- Email: [email protected]
The effect of shear on gravity thickening: Pilot scale modelling
ARTICLE IN PRESS Chemical Engineering Science 65 (2010) 4293–4301
Contents lists available at ScienceDirect
Chemical Engineering Science journal homepage: www.elsevier.com/locate/ces
The effect of shear on gravity thickening: Pilot scale modelling Brendan R. Gladman a, Murray Rudman b, Peter J. Scales a, a b
Particulate Fluids Processing Centre, Department of Chemical and Biomolecular Engineering, The University of Melbourne, Victoria, Australia CSIRO, Mathematical and Information Sciences, Clayton South, Victoria, Australia
a r t i c l e in fo
abstract
Article history: Received 20 April 2009 Received in revised form 28 January 2010 Accepted 12 April 2010 Available online 18 April 2010
Mathematical models are potentially a valuable tool for the prediction of continuous gravity thickener operation. However, experience shows that existing mathematical models underestimate dewatering in thickeners for flocculated feed materials when predictions are made of either the underflow solids concentration for a given solids feed flux density or the maximum solids feed flux density achievable for a minimum underflow solids concentration set point. One reason postulated for this discrepancy is shear enhancement of sedimentation and bed dewatering as a result of aggregate densification. This process is not taken into account in conventional 1-D thickener models. A pilot scale column, operated at low bed heights without the addition of mechanical shear, produced results that compared well with 1-D model predictions. The effect of mechanical shear and/or greater bed height was to significantly enhance thickener performance relative to model predictions (as measured by underflow density or maximum solids flux density achievable for a nominated underflow density). An experimental method was developed that enabled shear to be incorporated into the suspension dewatering characterisation. The results suggest an order of magnitude increase in solid flux density can be expected under controlled shear conditions with polymer flocculated aggregates. The results also indicate that mechanical shear is not the only factor that can enhance dewatering, since higher beds, and hence longer residence times, also improve the achievable solids flux density. This is despite the fact that the thickener is operating in a regime that is predicted to be limited by the sediment permeability and not its compressibility. This suggests an additional mechanism must be at play in full scale operation and points a direction for further experimentation. & 2010 Elsevier Ltd. All rights reserved.
Keywords: Thickening Solid–liquid separation Shear Raking Gravity separation Aggregate densification
1. Introduction Faced with a growing demand for potable water, industry faces a greater expectation to conserve, recover and reuse process water. Gravitational thickening is a widespread process, used to concentrate the solid phase and clarify the liquid phase of a particulate suspension. The recovered water can be recycled within the process or further treated for environmental release. The appeal of gravity thickening is its ability to treat highly variable feeds with low solids concentrations, both economically and efficiently. When operated appropriately, thickeners can produce a concentrated underflow stream that is substantially higher in solids than the feed. The use of gravity as the force driving separation ensures a relatively low operational and energy cost. Simplistically, a thickener can be envisaged as a cylindrical tank with a central feed-well in which solid particles are brought into contact with a polymeric flocculant, causing them to aggregate. As the solids settle from the feed-well, and if the
Corresponding author.
E-mail address: [email protected] (P.J. Scales). 0009-2509/$ - see front matter & 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.ces.2010.04.010
solids flux density permits, a networked bed of solid particles forms at the base of the tank. This base is usually sloped and a scraper or rake aids the transport of thickened solids to a central pump well (the underflow). Clarified liquor exits through a peripheral launder at the top of the tank. In terms of water recovery, the formation of a bed is desirable as the compressive/consolidation forces transmitted in the bed aid sedimentation, increasing the solids concentration in the underflow and thus improved water recovery. Although there are other material properties that can be important in the thickened material (i.e. pumpability of the underflow, angle of repose of the tailings if sent to a Central Tailings Discharge), fast settling to achieve a given underflow solids concentration (i.e. highest throughput for a given water recovery) is usually the design factor that is considered important in conventional thickening. Despite the prevalence of gravity thickeners, their design is usually empirical and their operation often based on a combination of experience and folklore. Kinematic thickener models based on Kynch (1952) theory predict area requirements for thickeners operated at their throughput limit, i.e. where the underflow solids concentration is not sufficient to form a networked bed of
ARTICLE IN PRESS 4294
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
particles and consolidation is absent. However, they do not provide an accurate indication of the expected solids concentration or the rheology of material exiting the underflow for thickeners operated at lower solids fluxes. In mining operations, there are ongoing efforts to continually improve water recovery and to produce thickened tailings, often with a high yield stress (sometimes erroneously termed ‘‘viscosity’’) to aid waste solids disposal. In these applications, the aim is to predict, design and operate thickeners close to the rheological limit; namely the point where the raking mechanism of the thickener is torque limited and a higher solids output would compromise the flow or pumping of solids from the thickener. Phenomenological thickener models encompassing both sedimentation and consolidation are more practical than the classical kinematic models as they take into account compressive forces in the networked bed of particles. Thickeners operating with a networked bed of particles show underflow slurry behaviour that is non-Newtonian and characterised by a yield stress. The ability to predict dewatering is advantageous both in thickener design and operation. On a more fundamental level, modelling can provide a better understanding of dewatering at a microscopic scale and elucidate how flocculation and hydrodynamic conditions can be optimised for throughput, underflow density, etc. A number of workers have modelled the thickening process (Tiller et al., 1987; Landman et al., 1988; Garrido et al., 2003a,b; Usher and Scales, 2005) but very few have reported the output of their model against the output of full scale operational thickeners (Gladman et al., 2006). To achieve a useful comparison, the dewaterability of the feed slurry needs to be characterised via parametric rheological functions termed the compressive yield stress Py(f) and hindered settling function R(f) (or an equivalent set of dewatering parameter descriptors). These functions are key inputs for a phenomenological dewatering model. A review of current thickener operations shows that the majority are predicted to operate in a regime in which the underflow rheology is close to Newtonian or perhaps exhibits a low yield stress, a direct consequence of the torque limitation of the raking mechanism in most cases. In this mode, the maximum operational flux is limited by the suspension permeability, not by its compressibility. Therefore, the bed height and the compressive forces imparted as a result of the bed are not predicted to be the rate-determining step. Both operationally and from a design perspective, it is advantageous to push the rheological limit, but a lack of correlation between models and thickener performance presently makes this an empirical exercise. The lack of a quantitative design process for thickening results from the fact that when laboratory measured dewatering parameters are used as inputs, current phenomenological models consistently underestimate the actual solids flux density required to achieve a given underflow solids concentration by anywhere between 2 and 200 times (Usher et al., 2005). It should be noted at this point that this discrepancy is rarely if ever observed for solids that are not aggregated. It is common practice, however, for the solids fed to a thickener to be aggregated through the addition of polymer flocculant and as a consequence, most current dewatering practice does not correlate with best practice in the modelling field. The work described herein pertains to the case for a flocculated solids thickener feed. The ratio between actual and predicted solids flux density for a given underflow solids concentration is termed the permeability enhancement (PE) factor of the thickener and is in general a function of the operating conditions (feed flux, flocculant dosage, bed height, residence time, etc.). Although factors up to 10 are typical, some level of permeability enhancement is common for
most thickeners. To a reader unfamiliar with gravity thickening, this might seem an extremely poor correlation, although it reflects the current state of the art (Usher et al., 2005). Worth noting is that the predicted underflow solids concentration for a given solids feed flux density is usually predicted to a much better degree with these models than the achievable flux. However, for operational reasons, a minimum underflow solids concentration and a solids feed rate (t h 1) is usually specified and for design reasons the thickener size will be dependent on the settling flux density (t h 1 m 2). Consequently, predicting the maximum achievable flux for a desired underflow density is a highly desirable feature of such models. Pilot scale studies have produced more encouraging results than those obtained from plant measurement. They provide far greater control over the measurement conditions, helping reduce the uncertainty associated with a full scale process. Good control also allows effects due to raking, sloped walls and poorly operated feed-wells to be isolated. In an earlier study by the same authors (Gladman et al., 2009), model validation was considered using a pilot scale dewatering column and flocculated calcite. The work showed that the predicted solids flux density through the column for a given underflow solids concentration was underestimated by factors between 1 (i.e. accurate) and 10 times. The reason why full scale predictions show a greater deviation from the models is difficult to establish. One explanation is that the methods used to measure the dewatering parameters, namely laboratory batch sedimentation (Lester et al., 2005) and filtration tests (de Kretser et al., 2001) are not representative of the hydrodynamic conditions or timescales under which a suspension dewaters in a thickener. Another is that the levels of shear stress on flocculated aggregates are higher in the full scale process and are imposed for longer periods due to the longer inherent solids residence times and finally, that channel formation and other in-homogeneities are not taken into account in the models. Many thickeners employ rakes to transport material to a common discharge point. Several studies in the literature show a qualitative improvement in dewaterability as a result of raking the suspension (Vesilind and Jones, 1993; Farrow et al., 2000; Comings et al., 1954; Novak and Bandak, 1994; Johnson et al., 2000; Holdich and Butt, 1996) although the exact mechanisms by which rakes or pickets enhance dewatering are not well understood. Careful examination of the results revealed that the degree of improvement depends on the speed of the rake and the slope angle of the conical thickener base (Comings et al., 1954). Both higher rake speed and a steeper conical section were associated with higher underflow densities and since the thickener was generally permeability limited, this implies that rake speed and cone angle also improved the permeability (Usher et al., 2005). As material settles and flows in the region of the sloped surface or past a rake blade, the flocculated material is sheared, leading to the conclusion that shear enhances the permeability of networked, flocculated suspensions. Investigations of the effect of shear on dewatering generally conclude that a low level of mechanical shear improves the rate and extent of dewatering in gravity settling (Gladman et al., 2005). Wall effects have been shown to be important; settling tests in a conical geometry revealed that settling in a sloped vessel was faster and achieved a lower equilibrium height than the equivalent test in a straight walled vessel (Usher et al., 2005). However, these conclusions are usually qualitative and do not provide a framework for incorporating shear into dewatering models. To quantify the role of mechanical shear, shear rates were targeted to match those typically encountered in a raked thickener bed (Rudman et al., 2008) and Py(f) and R(f) measured
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
as a function of the volume fraction of solids and shear in a Couette device (Gladman et al., 2005). The results showed that shear increased both the permeability and gel point of the suspension significantly. By accounting for shear in the model, it is hoped that simulation results for full scale thickeners (and here, for a pilot column) will be closer to those observed in practice. To understand how shear affects dewatering and therefore better predict thickening behaviour, changes in Py(f) and R(f) will be measured at different shear rates in our standard batch settling characterisation technique (Lester et al., 2005). A pilot thickening column, described previously (Gladman et al., 2009) but with the inclusion of shear elements, will also be used to validate the results. Pilot rather than full scale experiments provide more control over experimental parameters and the flexibility to vary these parameters. The overall aim of the work was to try and provide a quantitative relationship between the batch dewatering of flocculated suspensions subjected to known shear stresses in the laboratory, a one-dimensional phenomenological thickening model and the output of the pilot column under controlled shear conditions.
2. Experimental methods 2.1. Feed characterisation The procedure used to characterise the dewaterability of the calcite feed has been described elsewhere (Gladman et al., 2009, 2005) and only a brief description will be provided here. The solid used in this work was calcium carbonate (Omyacarb 10). A high molecular weight anionic polyacrylamide/polyacrylate copolymer (AN934SH from SNF) was used to flocculate the suspension. Stock polymer solutions were prepared at 5 g L 1. Working polymer solutions (0.2 g L 1) were prepared each day and any left over solution discarded. The dose rate of the polymer flocculant on a dry solids basis was 40 g t 1. Calcite (10 w/w%) was flocculated using a linear pipe reactor into a Couette shear cell, comprising three concentric cylinders: the inner and outer cylinders were fixed and a third cylinder in between, rotated to provide shear. The suspension–water interface height was measured for three different shear rates: 0, 4.1 and 10.8 s 1 and the dewatering material parameters Py(f) and R(f) were measured (Lester et al., 2005). The shear rate range covered in the laboratory batch settling characterisation is noted to be higher than that able to be explored in pilot scale (in this instance) and full scale studies (Rudman et al., 2008). It has already been established that shear rates in excess of 10–12 s 1 are detrimental to dewatering for this material and that there exists an optimum shear rate at solids concentrations less than the gel point of around 4 s 1 (Gladman et al., 2005). The resultant Py(f) and R(f) from these batch settling tests were then used to model dewatering in a continuously operated pilot thickening column.
4295
the thickener diameter (and in the case of bed heights in the
cone section of the thickener, the diameter as a function of height); the solids volume fraction of the feed ff; the density of the solid and fluid phases and the bed height.
2.3. Pilot column A continuous thickening pilot column (dimensions presented in Table 1) was used in the continuous dewatering experiments. The 287 mm diameter column was modified from a previous study (Gladman et al., 2009) as follows:
the column height was extended from 2.2 to 5.2 m, and a motor driven rotating cylinder arrangement was installed to provide steady shear. The cylinders were 100 cm in height and located at a distance of 40 cm above the cone apex. Fig. 1 shows a schematic of the column in plan and cross sectional view. The outer cylinder was rotated by attachment to a stainless steel shaft driven by an overhead-geared motor. This provided shear with respect to the wall and the stationary inner cylinder. To create Couette flow, the inner cylinder was kept stationary by attaching it to a stainless steel outer sheath, fitted over the shaft. Some level of shear was also imparted in the gap between the shaft and stationary inner cylinder. The motor allowed the shaft to be rotated at speeds between 0 and 20 rpm. Shear rates at 20 rpm were of the order of 1.6 s 1. Although these values are somewhat less than those used in the sheared batch settling tests, Table 1 Pilot column dimensions. Height (m) Truncated cone height (m) Column diameter (m) Cone angle (deg) Diameter of truncated cone at base (m) Cone volume (L)
5.2 0.205 0.285 60 0.05 4.4
5m 2.2. Thickener modelling
1m Using the approach described by Usher and Scales (2005) a 1-D thickener model developed by Landman et al. (1988) was solved numerically in combination with flux limiting calculations proposed by Coe and Clevenger (1916). Inputs to the model were:
287 mm 125 mm 50 mm
constitutive relationships for the compressibility and hindered settling function as functions of the solids volume fraction for the feed slurry, Py(f) and R(f);
Fig. 1. Cross-section and plan view of tall column with Couette style shear device. The inner and 125 mm cylinders rotate to provide the shear.
ARTICLE IN PRESS 4296
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
they are still within the range for which noticeable enhancement of permeability and gel point was observed. The vertical position of the cylinders could be adjusted by loosening two screws, which attached the cylinders to the shaft/sheath, allowing the column contents to be sheared at any height.
PID FT Mains Water
Flocculant solution
FT Isolating Valve
2.4. Column experimental procedure
Overflow
PID Concentrated calcite suspension was prepared at 40% w/w using local tap water (low dissolved solids and a conductivity of 0.177 mS cm 1). An approximate volume of 3500 L of suspension was prepared for each batch by adding 1845 kg of calcite to 2770 L of water. An overhead stirrer mixed the suspension for 24 h prior to use. The concentrated slurry was blended with tap water to produce a 10% w/w suspension. The chosen solid flux density was 0.3 t h–1 m 2 or volumetric feed rate of 3.0 L min 1. The feed to the pilot column was flocculated using a linear pipe reactor with an internal diameter of 12.5 mm and length of 1.8 m. The flocculant was dosed at 40 g t 1 of solids or 64.8 mL min 1. Each of the flocculant, dilution water and slurry flows were regulated using a PID feedback control system. Three bed heights were investigated: 1, 2 and 4 m. For each experiment, the flocculated calcite was introduced to the column approximately 0.5 m from the top. After operating for several minutes, three zones became apparent: a clarification zone above the feed point, a dilute hindered settling zone and an opaque bed zone. Here, the bed shall be defined as the interface between the hindered settling zone and the opaque zone in which the settlement of individual aggregates could no longer be distinguished. The bed formed a well defined interface and could be readily detected by eye. It should be noted that this ‘‘interface’’ does not necessarily indicate the point at which the bed of particles becomes networked and able to support normal stresses but is simply the concentration at which the hindered settling of the suspension becomes commensurate with the downward solids flux created by the underflow pump (Gladman et al., 2009). After start-up, the bed height gradually increased as solids accumulated. The bed pressure was measured using a pressure transducer located in the conical base. When the bed reached the desired height, the pressure was recorded and this value used as the set point for the pressure control loop. The control involved a feedback signal from the transducer to a variable frequency motor drive with onboard PID control. Any deviation was corrected by manipulating the underflow pump speed. Online density meters recorded the density of the feed suspension and the underflow exiting the column. Flow rates, densities, temperatures and pressure were logged so that the cause of any perturbation or disturbance could be traced. A digital video camera was also used to monitor the bed height. A schematic of the flow and control system for the column is shown in Fig. 2 and typical outputs will be discussed later. For each experiment, the column was operated continuously until steady state was attained. Then, a steady shear was applied by rotating the concentric cylinders at a fixed speed. For the 1-m bed case, three rotation rates were considered; 1, 10 and 20 rpm. For the higher bed heights, only 1 rotation rate (10 rpm) was studied. In the 1-m case, the control strategy maintained the bed height at 1.0 70.2 m. After approximately 19 h, steady-state was assumed and samples were collected ( 20 mL) through ball valves mounted along the pilot column at a vertical interval of 20 cm. For each sample, the solid concentration was determined by drying in an oven overnight. Once sampled, the concentric cylinders were rotated first at a speed of 1 rpm and the column operated until steady state was re-established. The bed sampling was repeated and then the rotation rate was increased further to
FT
Stock slurry
PT
PID Underflow
Fig. 2. Flow circuit incorporating feed back control and online density measurements. FT and PT are flow and pressure controllers, respectively.
10 rpm, and finally 20 rpm. The approximate shear rates in the column corresponding to the above rotation rates are presented in Table 2. Assuming all of the material in the Couette is being sheared, the nominal shear rate was estimated using:
g_
nR0 nRi R0 Ri
ð1Þ
3. Results and discussion 3.1. 1–D modelling 3.1.1. No shear The compressive yield stress (Py(f)) and hindered settling functions (R(f)) corresponding to conditions of zero applied shear were input into the 1–D model resulting in the solid flux curves in Fig. 3. At a feed solids flux of 0.3 t h 1 m 2, dewatering is predicted to depend primarily on the permeability of the suspension and thus the thickener underflow concentration is predicted to be the same, regardless of bed height. Furthermore, the bed zone in the pilot thickener is not expected to be networked since the predicted solids at the top of the bed are below the measured gel point (fg) of the flocculated calcite. 3.1.2. Model predictions with shear A previous study (Gladman et al., 2005) showed that dewatering of calcite flocculated under the conditions used here was improved at shear rates up to approximately 10–12 s 1. In that case, the shear was applied to a batch settling characterisation of a suspension in a Couette device and the effects on continuous thickening of sheared material predicted from the resulting rheological parameterisations. An important assumption in the modelling exercise was that there was ample time for the flocculated calcite aggregates to achieve a new aggregate state as a result of shear in the pilot column and thus the model was applicable in the continuous case (where the residence time at a given solids concentration is far greater). Consequently, only steady state and not transient modelling was conducted. The effect of a shear rate of 10.8 s 1 shear on the solids flux predicted by the 1-D model is shown in Fig. 4. The data for the unsheared case are also shown. For both flux curves, a 1 m bed was assumed. The predicted effect of shear is to shift the predicted solids flux density curves to higher underflow solid concentrations. For the solids flux density relevant to the column experiments, the dashed lines show that the underflow solids
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
4297
Table 2 Dimensions of concentric cylinders and nominal shear rate in pilot column. Radius (mm) R1 R2 R3 R4 R5 R6
9.5 22 25 59.5 62.5 143.5
Ri/Ro
Rotation rate (rpm)
Shear rate (s 1)
Rotation rate (rpm)
Shear rate (s 1)
0.4318
1
0.08
10
0.79
0.4201
1
0.08
10
0.76
0.4355
1
0.08
10
0.81
10
3000 Dilution water
1
Flow rate (mL min-1)
Solids Flux, (tonnes h-1m-2)
2500
0.1
0.5 metres 1.0 metres 1.5 metres 2.0 metres 5.0 metres 10 metres
0.01
1000 Slurry Flocculant
0.1 0.2 0.3 0.4 0.5 Underflow Solids Concentration, φ (v/v)
0.6
Fig. 3. Thickener solid flux versus underflow for different bed heights (predicted from an un-sheared batch settling test characterisation). The dashed line shows the operating flux density of 0.3 t h 1 m 2.
0 0
10
20
30 Time (hours)
40
50
Fig. 5. Variation of flow rates of controlled variables for a 0.3 t h 1 m 2 run at 1 m bed height in the pilot column.
shear is predicted to be a factor of 1.5 higher (i.e. predicted PE of 1.5) than the case with no shear.
10 0 s-1 Solids Flux, (tonnes h-1m-2)
Underflow
1500
500
0.001 0
2000
10.85 s-1
1
0.1
0.01
0.001 0
0.1 0.2 0.3 0.4 Underflow Solids Concentration, φ (v/v)
0.5
Fig. 4. Predicted effect of shear on thickener solid flux density versus underflow solids concentration for the case of a 1 m bed height and shear at & 0 s 1 and B 10.85 s 1.
volume fraction is predicted to increase from 0.105 to 0.12 (v/v) with the addition of shear. Alternatively, for an underflow concentration of 0.12 v/v, the solids flux density achievable with
3.1.3. Pilot column results To satisfy the model assumptions, it is important that the column is operating under steady state conditions. Previous work showed that this can take many hours depending on the bed height and solids flux and automation was necessary. To operate for longer periods it was necessary to automate the column. Fig. 5 plots flow-rates for a run where the column was operated at a solids flux density of 0.3 t h 1 m 2 and 1 m bed height. Both flocculant and slurry flow rates were effectively constant for the duration of the run. The water flow rate did drift slightly due to diurnal variations in the mains water pressure. However, the change was gradual and represented o4% of the total flow. The underflow rate varied as the underflow pump speed was used to control the bed height. The flow tended to oscillate due to the integral component of the PID control system. Complete damping was difficult because of the high gain necessary to compensate for the slow dynamical response. In the first 20 h, the average underflow rate increased from 700 mL min 1 to approximately 1400 mL min 1. This corresponds to a decrease in underflow solid concentration as the system moved to steady state. Fig. 6 shows the variation in pressure when the column was operated at 1, 2 and 4 m bed heights. For a 1 m bed height, the pressure was 52.8 kPa; for 2 m, 54.6 kPa; while for 4 m, the pressure was 58 kPa. Like the flow rates, the PID control of column pressure was effective, which was important in maintaining the bed height constant.
ARTICLE IN PRESS 4298
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
60 59
Column Pressure (kPa)
58 4 metre
57 56 55
2 metre
54 53
1 metre
52 51 50 0
10
20
30 Time (hours)
40
50
Fig. 6. Variation in column pressure for 1, 2 and 4 m high beds for a solids flux density of 0.3 t h 1 m 2 run in the pilot column.
1600
1500
Density (kg m-3)
Underflow density (4 m)
1400
10 rpm
1300 Underflow density (1 m)
1200 1 rpm
10 rpm
20 rpm
1100 Feed density (1 and 4m)
1000 0
10
20
30 Time (hours)
40
50
Fig. 7. Variation in feed and underflow density for the 1 and 4 m bed runs at a solids flux density of 0.3 t h 1 m 2 run in the pilot column. Data are shown for a range of shear/rotational rates.
Fig. 7 shows the feed and underflow (UF) densities for two different bed heights (1 and 4 m, unsheared). The results for the 2 m bed showed similar trends and have been omitted for the sake of clarity. For each run, the feed density was constant and equal to 1080 kg m–3. In Run 1, upon the bed reaching 1 m, the density in the UF was 1265 kg m 3 and as the underflow was turned on and the column allowed to reach steady state, this decreased to 1190 kg m 3. In Run 7, when the bed reached 4 m the UF density reached 1390 kg m 3, decreasing to 1350 kg m 3 at steady state. The higher underflow density with bed height is postulated to be due to aggregate rearrangement over time, a phenomenon that is not well understood but is known to depend on the solids residence time (Gladman et al., 2009). Importantly, such behaviour is not taken into account in the phenomenological
models used here. The residence times and bed volumes for each height are presented in Table 3. The underflow density tended to be less steady than the other measured variables with a number of concentration spikes occurring. These spikes were a feature of all pilot column experiments, although occurring more frequent in the early stages of the run ( o20 h) and at higher bed heights. The spikes are likely to be caused by intermittent dislodgement of solid cake from the surface of the shear cylinders or the conical section. Caking is more likely to occur when the suspension bed has a yield stress (solids concentration above 0.18 v/v), which provides an explanation of why the frequency of solid spikes increased with bed height. A scraper arm had been installed to the cone to prevent solids accumulation; however, it was not used as it was found to affect the underflow density. Whilst the concentration spikes are undesirable, when steady state was reached, the underflow solids concentration varied by no more than 3% for the 4 m run and even less for the 1 m bed. For a 1 m bed case, it took approximately 19 h to reach a constant underflow (UF) density of 1190 kg m 3 (0.12 v/v). When the material was sheared at a rate of 1 rpm, the UF density initially was unsteady. Eventually the fluctuation died out, giving an underflow density of 1205 kg m 3 or 0.13 v/v. The predicted value based on 1-D modelling for the sheared case is 0.12 v/v. A subsequent increase in rotation rate to 10 rpm caused the UF density to increase to 1215 kg m 3 (0.135 v/v). A rotation rate of 20 rpm caused the UF density to increase further to 1240 kg m 3 or 0.158 (v/v). For the 4 m bed case, steady state took considerably longer, approximately 31 h, to develop. This is more than four times the average solids residence time in the column. Upon filling, the UF density was 1390 kg m 3. Then for the following 12 h the UF density decreased, stabilising at 1350 kg m–3 where it remained fairly constant. Based on the model predictions, this higher value of underflow density was unexpected since at a solids flux density of 0.3 t h 1 m 2, dewatering is predicted to be permeability limited and for bed height to play no role. However, consistent with a previous study where the column was operated in the permeability limited regime, a clear bed height dependence is noted in the absence of shear (Gladman et al., 2009). For the 4 m bed, introduction of shear at a cylinder rotation rate of 10 rpm caused the UF density to rise to 1430 kg m 3 (0.21 v/v) where it remained for approximately 10 h. The output then became unstable assumed to be as a result of the gap in the cylinders becoming clogged. The run was then abandoned. Fig. 8 compares the experimentally measured underflow data (stand alone points) to the predicted solids flux curves as a function of underflow solids concentration for the 1 m bed where the suspension was (i) unsheared and (ii) sheared at 10.8 s 1. The measured underflow data for the case where the bed was sheared is at higher solids for each different bed height. At the experimental flux rate, the model-predicted data show only a small difference due to shear and no effect due to bed height (thus only the 1 m results are shown for clarity). It is clear from these results that the UF concentration depends on both the shear rate as well as the bed height in the column. It is interesting that this should be the case in a flux regime where the bed of aggregates is predicted to be un-networked. In reality, at least part of the bed is Table 3 Volume and residence time for the pilot column at a flux rate of 0.3 t h 1 m 2 for different bed heights. Bed height (m) Volume of bed (L) Residence time (min)
0.5 23.5 20
1 55.8 50
2 120.5 155
4 249.9 420
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
Solids Flux, (tonnes h-1m-2)
10
1
0.1
Sheared flux curve (1 metre)
0.01
Unsheared flux curve (1 metre)
0.001 0
0.1 0.2 0.3 0.4 Underflow Solids Concentration, φ (v/v)
0.5
Fig. 8. Experimental data for various cylinder rotation rates at a bed height of 1 (&), 2 (B) and 4 (D) m against the predicted data for zero shear or at a shear rate of 10.85 s 1 and a bed height of 1 m.
Rotation speed (rpm)
20
15
10
best batch settling case, the difference in shear rate magnitudes cannot be an explanation of the discrepancy between predicted and pilot scale results. Fig. 10 plots the UF solid concentration versus bed height for two different shear regimes (0 and 10 rpm). The figure shows that the relationship between UF solids concentration and bed height is non-linear. To allow comparison, the model prediction for both the zero (dashed line) and 10.8 s 1 shear rates (dot-dashed line) are also included. The figure clearly shows that for both shear rates, when the bed height increases, the column performance improves, whereas the 1–D model predicts no variation. Thus, discrepancies between thickener modelling and full scale plant operation are likely to include both shear effects and bed height effects that are not currently understood and consequently not included in thickener models. The results further imply that the thickener model is only accurate when the bed height is low ( o1 m) and the shear is negligible. A further observation from Fig. 10 is that at higher bed heights, shear had a less pronounced effect. The density increase Dfu due to shear was greatest at the smallest bed height. As the bed height increased, Dfu decreased, and at 4 m, the two cases are very similar. This implies that either the applied shear is too low to have any further effect or based on the conclusion from the trend in Fig. 9, some unknown residence time effect is able to act over a sufficient length of time to cause the same overall result. Fig. 11 shows the solids volume fraction profile measured in the column during the 2 m bed run under steady shear conditions. The result shows a continuous increase in underflow density with depth, indicating consolidation throughout the column, both above and within the sheared zone. In the solids region less than the gel point concentration, this cannot be compressional, in which case the result is indicative of an aggregate densification behaviour that is not captured appropriately in laboratory characterisation tests (note that the model predicts the solids concentration to be uniform in this region). The upper dashed line shows the position of the interface between the bed and hindered settling zone. The middle and bottom dashed lines represent the upper and lower vertical limits of the rotating cylinders. It is clear
0.23
5
Sheared (10 rpm)
0.21
0.12 0.14 0.16 0.18 UF Solids Concentration, u (v/v)
Fig. 9. Rotation speed versus underflow solid volume fraction for a 1 m bed height in the pilot column at an operational flux density of 0.3 t h 1 m 2.
likely networked in the 2 and 4 m bed height cases. It is also clear that applied shear is having more of an effect on the underflow density than can be predicted from sheared batch settling tests. Fig. 9 plots the measured experimental UF data against rotation speed for the 1 m bed case. The UF concentration increases monotonically with rotation rate and appear to asymptote to approximately 0.16 v/v. At some point (possibly well beyond the shear rates here), this curve must turn back with lower UF solids being attained for very high shear. It is concluded based on the data of Fig. 9 that for the flocculated calcium carbonate system studied here, that shear rates in excess of 1.6 s 1 (20 rpm) (for a 1 m bed height) will not cause any significant additional increase in fu. Given that the best performance in the batch settling corresponded to 10 s 1, and that shearing at 1.6 s 1 in the continuous pilot column improved performance well beyond the
Unsheared
0.19
0.2 UF solids volume fraction
0 0.1
4299
0.17 0.15 0.13 model predicted at 10.8 s-1 0.11
model predicted at 0 s-1
0.09 0.07 0.05 0
1
2 3 bed height (m)
4
5
Fig. 10. Steady state underflow solids concentration as a function of bed height for two different shear rates for the pilot column at an operational flux density of 0.3 t h 1 m 2.
ARTICLE IN PRESS 4300
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
2.5
Height (m)
2
interface between consolidating bed and settling suspension)
1.5
1 Region of shear 0.5
φ feed
φg from b.s test
0 0
0.05
0.1 0.15 0.2 0.25 Solids volume fraction
0.3
0.35
Fig. 11. Solid volume fraction profile of a 2 m bed under steady state bed shear for the pilot column operated at 0.3 t h 1 m 2. Horizontal lines representing the bed interface and the limit of operation of the shear device and vertical lines representing the feed and gel point concentrations are also shown.
that the rate of change of solid concentration is higher in the region of mechanical shear. At 0.4 m, the bed concentration was almost 0.3 (v/v), significantly higher than the measured underflow concentration (0.193 v/v). The reason for this behaviour is unclear but may reflect a cake build up in the region of the second lowest sampling point. Fig. 7 shows that the underflow density for the 1 and 4 m cases was reasonably consistent at steady state and it is more likely that the observed data point at higher concentrations is a column sampling artefact. The gel point of the slurry was measured as 0.18 v/v in the absence of shear. Based on this value, the bed is therefore a combination of a zone that might be networked over part (or all) of the range 0 ozo0.8 m and a zone that is definitely unnetworked from 0.8 oz o2.0 m (where z represents the height from the base of the column). The solids concentration at the top of the bed is slightly below the predicted underflow concentration from the 1-D model with zero shear. This is encouraging as this is the concentration at which the bed is predicted to form, except that no further increase in solids is predicted. It is clearly observed that the increase in solids concentration occurs as a function of residence time in the bed and that this response is not associated with compression, since the majority of the bed is un-networked. Indeed, the entire bed may be un-networked since the gel point has been observed to increase with application of low levels of shear, as was the case here (Gladman et al., 2005). The observed increase in underflow density with bed height remains to be explained. The largest PE ratio for the system here was approximately 5 for the 4 m bed height case. Normally, the role of bed height in dewatering is considered in terms of the amount of compressive force exerted by the bed. By increasing the bed height, the total ‘‘compressive squeeze’’ increases. Compression is most effective at low solid flux densities where residence times are sufficient for interstitial liquor to percolate through the network structure and for consolidation to subsequently occur. However, in the experiments here, the solids flux was too high for compression to play a major role in behaviour. The average volume fraction in the underflow is predicted to depend on the permeability, or the hydrodynamic drag exerted on particles due
to the up-flow of water and under steady state conditions, this is assumed to be a function only of the solids concentration. Thus, other phenomena must also be involved in dewatering. The data are consistent with the hypothesis that when a suspension is continuously sheared at a low rate, the permeability significantly increases (or conversely R(f) decreases). Furthermore, the data are also suggestive of R(f) showing a sensitivity to shear at solid volume fractions less than the gel point, suggesting that shear affects individual aggregates, not just aggregates which have formed a three dimensional network. If it is accepted that one action of shear is to increase the suspension permeability by affecting individual aggregates, essentially causing them to restructure into denser forms with faster settling rates, the next question is the timescale over which this process occurs. A theoretical analysis of this postulate shows that the observed behaviour could be easily explained by aggregate densification (Usher et al., 2009). In a sheared batch settling test conducted in the laboratory (Gladman et al., 2005), the relatively rapid settling rates at low solids means that aggregates only experience shear at solids volume fractions less than the gel point for a matter of seconds before that layer consolidates into a more concentrated one. If the timescale over which an aggregate restructures is greater than the time that aggregates are exposed to shear in batch settling (at a given solid volume fraction), the process of aggregate densification will not reach equilibrium in such a test. In the pilot column, by virtue of the significantly longer time scale and the continuous nature of the experiment, aggregates settle at concentrations less than the gel point in an un-networked bed zone for far longer periods (minutes to hours) than in a batch settling test. Therefore, although it is common to compare the settling rates of flocculated suspensions in the sheared and un-sheared state under batch settling, the transformation of the data to R(f) with subsequent model incorporation will not be representative of the longer time scale of the densification processes that occurs at concentrations less than the gel point. This time dependence in the change in aggregate dewatering clearly has a significant affect on the output solids concentration of a thickener operated in the permeability limited flux regime. The clear message from the data is that densification does not only take place in the region of the rotating cylinder. It is hypothesised that shear resulting from hydrodynamic gradients and particle collisions is sufficient for densification to occur. Work is ongoing in our laboratories in which un-networked flocculated slurries are fluidised and the densification and subsequent settling behaviour is monitored. Initial results indicate that this result is reproducible in a laboratory but is highly dependent on the flocculation state of the aggregate. Earlier work also demonstrated a strong dependence on flocculant dose rate of the change in dewatering (Gladman et al., 2009). This is in addition to the role of the rake in aiding dewatering in networked beds of particles, a scenario not considered here.
4. Conclusions The thickening of a flocculated calcite suspension has been examined in a pilot thickening column using controlled shear Couette style cylinders as the shear mechanism. The shear rate was maintained below 2 s 1 and the solids flux density through the column was maintained at 0.3 t h 1 m 2, which according to modelling using a phenomenological 1-D model, should produce an underflow solids that is less than the concentration at which the solids form a networked bed. The solids underflow demonstrated a significant dependence on both the residence time and shear rate that the flocculated aggregates spent in the
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
un-networked bed in the thickener. These results were not predicted from the 1-D model. At higher bed heights and shear rates, a transition to a networked bed was observed in the column. The model predictions were based upon both laboratory batch settling tests and filtration tests. Sheared settling tests were used as a mimic of the shear conditions in the continuous thickener. The work shows that the time scale of the shear imparted in typical laboratory batch settling characterisation tests at solids concentrations less than the gel point is not adequate to mimic the dewatering trends experienced in permeability limited thickening. The work further indicates that gentle shearing of the material from minutes to hours at low solids concentrations appears to be necessary to simulate at least part of the discrepancy between predicted and actual continuous thickening performance. Work to understand this mechanism is currently underway and will be reported in future publications.
Notation h Py R Ro Ri t vo vi z
time, hours compressive yield stress, kPa hindered settling function (resistance to dewatering), Pa s m 2 radius of outer cylinder of Couette shear cell, m radius of inner cylinder of Couette shear cell, m tonnes velocity of outer cylinder of Couette shear cell, m s 1 velocity of inner cylinder of Couette shear cell, m s 1 spatial ordinate, m
Greek letters
g j ff
jg fu
j0
shear rate, s 1 volume fraction of solids, m3 m 3 value of j in the thickener feed, m3 m 3 value of j at the gel point, m3 m 3 value of j in the thickener underflow, m3 m 3 initial value of j, m3 m 3
Acknowledgements This work was performed with the support of a consortium of companies coordinated through the Australian Mineral Industries Research Association (AMIRA) as project 266D and 266E. The financial and in-kind support of Albian Sands Energy, Alcan International, Alcoa World Alumina, AngloGold Ashanti, Anglo Platinum, Aughinish Alumina, Bateman Engineering BV, BHPBilliton, Cable Sands, Ciba Specialty Chemicals, Cytec Australia Holdings, De Beers Consolidated Mines, FL Smidth Minerals,
4301
Glencore AG, Hatch Engineers, Iluka Resources, Kumba Resources, Lightnin, Minara Resources, Metso Minerals, Nalco Chemicals, OMG, Outotec, Pechiney Aluminium, Phelps Dodge, Queensland Alumina, Queensland Nickel, Rio Tinto, WMC Resources and Zinifex Limited through the AMIRA coordinated project is gratefully acknowledged. The support of the Particulate Fluids Processing Centre, a Special Research Centre of the Australian Research Council is also acknowledged. The help and advice of Kosta Simic in the pilot experiments and Phil Fawell is greatly appreciated.
References Coe, H.S., Clevenger, G.H., 1916. Methods for determining the capacities of slime settling tanks. AIME Transactions 55, 356–384. Comings, E.W., Pruiss, C.E., DeBord, C., 1954. Continuous settling and thickening. Industrial and Engineering Chemistry 46, 1164–1172. de Kretser, R.G., Usher, S.P., Scales, P.J., Landman, K.A., Boger, D.V., 2001. Rapid measurement of dewatering design and optimization parameters. A.I.Ch.E. Journal 47, 1758–1769. Farrow, J., Johnston, R., Simic, K., Swift, J., 2000. Consolidation and aggregate densification during gravity thickening. Chemical Engineering Journal 80, 141–148. ¨ Garrido, P., Burgos, R., Concha, F., Burger, R., 2003a. Software for the design and simulation of gravity thickeners. Minerals Engineering 16, 85–92. ¨ Garrido, P., Concha, F., Burger, R., 2003b. Settling velocities of particulate systems: 14. Unified model of sedimentation, centrifugation and filtration of flocculated suspensions. International Journal of Mineral Processing 72, 57. Gladman, B., de Kretser, R.G., Rudman, M., Scales, P.J., 2005. Effect of shear on particulate suspension dewatering. Chemical Engineering Research & Design 83, 1–4. Gladman, B.R., Rudman, M.R., Scales, P.J., 2009. Validation of a 1-D continuous thickening model using a pilot column. Chemical Engineering Science (2010), doi:10.1016/j.ces.2010.03.029. Gladman, B., Usher, S.P. Scales, P.J., 2006. Understanding the thickening process. In: Ninth International Seminar on Paste and Thickened Tailings, Paste 2006, Limerick, Ireland. Holdich, R.G., Butt, G., 1996. Gravity sedimentation including compression and channeling. In: Seventh World Filtration Congress, Budapest, Hungary, May 20–23. Johnson, S.B., Dixon, D.R., Pascoe, M., Scales, P.J., 2000. Use of a super thickener device to concentrate potable water sludge. Water Research 34, 288–294. Kynch, G.J., 1952. A theory of sedimentation. Transactions of the Faraday Society 48, 166–176. Landman, K.A., White, L.R., Buscall, R., 1988. The continuous flow gravity thickener: steady state behaviour. A.I.Ch.E. Journal 34, 239–252. Lester, D., Usher, S.P., Scales, P.J., 2005. Estimation of the hindered settling function R(f) from batch settling tests. A.I.Ch.E. Journal 51, 1158–1168. Novak, J.T., Bandak, N., 1994. The effect of shear on the dewatering of water treatment residuals. Journal of American Water Works Association, 84–91. Rudman, M., Simic, K., Paterson, D.A., Strode, P., Brent, A., Sutalo, I.D., 2008. Raking in gravity thickeners. International Journal of Mineral Processing 86, 114–130. Tiller, F., Yeh, C.S., Tsai, C.D., Chen, W., 1987. Generalised approach to thickening, filtration and centrifugation. Filtration and Separation 24, 121–126. Usher, S.P., Scales, P.J., 2005. Steady state thickener modelling from the compressive yield stress and hindered settling function. Chemical Engineering Journal 111, 253–261. Usher, S.P., Gladman, B., Scales, P.J., 2005. New techniques for understanding settler/clarifier behaviour. In: American Filtration and Separations Society, 18th Annual Conference, Atlanta, USA, April 10–13. Usher, S.P., Spehar, R., Scales, P.J., 2009. Theoretical analysis of aggregate densification; impact on thickener performance. Chemical Engineering Journal 151, 202–208. Vesilind, P.A., Jones, G.N., 1993. Channelling in batch thickening. Water Science and Technology 28, 59–65.
Contents lists available at ScienceDirect
Chemical Engineering Science journal homepage: www.elsevier.com/locate/ces
The effect of shear on gravity thickening: Pilot scale modelling Brendan R. Gladman a, Murray Rudman b, Peter J. Scales a, a b
Particulate Fluids Processing Centre, Department of Chemical and Biomolecular Engineering, The University of Melbourne, Victoria, Australia CSIRO, Mathematical and Information Sciences, Clayton South, Victoria, Australia
a r t i c l e in fo
abstract
Article history: Received 20 April 2009 Received in revised form 28 January 2010 Accepted 12 April 2010 Available online 18 April 2010
Mathematical models are potentially a valuable tool for the prediction of continuous gravity thickener operation. However, experience shows that existing mathematical models underestimate dewatering in thickeners for flocculated feed materials when predictions are made of either the underflow solids concentration for a given solids feed flux density or the maximum solids feed flux density achievable for a minimum underflow solids concentration set point. One reason postulated for this discrepancy is shear enhancement of sedimentation and bed dewatering as a result of aggregate densification. This process is not taken into account in conventional 1-D thickener models. A pilot scale column, operated at low bed heights without the addition of mechanical shear, produced results that compared well with 1-D model predictions. The effect of mechanical shear and/or greater bed height was to significantly enhance thickener performance relative to model predictions (as measured by underflow density or maximum solids flux density achievable for a nominated underflow density). An experimental method was developed that enabled shear to be incorporated into the suspension dewatering characterisation. The results suggest an order of magnitude increase in solid flux density can be expected under controlled shear conditions with polymer flocculated aggregates. The results also indicate that mechanical shear is not the only factor that can enhance dewatering, since higher beds, and hence longer residence times, also improve the achievable solids flux density. This is despite the fact that the thickener is operating in a regime that is predicted to be limited by the sediment permeability and not its compressibility. This suggests an additional mechanism must be at play in full scale operation and points a direction for further experimentation. & 2010 Elsevier Ltd. All rights reserved.
Keywords: Thickening Solid–liquid separation Shear Raking Gravity separation Aggregate densification
1. Introduction Faced with a growing demand for potable water, industry faces a greater expectation to conserve, recover and reuse process water. Gravitational thickening is a widespread process, used to concentrate the solid phase and clarify the liquid phase of a particulate suspension. The recovered water can be recycled within the process or further treated for environmental release. The appeal of gravity thickening is its ability to treat highly variable feeds with low solids concentrations, both economically and efficiently. When operated appropriately, thickeners can produce a concentrated underflow stream that is substantially higher in solids than the feed. The use of gravity as the force driving separation ensures a relatively low operational and energy cost. Simplistically, a thickener can be envisaged as a cylindrical tank with a central feed-well in which solid particles are brought into contact with a polymeric flocculant, causing them to aggregate. As the solids settle from the feed-well, and if the
Corresponding author.
E-mail address: [email protected] (P.J. Scales). 0009-2509/$ - see front matter & 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.ces.2010.04.010
solids flux density permits, a networked bed of solid particles forms at the base of the tank. This base is usually sloped and a scraper or rake aids the transport of thickened solids to a central pump well (the underflow). Clarified liquor exits through a peripheral launder at the top of the tank. In terms of water recovery, the formation of a bed is desirable as the compressive/consolidation forces transmitted in the bed aid sedimentation, increasing the solids concentration in the underflow and thus improved water recovery. Although there are other material properties that can be important in the thickened material (i.e. pumpability of the underflow, angle of repose of the tailings if sent to a Central Tailings Discharge), fast settling to achieve a given underflow solids concentration (i.e. highest throughput for a given water recovery) is usually the design factor that is considered important in conventional thickening. Despite the prevalence of gravity thickeners, their design is usually empirical and their operation often based on a combination of experience and folklore. Kinematic thickener models based on Kynch (1952) theory predict area requirements for thickeners operated at their throughput limit, i.e. where the underflow solids concentration is not sufficient to form a networked bed of
ARTICLE IN PRESS 4294
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
particles and consolidation is absent. However, they do not provide an accurate indication of the expected solids concentration or the rheology of material exiting the underflow for thickeners operated at lower solids fluxes. In mining operations, there are ongoing efforts to continually improve water recovery and to produce thickened tailings, often with a high yield stress (sometimes erroneously termed ‘‘viscosity’’) to aid waste solids disposal. In these applications, the aim is to predict, design and operate thickeners close to the rheological limit; namely the point where the raking mechanism of the thickener is torque limited and a higher solids output would compromise the flow or pumping of solids from the thickener. Phenomenological thickener models encompassing both sedimentation and consolidation are more practical than the classical kinematic models as they take into account compressive forces in the networked bed of particles. Thickeners operating with a networked bed of particles show underflow slurry behaviour that is non-Newtonian and characterised by a yield stress. The ability to predict dewatering is advantageous both in thickener design and operation. On a more fundamental level, modelling can provide a better understanding of dewatering at a microscopic scale and elucidate how flocculation and hydrodynamic conditions can be optimised for throughput, underflow density, etc. A number of workers have modelled the thickening process (Tiller et al., 1987; Landman et al., 1988; Garrido et al., 2003a,b; Usher and Scales, 2005) but very few have reported the output of their model against the output of full scale operational thickeners (Gladman et al., 2006). To achieve a useful comparison, the dewaterability of the feed slurry needs to be characterised via parametric rheological functions termed the compressive yield stress Py(f) and hindered settling function R(f) (or an equivalent set of dewatering parameter descriptors). These functions are key inputs for a phenomenological dewatering model. A review of current thickener operations shows that the majority are predicted to operate in a regime in which the underflow rheology is close to Newtonian or perhaps exhibits a low yield stress, a direct consequence of the torque limitation of the raking mechanism in most cases. In this mode, the maximum operational flux is limited by the suspension permeability, not by its compressibility. Therefore, the bed height and the compressive forces imparted as a result of the bed are not predicted to be the rate-determining step. Both operationally and from a design perspective, it is advantageous to push the rheological limit, but a lack of correlation between models and thickener performance presently makes this an empirical exercise. The lack of a quantitative design process for thickening results from the fact that when laboratory measured dewatering parameters are used as inputs, current phenomenological models consistently underestimate the actual solids flux density required to achieve a given underflow solids concentration by anywhere between 2 and 200 times (Usher et al., 2005). It should be noted at this point that this discrepancy is rarely if ever observed for solids that are not aggregated. It is common practice, however, for the solids fed to a thickener to be aggregated through the addition of polymer flocculant and as a consequence, most current dewatering practice does not correlate with best practice in the modelling field. The work described herein pertains to the case for a flocculated solids thickener feed. The ratio between actual and predicted solids flux density for a given underflow solids concentration is termed the permeability enhancement (PE) factor of the thickener and is in general a function of the operating conditions (feed flux, flocculant dosage, bed height, residence time, etc.). Although factors up to 10 are typical, some level of permeability enhancement is common for
most thickeners. To a reader unfamiliar with gravity thickening, this might seem an extremely poor correlation, although it reflects the current state of the art (Usher et al., 2005). Worth noting is that the predicted underflow solids concentration for a given solids feed flux density is usually predicted to a much better degree with these models than the achievable flux. However, for operational reasons, a minimum underflow solids concentration and a solids feed rate (t h 1) is usually specified and for design reasons the thickener size will be dependent on the settling flux density (t h 1 m 2). Consequently, predicting the maximum achievable flux for a desired underflow density is a highly desirable feature of such models. Pilot scale studies have produced more encouraging results than those obtained from plant measurement. They provide far greater control over the measurement conditions, helping reduce the uncertainty associated with a full scale process. Good control also allows effects due to raking, sloped walls and poorly operated feed-wells to be isolated. In an earlier study by the same authors (Gladman et al., 2009), model validation was considered using a pilot scale dewatering column and flocculated calcite. The work showed that the predicted solids flux density through the column for a given underflow solids concentration was underestimated by factors between 1 (i.e. accurate) and 10 times. The reason why full scale predictions show a greater deviation from the models is difficult to establish. One explanation is that the methods used to measure the dewatering parameters, namely laboratory batch sedimentation (Lester et al., 2005) and filtration tests (de Kretser et al., 2001) are not representative of the hydrodynamic conditions or timescales under which a suspension dewaters in a thickener. Another is that the levels of shear stress on flocculated aggregates are higher in the full scale process and are imposed for longer periods due to the longer inherent solids residence times and finally, that channel formation and other in-homogeneities are not taken into account in the models. Many thickeners employ rakes to transport material to a common discharge point. Several studies in the literature show a qualitative improvement in dewaterability as a result of raking the suspension (Vesilind and Jones, 1993; Farrow et al., 2000; Comings et al., 1954; Novak and Bandak, 1994; Johnson et al., 2000; Holdich and Butt, 1996) although the exact mechanisms by which rakes or pickets enhance dewatering are not well understood. Careful examination of the results revealed that the degree of improvement depends on the speed of the rake and the slope angle of the conical thickener base (Comings et al., 1954). Both higher rake speed and a steeper conical section were associated with higher underflow densities and since the thickener was generally permeability limited, this implies that rake speed and cone angle also improved the permeability (Usher et al., 2005). As material settles and flows in the region of the sloped surface or past a rake blade, the flocculated material is sheared, leading to the conclusion that shear enhances the permeability of networked, flocculated suspensions. Investigations of the effect of shear on dewatering generally conclude that a low level of mechanical shear improves the rate and extent of dewatering in gravity settling (Gladman et al., 2005). Wall effects have been shown to be important; settling tests in a conical geometry revealed that settling in a sloped vessel was faster and achieved a lower equilibrium height than the equivalent test in a straight walled vessel (Usher et al., 2005). However, these conclusions are usually qualitative and do not provide a framework for incorporating shear into dewatering models. To quantify the role of mechanical shear, shear rates were targeted to match those typically encountered in a raked thickener bed (Rudman et al., 2008) and Py(f) and R(f) measured
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
as a function of the volume fraction of solids and shear in a Couette device (Gladman et al., 2005). The results showed that shear increased both the permeability and gel point of the suspension significantly. By accounting for shear in the model, it is hoped that simulation results for full scale thickeners (and here, for a pilot column) will be closer to those observed in practice. To understand how shear affects dewatering and therefore better predict thickening behaviour, changes in Py(f) and R(f) will be measured at different shear rates in our standard batch settling characterisation technique (Lester et al., 2005). A pilot thickening column, described previously (Gladman et al., 2009) but with the inclusion of shear elements, will also be used to validate the results. Pilot rather than full scale experiments provide more control over experimental parameters and the flexibility to vary these parameters. The overall aim of the work was to try and provide a quantitative relationship between the batch dewatering of flocculated suspensions subjected to known shear stresses in the laboratory, a one-dimensional phenomenological thickening model and the output of the pilot column under controlled shear conditions.
2. Experimental methods 2.1. Feed characterisation The procedure used to characterise the dewaterability of the calcite feed has been described elsewhere (Gladman et al., 2009, 2005) and only a brief description will be provided here. The solid used in this work was calcium carbonate (Omyacarb 10). A high molecular weight anionic polyacrylamide/polyacrylate copolymer (AN934SH from SNF) was used to flocculate the suspension. Stock polymer solutions were prepared at 5 g L 1. Working polymer solutions (0.2 g L 1) were prepared each day and any left over solution discarded. The dose rate of the polymer flocculant on a dry solids basis was 40 g t 1. Calcite (10 w/w%) was flocculated using a linear pipe reactor into a Couette shear cell, comprising three concentric cylinders: the inner and outer cylinders were fixed and a third cylinder in between, rotated to provide shear. The suspension–water interface height was measured for three different shear rates: 0, 4.1 and 10.8 s 1 and the dewatering material parameters Py(f) and R(f) were measured (Lester et al., 2005). The shear rate range covered in the laboratory batch settling characterisation is noted to be higher than that able to be explored in pilot scale (in this instance) and full scale studies (Rudman et al., 2008). It has already been established that shear rates in excess of 10–12 s 1 are detrimental to dewatering for this material and that there exists an optimum shear rate at solids concentrations less than the gel point of around 4 s 1 (Gladman et al., 2005). The resultant Py(f) and R(f) from these batch settling tests were then used to model dewatering in a continuously operated pilot thickening column.
4295
the thickener diameter (and in the case of bed heights in the
cone section of the thickener, the diameter as a function of height); the solids volume fraction of the feed ff; the density of the solid and fluid phases and the bed height.
2.3. Pilot column A continuous thickening pilot column (dimensions presented in Table 1) was used in the continuous dewatering experiments. The 287 mm diameter column was modified from a previous study (Gladman et al., 2009) as follows:
the column height was extended from 2.2 to 5.2 m, and a motor driven rotating cylinder arrangement was installed to provide steady shear. The cylinders were 100 cm in height and located at a distance of 40 cm above the cone apex. Fig. 1 shows a schematic of the column in plan and cross sectional view. The outer cylinder was rotated by attachment to a stainless steel shaft driven by an overhead-geared motor. This provided shear with respect to the wall and the stationary inner cylinder. To create Couette flow, the inner cylinder was kept stationary by attaching it to a stainless steel outer sheath, fitted over the shaft. Some level of shear was also imparted in the gap between the shaft and stationary inner cylinder. The motor allowed the shaft to be rotated at speeds between 0 and 20 rpm. Shear rates at 20 rpm were of the order of 1.6 s 1. Although these values are somewhat less than those used in the sheared batch settling tests, Table 1 Pilot column dimensions. Height (m) Truncated cone height (m) Column diameter (m) Cone angle (deg) Diameter of truncated cone at base (m) Cone volume (L)
5.2 0.205 0.285 60 0.05 4.4
5m 2.2. Thickener modelling
1m Using the approach described by Usher and Scales (2005) a 1-D thickener model developed by Landman et al. (1988) was solved numerically in combination with flux limiting calculations proposed by Coe and Clevenger (1916). Inputs to the model were:
287 mm 125 mm 50 mm
constitutive relationships for the compressibility and hindered settling function as functions of the solids volume fraction for the feed slurry, Py(f) and R(f);
Fig. 1. Cross-section and plan view of tall column with Couette style shear device. The inner and 125 mm cylinders rotate to provide the shear.
ARTICLE IN PRESS 4296
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
they are still within the range for which noticeable enhancement of permeability and gel point was observed. The vertical position of the cylinders could be adjusted by loosening two screws, which attached the cylinders to the shaft/sheath, allowing the column contents to be sheared at any height.
PID FT Mains Water
Flocculant solution
FT Isolating Valve
2.4. Column experimental procedure
Overflow
PID Concentrated calcite suspension was prepared at 40% w/w using local tap water (low dissolved solids and a conductivity of 0.177 mS cm 1). An approximate volume of 3500 L of suspension was prepared for each batch by adding 1845 kg of calcite to 2770 L of water. An overhead stirrer mixed the suspension for 24 h prior to use. The concentrated slurry was blended with tap water to produce a 10% w/w suspension. The chosen solid flux density was 0.3 t h–1 m 2 or volumetric feed rate of 3.0 L min 1. The feed to the pilot column was flocculated using a linear pipe reactor with an internal diameter of 12.5 mm and length of 1.8 m. The flocculant was dosed at 40 g t 1 of solids or 64.8 mL min 1. Each of the flocculant, dilution water and slurry flows were regulated using a PID feedback control system. Three bed heights were investigated: 1, 2 and 4 m. For each experiment, the flocculated calcite was introduced to the column approximately 0.5 m from the top. After operating for several minutes, three zones became apparent: a clarification zone above the feed point, a dilute hindered settling zone and an opaque bed zone. Here, the bed shall be defined as the interface between the hindered settling zone and the opaque zone in which the settlement of individual aggregates could no longer be distinguished. The bed formed a well defined interface and could be readily detected by eye. It should be noted that this ‘‘interface’’ does not necessarily indicate the point at which the bed of particles becomes networked and able to support normal stresses but is simply the concentration at which the hindered settling of the suspension becomes commensurate with the downward solids flux created by the underflow pump (Gladman et al., 2009). After start-up, the bed height gradually increased as solids accumulated. The bed pressure was measured using a pressure transducer located in the conical base. When the bed reached the desired height, the pressure was recorded and this value used as the set point for the pressure control loop. The control involved a feedback signal from the transducer to a variable frequency motor drive with onboard PID control. Any deviation was corrected by manipulating the underflow pump speed. Online density meters recorded the density of the feed suspension and the underflow exiting the column. Flow rates, densities, temperatures and pressure were logged so that the cause of any perturbation or disturbance could be traced. A digital video camera was also used to monitor the bed height. A schematic of the flow and control system for the column is shown in Fig. 2 and typical outputs will be discussed later. For each experiment, the column was operated continuously until steady state was attained. Then, a steady shear was applied by rotating the concentric cylinders at a fixed speed. For the 1-m bed case, three rotation rates were considered; 1, 10 and 20 rpm. For the higher bed heights, only 1 rotation rate (10 rpm) was studied. In the 1-m case, the control strategy maintained the bed height at 1.0 70.2 m. After approximately 19 h, steady-state was assumed and samples were collected ( 20 mL) through ball valves mounted along the pilot column at a vertical interval of 20 cm. For each sample, the solid concentration was determined by drying in an oven overnight. Once sampled, the concentric cylinders were rotated first at a speed of 1 rpm and the column operated until steady state was re-established. The bed sampling was repeated and then the rotation rate was increased further to
FT
Stock slurry
PT
PID Underflow
Fig. 2. Flow circuit incorporating feed back control and online density measurements. FT and PT are flow and pressure controllers, respectively.
10 rpm, and finally 20 rpm. The approximate shear rates in the column corresponding to the above rotation rates are presented in Table 2. Assuming all of the material in the Couette is being sheared, the nominal shear rate was estimated using:
g_
nR0 nRi R0 Ri
ð1Þ
3. Results and discussion 3.1. 1–D modelling 3.1.1. No shear The compressive yield stress (Py(f)) and hindered settling functions (R(f)) corresponding to conditions of zero applied shear were input into the 1–D model resulting in the solid flux curves in Fig. 3. At a feed solids flux of 0.3 t h 1 m 2, dewatering is predicted to depend primarily on the permeability of the suspension and thus the thickener underflow concentration is predicted to be the same, regardless of bed height. Furthermore, the bed zone in the pilot thickener is not expected to be networked since the predicted solids at the top of the bed are below the measured gel point (fg) of the flocculated calcite. 3.1.2. Model predictions with shear A previous study (Gladman et al., 2005) showed that dewatering of calcite flocculated under the conditions used here was improved at shear rates up to approximately 10–12 s 1. In that case, the shear was applied to a batch settling characterisation of a suspension in a Couette device and the effects on continuous thickening of sheared material predicted from the resulting rheological parameterisations. An important assumption in the modelling exercise was that there was ample time for the flocculated calcite aggregates to achieve a new aggregate state as a result of shear in the pilot column and thus the model was applicable in the continuous case (where the residence time at a given solids concentration is far greater). Consequently, only steady state and not transient modelling was conducted. The effect of a shear rate of 10.8 s 1 shear on the solids flux predicted by the 1-D model is shown in Fig. 4. The data for the unsheared case are also shown. For both flux curves, a 1 m bed was assumed. The predicted effect of shear is to shift the predicted solids flux density curves to higher underflow solid concentrations. For the solids flux density relevant to the column experiments, the dashed lines show that the underflow solids
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
4297
Table 2 Dimensions of concentric cylinders and nominal shear rate in pilot column. Radius (mm) R1 R2 R3 R4 R5 R6
9.5 22 25 59.5 62.5 143.5
Ri/Ro
Rotation rate (rpm)
Shear rate (s 1)
Rotation rate (rpm)
Shear rate (s 1)
0.4318
1
0.08
10
0.79
0.4201
1
0.08
10
0.76
0.4355
1
0.08
10
0.81
10
3000 Dilution water
1
Flow rate (mL min-1)
Solids Flux, (tonnes h-1m-2)
2500
0.1
0.5 metres 1.0 metres 1.5 metres 2.0 metres 5.0 metres 10 metres
0.01
1000 Slurry Flocculant
0.1 0.2 0.3 0.4 0.5 Underflow Solids Concentration, φ (v/v)
0.6
Fig. 3. Thickener solid flux versus underflow for different bed heights (predicted from an un-sheared batch settling test characterisation). The dashed line shows the operating flux density of 0.3 t h 1 m 2.
0 0
10
20
30 Time (hours)
40
50
Fig. 5. Variation of flow rates of controlled variables for a 0.3 t h 1 m 2 run at 1 m bed height in the pilot column.
shear is predicted to be a factor of 1.5 higher (i.e. predicted PE of 1.5) than the case with no shear.
10 0 s-1 Solids Flux, (tonnes h-1m-2)
Underflow
1500
500
0.001 0
2000
10.85 s-1
1
0.1
0.01
0.001 0
0.1 0.2 0.3 0.4 Underflow Solids Concentration, φ (v/v)
0.5
Fig. 4. Predicted effect of shear on thickener solid flux density versus underflow solids concentration for the case of a 1 m bed height and shear at & 0 s 1 and B 10.85 s 1.
volume fraction is predicted to increase from 0.105 to 0.12 (v/v) with the addition of shear. Alternatively, for an underflow concentration of 0.12 v/v, the solids flux density achievable with
3.1.3. Pilot column results To satisfy the model assumptions, it is important that the column is operating under steady state conditions. Previous work showed that this can take many hours depending on the bed height and solids flux and automation was necessary. To operate for longer periods it was necessary to automate the column. Fig. 5 plots flow-rates for a run where the column was operated at a solids flux density of 0.3 t h 1 m 2 and 1 m bed height. Both flocculant and slurry flow rates were effectively constant for the duration of the run. The water flow rate did drift slightly due to diurnal variations in the mains water pressure. However, the change was gradual and represented o4% of the total flow. The underflow rate varied as the underflow pump speed was used to control the bed height. The flow tended to oscillate due to the integral component of the PID control system. Complete damping was difficult because of the high gain necessary to compensate for the slow dynamical response. In the first 20 h, the average underflow rate increased from 700 mL min 1 to approximately 1400 mL min 1. This corresponds to a decrease in underflow solid concentration as the system moved to steady state. Fig. 6 shows the variation in pressure when the column was operated at 1, 2 and 4 m bed heights. For a 1 m bed height, the pressure was 52.8 kPa; for 2 m, 54.6 kPa; while for 4 m, the pressure was 58 kPa. Like the flow rates, the PID control of column pressure was effective, which was important in maintaining the bed height constant.
ARTICLE IN PRESS 4298
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
60 59
Column Pressure (kPa)
58 4 metre
57 56 55
2 metre
54 53
1 metre
52 51 50 0
10
20
30 Time (hours)
40
50
Fig. 6. Variation in column pressure for 1, 2 and 4 m high beds for a solids flux density of 0.3 t h 1 m 2 run in the pilot column.
1600
1500
Density (kg m-3)
Underflow density (4 m)
1400
10 rpm
1300 Underflow density (1 m)
1200 1 rpm
10 rpm
20 rpm
1100 Feed density (1 and 4m)
1000 0
10
20
30 Time (hours)
40
50
Fig. 7. Variation in feed and underflow density for the 1 and 4 m bed runs at a solids flux density of 0.3 t h 1 m 2 run in the pilot column. Data are shown for a range of shear/rotational rates.
Fig. 7 shows the feed and underflow (UF) densities for two different bed heights (1 and 4 m, unsheared). The results for the 2 m bed showed similar trends and have been omitted for the sake of clarity. For each run, the feed density was constant and equal to 1080 kg m–3. In Run 1, upon the bed reaching 1 m, the density in the UF was 1265 kg m 3 and as the underflow was turned on and the column allowed to reach steady state, this decreased to 1190 kg m 3. In Run 7, when the bed reached 4 m the UF density reached 1390 kg m 3, decreasing to 1350 kg m 3 at steady state. The higher underflow density with bed height is postulated to be due to aggregate rearrangement over time, a phenomenon that is not well understood but is known to depend on the solids residence time (Gladman et al., 2009). Importantly, such behaviour is not taken into account in the phenomenological
models used here. The residence times and bed volumes for each height are presented in Table 3. The underflow density tended to be less steady than the other measured variables with a number of concentration spikes occurring. These spikes were a feature of all pilot column experiments, although occurring more frequent in the early stages of the run ( o20 h) and at higher bed heights. The spikes are likely to be caused by intermittent dislodgement of solid cake from the surface of the shear cylinders or the conical section. Caking is more likely to occur when the suspension bed has a yield stress (solids concentration above 0.18 v/v), which provides an explanation of why the frequency of solid spikes increased with bed height. A scraper arm had been installed to the cone to prevent solids accumulation; however, it was not used as it was found to affect the underflow density. Whilst the concentration spikes are undesirable, when steady state was reached, the underflow solids concentration varied by no more than 3% for the 4 m run and even less for the 1 m bed. For a 1 m bed case, it took approximately 19 h to reach a constant underflow (UF) density of 1190 kg m 3 (0.12 v/v). When the material was sheared at a rate of 1 rpm, the UF density initially was unsteady. Eventually the fluctuation died out, giving an underflow density of 1205 kg m 3 or 0.13 v/v. The predicted value based on 1-D modelling for the sheared case is 0.12 v/v. A subsequent increase in rotation rate to 10 rpm caused the UF density to increase to 1215 kg m 3 (0.135 v/v). A rotation rate of 20 rpm caused the UF density to increase further to 1240 kg m 3 or 0.158 (v/v). For the 4 m bed case, steady state took considerably longer, approximately 31 h, to develop. This is more than four times the average solids residence time in the column. Upon filling, the UF density was 1390 kg m 3. Then for the following 12 h the UF density decreased, stabilising at 1350 kg m–3 where it remained fairly constant. Based on the model predictions, this higher value of underflow density was unexpected since at a solids flux density of 0.3 t h 1 m 2, dewatering is predicted to be permeability limited and for bed height to play no role. However, consistent with a previous study where the column was operated in the permeability limited regime, a clear bed height dependence is noted in the absence of shear (Gladman et al., 2009). For the 4 m bed, introduction of shear at a cylinder rotation rate of 10 rpm caused the UF density to rise to 1430 kg m 3 (0.21 v/v) where it remained for approximately 10 h. The output then became unstable assumed to be as a result of the gap in the cylinders becoming clogged. The run was then abandoned. Fig. 8 compares the experimentally measured underflow data (stand alone points) to the predicted solids flux curves as a function of underflow solids concentration for the 1 m bed where the suspension was (i) unsheared and (ii) sheared at 10.8 s 1. The measured underflow data for the case where the bed was sheared is at higher solids for each different bed height. At the experimental flux rate, the model-predicted data show only a small difference due to shear and no effect due to bed height (thus only the 1 m results are shown for clarity). It is clear from these results that the UF concentration depends on both the shear rate as well as the bed height in the column. It is interesting that this should be the case in a flux regime where the bed of aggregates is predicted to be un-networked. In reality, at least part of the bed is Table 3 Volume and residence time for the pilot column at a flux rate of 0.3 t h 1 m 2 for different bed heights. Bed height (m) Volume of bed (L) Residence time (min)
0.5 23.5 20
1 55.8 50
2 120.5 155
4 249.9 420
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
Solids Flux, (tonnes h-1m-2)
10
1
0.1
Sheared flux curve (1 metre)
0.01
Unsheared flux curve (1 metre)
0.001 0
0.1 0.2 0.3 0.4 Underflow Solids Concentration, φ (v/v)
0.5
Fig. 8. Experimental data for various cylinder rotation rates at a bed height of 1 (&), 2 (B) and 4 (D) m against the predicted data for zero shear or at a shear rate of 10.85 s 1 and a bed height of 1 m.
Rotation speed (rpm)
20
15
10
best batch settling case, the difference in shear rate magnitudes cannot be an explanation of the discrepancy between predicted and pilot scale results. Fig. 10 plots the UF solid concentration versus bed height for two different shear regimes (0 and 10 rpm). The figure shows that the relationship between UF solids concentration and bed height is non-linear. To allow comparison, the model prediction for both the zero (dashed line) and 10.8 s 1 shear rates (dot-dashed line) are also included. The figure clearly shows that for both shear rates, when the bed height increases, the column performance improves, whereas the 1–D model predicts no variation. Thus, discrepancies between thickener modelling and full scale plant operation are likely to include both shear effects and bed height effects that are not currently understood and consequently not included in thickener models. The results further imply that the thickener model is only accurate when the bed height is low ( o1 m) and the shear is negligible. A further observation from Fig. 10 is that at higher bed heights, shear had a less pronounced effect. The density increase Dfu due to shear was greatest at the smallest bed height. As the bed height increased, Dfu decreased, and at 4 m, the two cases are very similar. This implies that either the applied shear is too low to have any further effect or based on the conclusion from the trend in Fig. 9, some unknown residence time effect is able to act over a sufficient length of time to cause the same overall result. Fig. 11 shows the solids volume fraction profile measured in the column during the 2 m bed run under steady shear conditions. The result shows a continuous increase in underflow density with depth, indicating consolidation throughout the column, both above and within the sheared zone. In the solids region less than the gel point concentration, this cannot be compressional, in which case the result is indicative of an aggregate densification behaviour that is not captured appropriately in laboratory characterisation tests (note that the model predicts the solids concentration to be uniform in this region). The upper dashed line shows the position of the interface between the bed and hindered settling zone. The middle and bottom dashed lines represent the upper and lower vertical limits of the rotating cylinders. It is clear
0.23
5
Sheared (10 rpm)
0.21
0.12 0.14 0.16 0.18 UF Solids Concentration, u (v/v)
Fig. 9. Rotation speed versus underflow solid volume fraction for a 1 m bed height in the pilot column at an operational flux density of 0.3 t h 1 m 2.
likely networked in the 2 and 4 m bed height cases. It is also clear that applied shear is having more of an effect on the underflow density than can be predicted from sheared batch settling tests. Fig. 9 plots the measured experimental UF data against rotation speed for the 1 m bed case. The UF concentration increases monotonically with rotation rate and appear to asymptote to approximately 0.16 v/v. At some point (possibly well beyond the shear rates here), this curve must turn back with lower UF solids being attained for very high shear. It is concluded based on the data of Fig. 9 that for the flocculated calcium carbonate system studied here, that shear rates in excess of 1.6 s 1 (20 rpm) (for a 1 m bed height) will not cause any significant additional increase in fu. Given that the best performance in the batch settling corresponded to 10 s 1, and that shearing at 1.6 s 1 in the continuous pilot column improved performance well beyond the
Unsheared
0.19
0.2 UF solids volume fraction
0 0.1
4299
0.17 0.15 0.13 model predicted at 10.8 s-1 0.11
model predicted at 0 s-1
0.09 0.07 0.05 0
1
2 3 bed height (m)
4
5
Fig. 10. Steady state underflow solids concentration as a function of bed height for two different shear rates for the pilot column at an operational flux density of 0.3 t h 1 m 2.
ARTICLE IN PRESS 4300
B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
2.5
Height (m)
2
interface between consolidating bed and settling suspension)
1.5
1 Region of shear 0.5
φ feed
φg from b.s test
0 0
0.05
0.1 0.15 0.2 0.25 Solids volume fraction
0.3
0.35
Fig. 11. Solid volume fraction profile of a 2 m bed under steady state bed shear for the pilot column operated at 0.3 t h 1 m 2. Horizontal lines representing the bed interface and the limit of operation of the shear device and vertical lines representing the feed and gel point concentrations are also shown.
that the rate of change of solid concentration is higher in the region of mechanical shear. At 0.4 m, the bed concentration was almost 0.3 (v/v), significantly higher than the measured underflow concentration (0.193 v/v). The reason for this behaviour is unclear but may reflect a cake build up in the region of the second lowest sampling point. Fig. 7 shows that the underflow density for the 1 and 4 m cases was reasonably consistent at steady state and it is more likely that the observed data point at higher concentrations is a column sampling artefact. The gel point of the slurry was measured as 0.18 v/v in the absence of shear. Based on this value, the bed is therefore a combination of a zone that might be networked over part (or all) of the range 0 ozo0.8 m and a zone that is definitely unnetworked from 0.8 oz o2.0 m (where z represents the height from the base of the column). The solids concentration at the top of the bed is slightly below the predicted underflow concentration from the 1-D model with zero shear. This is encouraging as this is the concentration at which the bed is predicted to form, except that no further increase in solids is predicted. It is clearly observed that the increase in solids concentration occurs as a function of residence time in the bed and that this response is not associated with compression, since the majority of the bed is un-networked. Indeed, the entire bed may be un-networked since the gel point has been observed to increase with application of low levels of shear, as was the case here (Gladman et al., 2005). The observed increase in underflow density with bed height remains to be explained. The largest PE ratio for the system here was approximately 5 for the 4 m bed height case. Normally, the role of bed height in dewatering is considered in terms of the amount of compressive force exerted by the bed. By increasing the bed height, the total ‘‘compressive squeeze’’ increases. Compression is most effective at low solid flux densities where residence times are sufficient for interstitial liquor to percolate through the network structure and for consolidation to subsequently occur. However, in the experiments here, the solids flux was too high for compression to play a major role in behaviour. The average volume fraction in the underflow is predicted to depend on the permeability, or the hydrodynamic drag exerted on particles due
to the up-flow of water and under steady state conditions, this is assumed to be a function only of the solids concentration. Thus, other phenomena must also be involved in dewatering. The data are consistent with the hypothesis that when a suspension is continuously sheared at a low rate, the permeability significantly increases (or conversely R(f) decreases). Furthermore, the data are also suggestive of R(f) showing a sensitivity to shear at solid volume fractions less than the gel point, suggesting that shear affects individual aggregates, not just aggregates which have formed a three dimensional network. If it is accepted that one action of shear is to increase the suspension permeability by affecting individual aggregates, essentially causing them to restructure into denser forms with faster settling rates, the next question is the timescale over which this process occurs. A theoretical analysis of this postulate shows that the observed behaviour could be easily explained by aggregate densification (Usher et al., 2009). In a sheared batch settling test conducted in the laboratory (Gladman et al., 2005), the relatively rapid settling rates at low solids means that aggregates only experience shear at solids volume fractions less than the gel point for a matter of seconds before that layer consolidates into a more concentrated one. If the timescale over which an aggregate restructures is greater than the time that aggregates are exposed to shear in batch settling (at a given solid volume fraction), the process of aggregate densification will not reach equilibrium in such a test. In the pilot column, by virtue of the significantly longer time scale and the continuous nature of the experiment, aggregates settle at concentrations less than the gel point in an un-networked bed zone for far longer periods (minutes to hours) than in a batch settling test. Therefore, although it is common to compare the settling rates of flocculated suspensions in the sheared and un-sheared state under batch settling, the transformation of the data to R(f) with subsequent model incorporation will not be representative of the longer time scale of the densification processes that occurs at concentrations less than the gel point. This time dependence in the change in aggregate dewatering clearly has a significant affect on the output solids concentration of a thickener operated in the permeability limited flux regime. The clear message from the data is that densification does not only take place in the region of the rotating cylinder. It is hypothesised that shear resulting from hydrodynamic gradients and particle collisions is sufficient for densification to occur. Work is ongoing in our laboratories in which un-networked flocculated slurries are fluidised and the densification and subsequent settling behaviour is monitored. Initial results indicate that this result is reproducible in a laboratory but is highly dependent on the flocculation state of the aggregate. Earlier work also demonstrated a strong dependence on flocculant dose rate of the change in dewatering (Gladman et al., 2009). This is in addition to the role of the rake in aiding dewatering in networked beds of particles, a scenario not considered here.
4. Conclusions The thickening of a flocculated calcite suspension has been examined in a pilot thickening column using controlled shear Couette style cylinders as the shear mechanism. The shear rate was maintained below 2 s 1 and the solids flux density through the column was maintained at 0.3 t h 1 m 2, which according to modelling using a phenomenological 1-D model, should produce an underflow solids that is less than the concentration at which the solids form a networked bed. The solids underflow demonstrated a significant dependence on both the residence time and shear rate that the flocculated aggregates spent in the
ARTICLE IN PRESS B.R. Gladman et al. / Chemical Engineering Science 65 (2010) 4293–4301
un-networked bed in the thickener. These results were not predicted from the 1-D model. At higher bed heights and shear rates, a transition to a networked bed was observed in the column. The model predictions were based upon both laboratory batch settling tests and filtration tests. Sheared settling tests were used as a mimic of the shear conditions in the continuous thickener. The work shows that the time scale of the shear imparted in typical laboratory batch settling characterisation tests at solids concentrations less than the gel point is not adequate to mimic the dewatering trends experienced in permeability limited thickening. The work further indicates that gentle shearing of the material from minutes to hours at low solids concentrations appears to be necessary to simulate at least part of the discrepancy between predicted and actual continuous thickening performance. Work to understand this mechanism is currently underway and will be reported in future publications.
Notation h Py R Ro Ri t vo vi z
time, hours compressive yield stress, kPa hindered settling function (resistance to dewatering), Pa s m 2 radius of outer cylinder of Couette shear cell, m radius of inner cylinder of Couette shear cell, m tonnes velocity of outer cylinder of Couette shear cell, m s 1 velocity of inner cylinder of Couette shear cell, m s 1 spatial ordinate, m
Greek letters
g j ff
jg fu
j0
shear rate, s 1 volume fraction of solids, m3 m 3 value of j in the thickener feed, m3 m 3 value of j at the gel point, m3 m 3 value of j in the thickener underflow, m3 m 3 initial value of j, m3 m 3
Acknowledgements This work was performed with the support of a consortium of companies coordinated through the Australian Mineral Industries Research Association (AMIRA) as project 266D and 266E. The financial and in-kind support of Albian Sands Energy, Alcan International, Alcoa World Alumina, AngloGold Ashanti, Anglo Platinum, Aughinish Alumina, Bateman Engineering BV, BHPBilliton, Cable Sands, Ciba Specialty Chemicals, Cytec Australia Holdings, De Beers Consolidated Mines, FL Smidth Minerals,
4301
Glencore AG, Hatch Engineers, Iluka Resources, Kumba Resources, Lightnin, Minara Resources, Metso Minerals, Nalco Chemicals, OMG, Outotec, Pechiney Aluminium, Phelps Dodge, Queensland Alumina, Queensland Nickel, Rio Tinto, WMC Resources and Zinifex Limited through the AMIRA coordinated project is gratefully acknowledged. The support of the Particulate Fluids Processing Centre, a Special Research Centre of the Australian Research Council is also acknowledged. The help and advice of Kosta Simic in the pilot experiments and Phil Fawell is greatly appreciated.
References Coe, H.S., Clevenger, G.H., 1916. Methods for determining the capacities of slime settling tanks. AIME Transactions 55, 356–384. Comings, E.W., Pruiss, C.E., DeBord, C., 1954. Continuous settling and thickening. Industrial and Engineering Chemistry 46, 1164–1172. de Kretser, R.G., Usher, S.P., Scales, P.J., Landman, K.A., Boger, D.V., 2001. Rapid measurement of dewatering design and optimization parameters. A.I.Ch.E. Journal 47, 1758–1769. Farrow, J., Johnston, R., Simic, K., Swift, J., 2000. Consolidation and aggregate densification during gravity thickening. Chemical Engineering Journal 80, 141–148. ¨ Garrido, P., Burgos, R., Concha, F., Burger, R., 2003a. Software for the design and simulation of gravity thickeners. Minerals Engineering 16, 85–92. ¨ Garrido, P., Concha, F., Burger, R., 2003b. Settling velocities of particulate systems: 14. Unified model of sedimentation, centrifugation and filtration of flocculated suspensions. International Journal of Mineral Processing 72, 57. Gladman, B., de Kretser, R.G., Rudman, M., Scales, P.J., 2005. Effect of shear on particulate suspension dewatering. Chemical Engineering Research & Design 83, 1–4. Gladman, B.R., Rudman, M.R., Scales, P.J., 2009. Validation of a 1-D continuous thickening model using a pilot column. Chemical Engineering Science (2010), doi:10.1016/j.ces.2010.03.029. Gladman, B., Usher, S.P. Scales, P.J., 2006. Understanding the thickening process. In: Ninth International Seminar on Paste and Thickened Tailings, Paste 2006, Limerick, Ireland. Holdich, R.G., Butt, G., 1996. Gravity sedimentation including compression and channeling. In: Seventh World Filtration Congress, Budapest, Hungary, May 20–23. Johnson, S.B., Dixon, D.R., Pascoe, M., Scales, P.J., 2000. Use of a super thickener device to concentrate potable water sludge. Water Research 34, 288–294. Kynch, G.J., 1952. A theory of sedimentation. Transactions of the Faraday Society 48, 166–176. Landman, K.A., White, L.R., Buscall, R., 1988. The continuous flow gravity thickener: steady state behaviour. A.I.Ch.E. Journal 34, 239–252. Lester, D., Usher, S.P., Scales, P.J., 2005. Estimation of the hindered settling function R(f) from batch settling tests. A.I.Ch.E. Journal 51, 1158–1168. Novak, J.T., Bandak, N., 1994. The effect of shear on the dewatering of water treatment residuals. Journal of American Water Works Association, 84–91. Rudman, M., Simic, K., Paterson, D.A., Strode, P., Brent, A., Sutalo, I.D., 2008. Raking in gravity thickeners. International Journal of Mineral Processing 86, 114–130. Tiller, F., Yeh, C.S., Tsai, C.D., Chen, W., 1987. Generalised approach to thickening, filtration and centrifugation. Filtration and Separation 24, 121–126. Usher, S.P., Scales, P.J., 2005. Steady state thickener modelling from the compressive yield stress and hindered settling function. Chemical Engineering Journal 111, 253–261. Usher, S.P., Gladman, B., Scales, P.J., 2005. New techniques for understanding settler/clarifier behaviour. In: American Filtration and Separations Society, 18th Annual Conference, Atlanta, USA, April 10–13. Usher, S.P., Spehar, R., Scales, P.J., 2009. Theoretical analysis of aggregate densification; impact on thickener performance. Chemical Engineering Journal 151, 202–208. Vesilind, P.A., Jones, G.N., 1993. Channelling in batch thickening. Water Science and Technology 28, 59–65.