different mechanisms were proposed. Some authors claim that a carbothermic reduction of alumina (with iron as a catalyst) could provide the necessary alumina sub-oxides and carbon monoxide, which migrate through the coating and into the steel melt, remaining available for further reactions [2, 27, 28]. However, thermodynamic studies by Zienert et al. indicated that carbothermic reduction of alumina might not occur under the given experimental conditions [29]. They suggested, instead, a partial dissolution of alumina into the molten steel. No stable equilibrium between alumina and iron liquid is reached because of the constant production of CO gas from the dissolved oxygen and the presence of carbon. Al supersaturation in proximity of the filter surface results in the reprecipitation of alumina. It was observed that the secondary thin layer thickness and the polycrystal size did not increase with immersion time. Aluminum and oxygen showed a concentration gradient through the layer thickness, suggesting that the formation of this structure is at some point limited by diffusion [9]. This would explain why this layer always shows a thickness <1 µm, regardless of the type of functionalization (active or reactive) employed and the immersion time. In addition, no evidence of such a layer was found on filters with dense coatings prepared by flame-spraying. On the other hand, several studies showed that the amount of platelets or particles with complex shapes (“clogging phase”) which are collected on this layer slowly increases with the immersion time [17, 19]. The progress of clogging suggests that the interactions at the filter/steel interface still take place, but at a much slower rate. At some point in time, it is possible that dissolution and reprecipitation reach some kind of equilibrium. This would explain why filters delivered a worse performance with longer contact times. This is where the filter behavior changes from “reactive” to “active”, according to the definitions proposed by the CRC 920. Endogenous inclusions from the melt are now mainly attracted by the filter surface due to the similar chemistry. The increased roughness and specific surface benefit the filtration process.

From several immersion tests, the Al content as well as the amount of alumina inclusions decreased in all steel samples for short contact times. In addition, other elements such as C, Mn, Ni, S and P were also present in lower amounts after the experiments. With increasing immersion time (60 + s), a higher inclusion density was registered and the filtration efficiency dropped. This was probably due to detachment of particles from the filter surface due to the melt flow. Especially in the case of nano-coatings, a higher amount of large inclusions was detected after the experiments. Since oxides tend to behave as nucleation sites for sulfides, a lower oxygen content (induced by the nano-additives) resulted in larger sulfides and at the same time in a lower inclusion concentration [25]. Apart from the grain growth, the strong reaction with the MWCNTs could also explain the generation of silicates. Carbon and oxygen from the filter material may influence the Si activity in the steel, leading to local saturation and precipitation of silica/silicate inclusions.

In summary, the filters were most efficient during the reactive phase. Multiple studies dealing with the turbulent melt flow driven by the electromagnetic force in an induction furnace were performed within the CRC 920 [30‐33]. It was demonstrated that the immersion of a ceramic filter, depending on its permeability and position, has a considerable effect on the flow pattern. Moreover, the model was used to determine the filtration efficiency of ceramic filters. In comparison with the observed performance (30% up to over 95%) from the steel casting simulator experiments, low efficiency values were obtained at first [31]. However, the first simulations did not consider any reactions at the interface, nor the filter rotation. Further studies showed that the formation of CO bubbles on the surface of inclusions increases their rising velocity toward the free surface of the melt. According to the numerical results, about 30% of the starting inclusions are removed after 10 s. This can considerably increase the inclusion removal up to the observed levels and therefore improve the cleanliness of the steel melt [11, 34, 35]. On the other hand, when the bubbling effect of carbon-bonded alumina (due to evolution of CO gas) must be avoided, such as during a continuous casting process, the application of a dense flame-sprayed coating would be beneficial.