What are the involved phenomena in steel slag filters?

When this project was started, it appeared from the literature, that slag was leaching CaO, resulting in a pH rise in the filter. It was also reported that phosphorus was mainly removed by hydroxyapatite precipitation in steel slag filters following pH rise. These two observations, quite simple ones, remained the basis of the model. The challenge was to model correctly these two phenomena.

The first try for CaO slag leaching modeling was to understand slag composition. Major compounds composition (as Ca content (%), Fe content (%), etc) would not be appropriate for defining the leaching behavior, as calcium oxides may be present as various complex oxides with various properties. Real slag calcium oxide is therefore not well represented by CaO as it is divided into several long-name complex calcium oxides, with different solubility. The incidence of this for wastewater treatment modeling was clear in an excellent study of Kostura et al. (2005), who associated the buffering capacity of slag with its mineralogical composition via acid neutralization capacity curves. One possibility was to model accurately every mineralogical calcium oxide, with its individual properties (as solubility or kinetic dissolution rate). This strategy would reproduce the exhaustion behavior of slag, as fast and readily soluble oxides would dissolve first, then slower oxides would follow. This avenue appeared too complex and risky (experimental data would not easily follow predicted behavior). A second avenue was the intra-particle diffusion model (Kostura et al., 2005), interesting because it considered slow release of deep oxides from grain core, but it did not consider slag exhaustion, which is a crucial aspect of slag filters.

Following these observations and literature review, the hypothesis that readily soluble oxides are leached first, followed by slower-release oxides, was proposed. The objective was to develop an empirical approach for representing the oxide complexity of slag, without needing accurate measurement of slag oxide mineralogical composition. This approach would involve kinetic rates like those used for biological wastewater treatment modeling (Metcalf & Eddy et al., 2014).

Saturation functions used for biomass growth were especially interesting as a starting point, because slag is leaching until a saturation state is reached. The turning point came with the first batch test simulations (paper #3), when a methodology was proposed for simulating and controlling slag aging with acid baths, then repeating the CaO leaching parameter measurements. This principle was the basis of the experimental protocol for exhaustion function measurements, proposed in paper #3, and validated in paper #4. Exhaustion functions of the P-Hydroslag model are an empirical (and easily measurable) manner of representing slag calcium oxide complexity, and exhaustion behavior. Note that slag mineralogical analysis was performed in paper #2, while readings and findings related to slag mineralogy were still going on. When the method for exhaustion functions was developed, mineralogical analysis of slag was abandoned, explaining why this analysis is not part of papers #3 to #4.

The first literature review work for hydroxyapatite precipitation was reading a master thesis of a former project related to steel slag filters (Forget, 2001). This work described adsorption isotherms (Fetter, 1999) and possible surface removal mechanisms. From the beginning, it was clear that adsorption isotherms would not be suitable for steel slag filters modeling, because hydroxyapatite precipitation is a volume-based and not surface-based phenomenon. P removal mechanisms were roughly sorted into sorption for neutral-pH materials, and precipitation for alkaline pH materials (Baker et al., 1998).

Three important features were developed for hydroxyapatite precipitation: solubility of fine particles, crystal growth and hydroxyapatite structures. The equation of fine particles solubility was found while reading about precipitation in Stumm and Morgan (1996) for a literature review of hydroxyapatite solubility. After studying more deeply this equation, it was found that considering the small hydroxyapatite size in steel slag filters, this equation would explain alone the apparent supersaturation observed in previous studies (Baker et al., 1998), and also in data from paper #2. It was concluded that steel slag filters are not supersaturated with hydroxyapatite, they are equilibrated with a hydroxyapatite phase much more soluble then tabulated (bulk) hydroxyapatite, because of crystal smallness.

In the former master work of the candidate, crystal growth was observed in steel slag filters with limited XRD crystal size measurements (Claveau-Mallet et al., 2012). XRD was preferred to sequential extraction (Audette et al., 2016) for crystal characterization as it is possible to measure crystal size with this method. Following this work, a more exhaustive crystal sampling campaign was performed in this Ph.D. work (paper #2), with the objective to confirm crystal growth as a removal mechanism. In this project (paper #2), the effect of influent water composition on crystal growth was also confirmed, as two different wastewater influents were used. These observations led to the use of precipitation rates from the mineralogy science (Oelkers et al., 2009), as they were defined using the saturation index (therefore water composition). The project resulting in paper #2 also helped to understand connected precipitation issues involving metals and fluoride. Metal precipitation more especially helped to understand the critical role of pH in a precipitation process.

Development steps for fine-particle solubility and crystal growth were completed at the end of the first year, and were included in paper #3 and complementary paper.

At the beginning of this Ph.D. project, it was already outlined that hydroxyapatite structures played a role in phosphorus removal mechanisms, as an effect of crystal organization on efficient volume use in the filter was indirectly observed in a former project (Claveau-Mallet et al., 2012). This work resulted in this hypothesis: water velocity affects crystal organization, and crystal organization affects the efficient use of the slag filter volume by clogging. Later, in 2012, a crystal precipitation study conducted by the candidate (not presented in the thesis) confirmed this hypothesis with SEM observations of crystal structures on used slag grains. This work resulted in a qualitative schematic showing the effect of water velocity on crystal structures (Figure 9.1). An important step for understanding of HAP structures was a review of literature on implant biocompatibility, where a lot of information related to HAP precipitation is present, including crystal shape and organization under various organic or inorganic conditions. This research branch showed that HAP is extremely versatile in shapes and structures. It would be therefore wiser to find a way to represent it in a macroscopic way instead of modeling crystal structures at small scale. This observation led to the development of the diffusion through a uniform thin crystal layer approximation. The first development on this thin crystal layer approximation was done early in the Ph.D. project, but it was included only in paper #4, because full mathematical development was completed in the third year, and it involved calibration work that was not possible in paper #3 and complementary paper.

Figure 9.1 : Representative SEM photography and schematic representation of different organization levels in filters (unpublished work)

9.1.2 How could phosphorus removal mechanisms in slag filters be transposed